VME50121 commited on
Commit
30409a0
·
verified ·
1 Parent(s): 5bf3543

Add PyRUA-Dex evaluation logs (part 2)

Browse files

Evaluation logs for the Bench2Dex leaderboard entry `pyrua-dex` (PKU DAGroup), a coding-agent system that uses no human demonstration data. GitHub PR: https://github.com/Bench2Dex/Bench2Dex-Leaderboard/pull/7

- `pyrua-dex/README.md`, `pyrua-dex/leaderboard_pyrua-dex.json` (identical to the GitHub results JSON)
- `pyrua-dex/per_task_eval/<task>/<None|Equi|Inv|Full>/per_episode.jsonl` and `summary.json`: the original Bench2Dex outputs of each channel's 50 episodes; `native_shards/000` and `001` hold the two batches of 25 as Bench2Dex wrote them.
- Sanitization: absolute asset paths are made relative (`dex2bench_dataset/...`); nothing else is changed (structure, fields, units, seeds, episode counts as written).
- Coverage: 26 tasks x 4 channels x 50 episodes = 5,200 episodes; Bench2Dex `ae870fb0e1f6237f20c8302198ee3850118892e1`. Overall SR 44.62, LSCR 67.02, Eff 0.8889 (Score 59.126).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. pyrua-dex/per_task_eval/27_ball_box_loading/Full/native_shards/001/per_episode.jsonl +0 -0
  2. pyrua-dex/per_task_eval/27_ball_box_loading/Full/native_shards/001/summary.json +230 -0
  3. pyrua-dex/per_task_eval/27_ball_box_loading/Full/per_episode.jsonl +0 -0
  4. pyrua-dex/per_task_eval/27_ball_box_loading/Full/summary.json +235 -0
  5. pyrua-dex/per_task_eval/27_ball_box_loading/Inv/native_shards/000/per_episode.jsonl +0 -0
  6. pyrua-dex/per_task_eval/27_ball_box_loading/Inv/native_shards/000/summary.json +231 -0
  7. pyrua-dex/per_task_eval/27_ball_box_loading/Inv/native_shards/001/per_episode.jsonl +0 -0
  8. pyrua-dex/per_task_eval/27_ball_box_loading/Inv/native_shards/001/summary.json +231 -0
  9. pyrua-dex/per_task_eval/27_ball_box_loading/Inv/per_episode.jsonl +0 -0
  10. pyrua-dex/per_task_eval/27_ball_box_loading/Inv/summary.json +233 -0
  11. pyrua-dex/per_task_eval/27_ball_box_loading/None/native_shards/000/per_episode.jsonl +0 -0
  12. pyrua-dex/per_task_eval/27_ball_box_loading/None/native_shards/000/summary.json +227 -0
  13. pyrua-dex/per_task_eval/27_ball_box_loading/None/native_shards/001/per_episode.jsonl +0 -0
  14. pyrua-dex/per_task_eval/27_ball_box_loading/None/native_shards/001/summary.json +236 -0
  15. pyrua-dex/per_task_eval/27_ball_box_loading/None/per_episode.jsonl +0 -0
  16. pyrua-dex/per_task_eval/27_ball_box_loading/None/summary.json +236 -0
  17. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/native_shards/000/per_episode.jsonl +0 -0
  18. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/native_shards/000/summary.json +230 -0
  19. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/native_shards/001/per_episode.jsonl +0 -0
  20. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/native_shards/001/summary.json +230 -0
  21. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/per_episode.jsonl +0 -0
  22. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/summary.json +236 -0
  23. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/native_shards/000/per_episode.jsonl +0 -0
  24. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/native_shards/000/summary.json +238 -0
  25. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/native_shards/001/per_episode.jsonl +0 -0
  26. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/native_shards/001/summary.json +230 -0
  27. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/per_episode.jsonl +0 -0
  28. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/summary.json +240 -0
  29. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/native_shards/000/per_episode.jsonl +0 -0
  30. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/native_shards/000/summary.json +227 -0
  31. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/native_shards/001/per_episode.jsonl +0 -0
  32. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/native_shards/001/summary.json +225 -0
  33. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/per_episode.jsonl +0 -0
  34. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/summary.json +230 -0
  35. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/native_shards/000/per_episode.jsonl +0 -0
  36. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/native_shards/000/summary.json +228 -0
  37. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/native_shards/001/per_episode.jsonl +0 -0
  38. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/native_shards/001/summary.json +234 -0
  39. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/per_episode.jsonl +0 -0
  40. pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/summary.json +240 -0
  41. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/native_shards/000/per_episode.jsonl +0 -0
  42. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/native_shards/000/summary.json +247 -0
  43. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/native_shards/001/per_episode.jsonl +0 -0
  44. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/native_shards/001/summary.json +249 -0
  45. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/per_episode.jsonl +0 -0
  46. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/summary.json +253 -0
  47. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Full/native_shards/000/per_episode.jsonl +0 -0
  48. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Full/native_shards/000/summary.json +246 -0
  49. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Full/native_shards/001/per_episode.jsonl +0 -0
  50. pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Full/native_shards/001/summary.json +247 -0
pyrua-dex/per_task_eval/27_ball_box_loading/Full/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/27_ball_box_loading/Full/native_shards/001/summary.json ADDED
@@ -0,0 +1,230 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA ring_prefix_thumb_release-f1ac959cf9",
5
+ "base_seed": 202700000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 19,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3570,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.76,
50
+ "success_rate_ci": {
51
+ "mean": 0.76,
52
+ "ci_low": 0.6,
53
+ "ci_high": 0.92
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.93
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 52.200877192980776
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.76,
64
+ "hard_violation_rate": 0.0,
65
+ "drop_rate": 0.0,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 0.0
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.76,
74
+ "at_end_success_rate": 0.76,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.76
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.92,
81
+ "current_chain_depth_progress_score": 0.98,
82
+ "latched_chain_depth_progress_score": 0.98,
83
+ "avg_current_chain_depth": 1.96,
84
+ "avg_latched_chain_depth": 1.96
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7600997727118974,
88
+ "expert_time_s": 39.666666666666664,
89
+ "expert_time_step": 2380,
90
+ "avg_steps_to_success": 3132.0526315789475,
91
+ "avg_policy_steps_to_success": 1044.3684210526317,
92
+ "avg_policy_queries_to_success": 1044.3684210526317
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.0,
96
+ "violation_counts": {},
97
+ "robot_constraint_violation_rate": 1.0,
98
+ "finger_joint_limit_saturation_rate": 0.96,
99
+ "joint_limit_active_step_rate": 0.062375662617850214,
100
+ "finger_joint_limit_saturation_step_rate": 0.0484128062869923,
101
+ "max_joint_limit_excess_rad": 0.9298250675201416,
102
+ "robot_constraint_diagnostics": {
103
+ "joint_limit_soft_counts": {
104
+ "left_finger4_joint1": 237,
105
+ "right_finger1_joint1": 925,
106
+ "right_finger1_joint2": 1,
107
+ "right_finger1_joint3": 109,
108
+ "right_finger2_joint1": 1100,
109
+ "right_finger3_joint1": 449,
110
+ "right_finger3_joint3": 46,
111
+ "right_finger4_joint1": 1526,
112
+ "right_finger4_joint2": 20,
113
+ "right_finger4_joint3": 768,
114
+ "right_finger5_joint1": 344
115
+ },
116
+ "joint_limit_hard_counts": {
117
+ "left_finger4_joint1": 126,
118
+ "right_finger1_joint1": 8,
119
+ "right_finger1_joint2": 6,
120
+ "right_finger1_joint3": 195,
121
+ "right_finger2_joint1": 34,
122
+ "right_finger3_joint1": 12,
123
+ "right_finger3_joint3": 1,
124
+ "right_finger4_joint1": 192,
125
+ "right_finger4_joint2": 13,
126
+ "right_finger4_joint3": 23,
127
+ "right_finger5_joint1": 16,
128
+ "right_finger5_joint3": 3
129
+ },
130
+ "joint_limit_hard_episode_events": 561,
131
+ "finger_joint_limit_saturation_counts": {
132
+ "right_finger1_joint1": 810,
133
+ "right_finger2_joint1": 676,
134
+ "right_finger3_joint1": 190,
135
+ "right_finger4_joint1": 1450,
136
+ "right_finger4_joint3": 610,
137
+ "right_finger5_joint1": 303
138
+ },
139
+ "joint_limit_max_excess_by_joint": {
140
+ "left_finger1_joint1": 0.00015211105346679688,
141
+ "left_finger1_joint4": 2.9802322387695312e-08,
142
+ "left_finger4_joint1": 0.07161253690719604,
143
+ "left_finger4_joint2": 0.0005708634853363037,
144
+ "right_finger1_joint1": 0.526412844657898,
145
+ "right_finger1_joint2": 0.010310426354408264,
146
+ "right_finger1_joint3": 0.031512558460235596,
147
+ "right_finger2_joint1": 0.4000329375267029,
148
+ "right_finger2_joint2": 0.0002550780773162842,
149
+ "right_finger2_joint3": 0.0007015466690063477,
150
+ "right_finger2_joint4": 0.000494837760925293,
151
+ "right_finger3_joint1": 0.25422683358192444,
152
+ "right_finger3_joint2": 0.0007892251014709473,
153
+ "right_finger3_joint3": 0.09393310546875,
154
+ "right_finger3_joint4": 0.0006798505783081055,
155
+ "right_finger4_joint1": 0.40702950954437256,
156
+ "right_finger4_joint2": 0.06009802222251892,
157
+ "right_finger4_joint3": 0.9298250675201416,
158
+ "right_finger5_joint1": 0.8028194308280945,
159
+ "right_finger5_joint2": 0.0009196996688842773,
160
+ "right_finger5_joint3": 0.006759285926818848,
161
+ "right_finger5_joint4": 0.0030918121337890625
162
+ },
163
+ "coverage": {
164
+ "hard_limit_source": {
165
+ "available": 25,
166
+ "unavailable": 0,
167
+ "unknown": 0
168
+ },
169
+ "joint_role_manifest": {
170
+ "available": 25,
171
+ "unavailable": 0,
172
+ "unknown": 0
173
+ }
174
+ },
175
+ "hard_limit_sources": {
176
+ "runtime.joint_pos_limits": 25
177
+ },
178
+ "soft_limit_sources": {
179
+ "runtime.soft_joint_pos_limits": 25
180
+ },
181
+ "unknown_runtime_joint_names": [],
182
+ "baseline_checks": {
183
+ "post_home": {
184
+ "checked": 25,
185
+ "hard_violation_episodes": 0
186
+ },
187
+ "post_reset": {
188
+ "checked": 25,
189
+ "hard_violation_episodes": 0
190
+ }
191
+ }
192
+ }
193
+ },
194
+ "robot_motion": {
195
+ "joint_acceleration_rms": 11.681390233575364,
196
+ "joint_effort_rms": 20.258262157252783,
197
+ "joint_jerk_rms": 1166.1639815564438,
198
+ "joint_velocity_rms": 0.39523109306006143
199
+ },
200
+ "grasp": {
201
+ "kinematic_grasp_stability_index": 0.8400038908520694,
202
+ "mean_kinematic_grasp_stability": 0.4797361230854294,
203
+ "per_object_best_kinematic_grasp_stability": {
204
+ "obj_053_mini_soccer_ball_2": 0.3333164391277472,
205
+ "obj_056_tennis_ball_3": 0.5067063265301391,
206
+ "obj_058_golf_ball_4": 0.7491851291367023,
207
+ "obj_144_table_tennis_5": 0.4304948368556557
208
+ },
209
+ "grasp_gsi_diagnostics": {
210
+ "mean_anti_slip_score": 0.5887223681588738,
211
+ "mean_duration_score": 0.4613571892130512,
212
+ "mean_hold_score": 0.2717508145905905,
213
+ "mean_kinematic_stability_score": 0.04625506526550393,
214
+ "mean_legacy_gsi": 0.36919697560124354,
215
+ "mean_lift_score": 0.31653662871009075,
216
+ "mean_motion_stability_score": 0.25360904394283623
217
+ }
218
+ },
219
+ "tool": {
220
+ "tool_selection_accuracy": null,
221
+ "tool_switch_success_rate": null,
222
+ "tool_switch_total": 0,
223
+ "tool_switch_successful": 0
224
+ },
225
+ "runtime": {
226
+ "avg_steps": 3237.16,
227
+ "error_count": 0
228
+ }
229
+ }
230
+ }
pyrua-dex/per_task_eval/27_ball_box_loading/Full/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/27_ball_box_loading/Full/summary.json ADDED
@@ -0,0 +1,235 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA ring_prefix_thumb_release-f1ac959cf9",
5
+ "base_seed": 202700000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 50,
9
+ "successes": 35,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3570,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.7,
50
+ "success_rate_ci": {
51
+ "mean": 0.7,
52
+ "ci_low": 0.56,
53
+ "ci_high": 0.82
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.89
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 52.39857142856974
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.68,
64
+ "hard_violation_rate": 0.04,
65
+ "drop_rate": 0.02,
66
+ "high_speed_violation_rate": 0.02,
67
+ "safety_violation_step_rate": 1.2225910371851064e-05
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.7,
74
+ "at_end_success_rate": 0.7,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.7
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.88,
81
+ "current_chain_depth_progress_score": 0.95,
82
+ "latched_chain_depth_progress_score": 0.95,
83
+ "avg_current_chain_depth": 1.9,
84
+ "avg_latched_chain_depth": 1.9
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7572564865601721,
88
+ "expert_time_s": 39.666666666666664,
89
+ "expert_time_step": 2380,
90
+ "avg_steps_to_success": 3143.9142857142856,
91
+ "avg_policy_steps_to_success": 1048.3142857142857,
92
+ "avg_policy_queries_to_success": 1048.3142857142857
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 1.2225910371851064e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_056_tennis_ball_3": 1,
98
+ "high_speed_obj_144_table_tennis_5": 1
99
+ },
100
+ "robot_constraint_violation_rate": 0.98,
101
+ "finger_joint_limit_saturation_rate": 0.94,
102
+ "joint_limit_active_step_rate": 0.04640955577154664,
103
+ "finger_joint_limit_saturation_step_rate": 0.03258205114098309,
104
+ "max_joint_limit_excess_rad": 0.9298250675201416,
105
+ "robot_constraint_diagnostics": {
106
+ "joint_limit_soft_counts": {
107
+ "left_finger4_joint1": 501,
108
+ "right_finger1_joint1": 1071,
109
+ "right_finger1_joint2": 1,
110
+ "right_finger1_joint3": 200,
111
+ "right_finger2_joint1": 2259,
112
+ "right_finger3_joint1": 943,
113
+ "right_finger3_joint3": 75,
114
+ "right_finger4_joint1": 1602,
115
+ "right_finger4_joint2": 30,
116
+ "right_finger4_joint3": 1525,
117
+ "right_finger5_joint1": 562,
118
+ "right_finger5_joint3": 1
119
+ },
120
+ "joint_limit_hard_counts": {
121
+ "left_finger4_joint1": 189,
122
+ "right_finger1_joint1": 18,
123
+ "right_finger1_joint2": 12,
124
+ "right_finger1_joint3": 455,
125
+ "right_finger2_joint1": 66,
126
+ "right_finger3_joint1": 26,
127
+ "right_finger3_joint3": 2,
128
+ "right_finger4_joint1": 359,
129
+ "right_finger4_joint2": 30,
130
+ "right_finger4_joint3": 46,
131
+ "right_finger5_joint1": 31,
132
+ "right_finger5_joint3": 5
133
+ },
134
+ "joint_limit_hard_episode_events": 1109,
135
+ "finger_joint_limit_saturation_counts": {
136
+ "right_finger1_joint1": 893,
137
+ "right_finger1_joint3": 19,
138
+ "right_finger2_joint1": 1322,
139
+ "right_finger3_joint1": 458,
140
+ "right_finger4_joint1": 1453,
141
+ "right_finger4_joint3": 1149,
142
+ "right_finger5_joint1": 425
143
+ },
144
+ "joint_limit_max_excess_by_joint": {
145
+ "left_finger1_joint1": 0.00023186206817626953,
146
+ "left_finger1_joint4": 2.9802322387695312e-08,
147
+ "left_finger4_joint1": 0.07161253690719604,
148
+ "left_finger4_joint2": 0.0005708634853363037,
149
+ "right_finger1_joint1": 0.526412844657898,
150
+ "right_finger1_joint2": 0.010310426354408264,
151
+ "right_finger1_joint3": 0.41223740577697754,
152
+ "right_finger2_joint1": 0.4000329375267029,
153
+ "right_finger2_joint2": 0.0005185902118682861,
154
+ "right_finger2_joint3": 0.0007307529449462891,
155
+ "right_finger2_joint4": 0.0015088319778442383,
156
+ "right_finger3_joint1": 0.7560850977897644,
157
+ "right_finger3_joint2": 0.0007892251014709473,
158
+ "right_finger3_joint3": 0.1357850432395935,
159
+ "right_finger3_joint4": 0.0009046792984008789,
160
+ "right_finger4_joint1": 0.40702950954437256,
161
+ "right_finger4_joint2": 0.06009802222251892,
162
+ "right_finger4_joint3": 0.9298250675201416,
163
+ "right_finger5_joint1": 0.8028194308280945,
164
+ "right_finger5_joint2": 0.003274083137512207,
165
+ "right_finger5_joint3": 0.027802586555480957,
166
+ "right_finger5_joint4": 0.0030918121337890625
167
+ },
168
+ "coverage": {
169
+ "hard_limit_source": {
170
+ "available": 50,
171
+ "unavailable": 0,
172
+ "unknown": 0
173
+ },
174
+ "joint_role_manifest": {
175
+ "available": 50,
176
+ "unavailable": 0,
177
+ "unknown": 0
178
+ }
179
+ },
180
+ "hard_limit_sources": {
181
+ "runtime.joint_pos_limits": 50
182
+ },
183
+ "soft_limit_sources": {
184
+ "runtime.soft_joint_pos_limits": 50
185
+ },
186
+ "unknown_runtime_joint_names": [],
187
+ "baseline_checks": {
188
+ "post_home": {
189
+ "checked": 50,
190
+ "hard_violation_episodes": 0
191
+ },
192
+ "post_reset": {
193
+ "checked": 50,
194
+ "hard_violation_episodes": 0
195
+ }
196
+ }
197
+ }
198
+ },
199
+ "robot_motion": {
200
+ "joint_acceleration_rms": 11.386419238847855,
201
+ "joint_effort_rms": 19.162452975711275,
202
+ "joint_jerk_rms": 1133.2395654768552,
203
+ "joint_velocity_rms": 0.38632805470062015
204
+ },
205
+ "grasp": {
206
+ "kinematic_grasp_stability_index": 0.7912437076345341,
207
+ "mean_kinematic_grasp_stability": 0.4593134053522152,
208
+ "per_object_best_kinematic_grasp_stability": {
209
+ "obj_053_mini_soccer_ball_2": 0.3361001220351561,
210
+ "obj_056_tennis_ball_3": 0.4811638942681929,
211
+ "obj_058_golf_ball_4": 0.6434686246924374,
212
+ "obj_144_table_tennis_5": 0.4162624473019325
213
+ },
214
+ "grasp_gsi_diagnostics": {
215
+ "mean_anti_slip_score": 0.6121435797364286,
216
+ "mean_duration_score": 0.4434314641257751,
217
+ "mean_hold_score": 0.2641683451848166,
218
+ "mean_kinematic_stability_score": 0.0466277157633013,
219
+ "mean_legacy_gsi": 0.36550747668901024,
220
+ "mean_lift_score": 0.30793218201706596,
221
+ "mean_motion_stability_score": 0.24589708571341679
222
+ }
223
+ },
224
+ "tool": {
225
+ "tool_selection_accuracy": null,
226
+ "tool_switch_success_rate": null,
227
+ "tool_switch_total": 0,
228
+ "tool_switch_successful": 0
229
+ },
230
+ "runtime": {
231
+ "avg_steps": 3271.74,
232
+ "error_count": 0
233
+ }
234
+ }
235
+ }
pyrua-dex/per_task_eval/27_ball_box_loading/Inv/native_shards/000/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/27_ball_box_loading/Inv/native_shards/000/summary.json ADDED
@@ -0,0 +1,231 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA ring_prefix_thumb_release-f1ac959cf9",
5
+ "base_seed": 102700000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 18,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3570,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.72,
50
+ "success_rate_ci": {
51
+ "mean": 0.72,
52
+ "ci_low": 0.52,
53
+ "ci_high": 0.88
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.9
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 52.66574074073903
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.72,
64
+ "hard_violation_rate": 0.12,
65
+ "drop_rate": 0.08,
66
+ "high_speed_violation_rate": 0.04,
67
+ "safety_violation_step_rate": 6.10731778817379e-05
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.72,
74
+ "at_end_success_rate": 0.72,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.72
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.89,
81
+ "current_chain_depth_progress_score": 0.94,
82
+ "latched_chain_depth_progress_score": 0.94,
83
+ "avg_current_chain_depth": 1.88,
84
+ "avg_latched_chain_depth": 1.88
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7534787087708198,
88
+ "expert_time_s": 39.666666666666664,
89
+ "expert_time_step": 2380,
90
+ "avg_steps_to_success": 3159.9444444444443,
91
+ "avg_policy_steps_to_success": 1053.611111111111,
92
+ "avg_policy_queries_to_success": 1053.611111111111
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 3.6643906729042735e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_144_table_tennis_5": 2,
98
+ "high_speed_obj_058_golf_ball_4": 1
99
+ },
100
+ "robot_constraint_violation_rate": 0.96,
101
+ "finger_joint_limit_saturation_rate": 0.96,
102
+ "joint_limit_active_step_rate": 0.04645225909684984,
103
+ "finger_joint_limit_saturation_step_rate": 0.02391625645848856,
104
+ "max_joint_limit_excess_rad": 0.9570550918579102,
105
+ "robot_constraint_diagnostics": {
106
+ "joint_limit_soft_counts": {
107
+ "left_finger4_joint1": 238,
108
+ "right_finger1_joint1": 115,
109
+ "right_finger1_joint3": 171,
110
+ "right_finger2_joint1": 1559,
111
+ "right_finger3_joint1": 372,
112
+ "right_finger4_joint1": 76,
113
+ "right_finger4_joint2": 22,
114
+ "right_finger4_joint3": 764,
115
+ "right_finger5_joint1": 1001,
116
+ "right_finger5_joint3": 1
117
+ },
118
+ "joint_limit_hard_counts": {
119
+ "left_finger4_joint1": 130,
120
+ "right_finger1_joint1": 7,
121
+ "right_finger1_joint2": 2,
122
+ "right_finger1_joint3": 216,
123
+ "right_finger2_joint1": 35,
124
+ "right_finger3_joint1": 10,
125
+ "right_finger4_joint1": 207,
126
+ "right_finger4_joint2": 16,
127
+ "right_finger4_joint3": 25,
128
+ "right_finger5_joint1": 18,
129
+ "right_finger5_joint3": 1
130
+ },
131
+ "joint_limit_hard_episode_events": 617,
132
+ "finger_joint_limit_saturation_counts": {
133
+ "right_finger1_joint1": 71,
134
+ "right_finger1_joint3": 22,
135
+ "right_finger2_joint1": 1015,
136
+ "right_finger3_joint1": 107,
137
+ "right_finger4_joint1": 38,
138
+ "right_finger4_joint3": 604,
139
+ "right_finger5_joint1": 285
140
+ },
141
+ "joint_limit_max_excess_by_joint": {
142
+ "left_finger1_joint1": 0.00020837783813476562,
143
+ "left_finger4_joint1": 0.07036568224430084,
144
+ "left_finger4_joint2": 0.0005677640438079834,
145
+ "right_finger1_joint1": 0.3365767002105713,
146
+ "right_finger1_joint2": 0.00848989188671112,
147
+ "right_finger1_joint3": 0.5048845410346985,
148
+ "right_finger2_joint1": 0.7581371665000916,
149
+ "right_finger2_joint2": 0.00048828125,
150
+ "right_finger2_joint3": 0.0003578662872314453,
151
+ "right_finger2_joint4": 0.0008252859115600586,
152
+ "right_finger3_joint1": 0.26722079515457153,
153
+ "right_finger3_joint2": 0.0012336671352386475,
154
+ "right_finger3_joint3": 0.0008697509765625,
155
+ "right_finger3_joint4": 0.00047791004180908203,
156
+ "right_finger4_joint1": 0.4625527262687683,
157
+ "right_finger4_joint2": 0.07641583681106567,
158
+ "right_finger4_joint3": 0.9570550918579102,
159
+ "right_finger5_joint1": 0.9212390184402466,
160
+ "right_finger5_joint2": 0.0015336275100708008,
161
+ "right_finger5_joint3": 0.015417933464050293,
162
+ "right_finger5_joint4": 0.0030771493911743164
163
+ },
164
+ "coverage": {
165
+ "hard_limit_source": {
166
+ "available": 25,
167
+ "unavailable": 0,
168
+ "unknown": 0
169
+ },
170
+ "joint_role_manifest": {
171
+ "available": 25,
172
+ "unavailable": 0,
173
+ "unknown": 0
174
+ }
175
+ },
176
+ "hard_limit_sources": {
177
+ "runtime.joint_pos_limits": 25
178
+ },
179
+ "soft_limit_sources": {
180
+ "runtime.soft_joint_pos_limits": 25
181
+ },
182
+ "unknown_runtime_joint_names": [],
183
+ "baseline_checks": {
184
+ "post_home": {
185
+ "checked": 25,
186
+ "hard_violation_episodes": 0
187
+ },
188
+ "post_reset": {
189
+ "checked": 25,
190
+ "hard_violation_episodes": 0
191
+ }
192
+ }
193
+ }
194
+ },
195
+ "robot_motion": {
196
+ "joint_acceleration_rms": 11.364312263939317,
197
+ "joint_effort_rms": 16.36048201640793,
198
+ "joint_jerk_rms": 1135.9439162263275,
199
+ "joint_velocity_rms": 0.35772022031067374
200
+ },
201
+ "grasp": {
202
+ "kinematic_grasp_stability_index": 0.8083676216799426,
203
+ "mean_kinematic_grasp_stability": 0.36102000234639553,
204
+ "per_object_best_kinematic_grasp_stability": {
205
+ "obj_053_mini_soccer_ball_2": 0.3199227998252387,
206
+ "obj_056_tennis_ball_3": 0.47936366311357675,
207
+ "obj_058_golf_ball_4": 0.5967563358881425,
208
+ "obj_144_table_tennis_5": 0.5825035471677201
209
+ },
210
+ "grasp_gsi_diagnostics": {
211
+ "mean_anti_slip_score": 0.6134781164281118,
212
+ "mean_duration_score": 0.42716710312684236,
213
+ "mean_hold_score": 0.2234927824361282,
214
+ "mean_kinematic_stability_score": 0.03360323892256072,
215
+ "mean_legacy_gsi": 0.3399867225492044,
216
+ "mean_lift_score": 0.2648818299487337,
217
+ "mean_motion_stability_score": 0.20982731767101856
218
+ }
219
+ },
220
+ "tool": {
221
+ "tool_selection_accuracy": null,
222
+ "tool_switch_success_rate": null,
223
+ "tool_switch_total": 0,
224
+ "tool_switch_successful": 0
225
+ },
226
+ "runtime": {
227
+ "avg_steps": 3274.76,
228
+ "error_count": 0
229
+ }
230
+ }
231
+ }
pyrua-dex/per_task_eval/27_ball_box_loading/Inv/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/27_ball_box_loading/Inv/native_shards/001/summary.json ADDED
@@ -0,0 +1,231 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA ring_prefix_thumb_release-f1ac959cf9",
5
+ "base_seed": 102700000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 16,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3570,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.64,
50
+ "success_rate_ci": {
51
+ "mean": 0.64,
52
+ "ci_low": 0.44,
53
+ "ci_high": 0.84
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.91
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 52.308333333331646
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.64,
64
+ "hard_violation_rate": 0.12,
65
+ "drop_rate": 0.12,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 6.071940349258008e-05
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.64,
74
+ "at_end_success_rate": 0.64,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.64
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.84,
81
+ "current_chain_depth_progress_score": 0.92,
82
+ "latched_chain_depth_progress_score": 1.0,
83
+ "avg_current_chain_depth": 1.84,
84
+ "avg_latched_chain_depth": 2
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7585180085709265,
88
+ "expert_time_s": 39.666666666666664,
89
+ "expert_time_step": 2380,
90
+ "avg_steps_to_success": 3138.5,
91
+ "avg_policy_steps_to_success": 1046.5625,
92
+ "avg_policy_queries_to_success": 1046.5625
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 6.071940349258008e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_053_mini_soccer_ball_2": 3,
98
+ "drop_to_table_obj_056_tennis_ball_3": 2
99
+ },
100
+ "robot_constraint_violation_rate": 1.0,
101
+ "finger_joint_limit_saturation_rate": 1.0,
102
+ "joint_limit_active_step_rate": 0.07655502392344497,
103
+ "finger_joint_limit_saturation_step_rate": 0.04381512156024579,
104
+ "max_joint_limit_excess_rad": 1.2985665798187256,
105
+ "robot_constraint_diagnostics": {
106
+ "joint_limit_soft_counts": {
107
+ "left_finger4_joint1": 288,
108
+ "right_finger1_joint1": 255,
109
+ "right_finger1_joint3": 525,
110
+ "right_finger2_joint1": 1040,
111
+ "right_finger3_joint1": 307,
112
+ "right_finger4_joint1": 1100,
113
+ "right_finger4_joint2": 15,
114
+ "right_finger4_joint3": 2204,
115
+ "right_finger5_joint1": 1112,
116
+ "right_finger5_joint3": 1
117
+ },
118
+ "joint_limit_hard_counts": {
119
+ "left_finger4_joint1": 57,
120
+ "right_finger1_joint1": 8,
121
+ "right_finger1_joint2": 1,
122
+ "right_finger1_joint3": 263,
123
+ "right_finger2_joint1": 32,
124
+ "right_finger3_joint1": 7,
125
+ "right_finger4_joint1": 229,
126
+ "right_finger4_joint2": 23,
127
+ "right_finger4_joint3": 24,
128
+ "right_finger5_joint1": 23,
129
+ "right_finger5_joint3": 8
130
+ },
131
+ "joint_limit_hard_episode_events": 589,
132
+ "finger_joint_limit_saturation_counts": {
133
+ "right_finger1_joint1": 174,
134
+ "right_finger1_joint3": 71,
135
+ "right_finger2_joint1": 679,
136
+ "right_finger3_joint1": 133,
137
+ "right_finger4_joint1": 355,
138
+ "right_finger4_joint3": 2067,
139
+ "right_finger5_joint1": 379
140
+ },
141
+ "joint_limit_max_excess_by_joint": {
142
+ "left_finger1_joint1": 0.00019681453704833984,
143
+ "left_finger4_joint1": 0.07155676186084747,
144
+ "left_finger4_joint2": 0.0005711913108825684,
145
+ "right_finger1_joint1": 0.4934276342391968,
146
+ "right_finger1_joint2": 0.008814394474029541,
147
+ "right_finger1_joint3": 1.2985665798187256,
148
+ "right_finger2_joint1": 0.5049048066139221,
149
+ "right_finger2_joint2": 0.0015058517456054688,
150
+ "right_finger2_joint3": 0.0027426481246948242,
151
+ "right_finger2_joint4": 0.0008242130279541016,
152
+ "right_finger3_joint1": 0.6439911723136902,
153
+ "right_finger3_joint2": 0.0025108754634857178,
154
+ "right_finger3_joint3": 0.001021265983581543,
155
+ "right_finger3_joint4": 0.0005322694778442383,
156
+ "right_finger4_joint1": 0.8194855451583862,
157
+ "right_finger4_joint2": 0.060319215059280396,
158
+ "right_finger4_joint3": 0.9702248573303223,
159
+ "right_finger5_joint1": 0.8706169128417969,
160
+ "right_finger5_joint2": 0.0028456151485443115,
161
+ "right_finger5_joint3": 0.017028212547302246,
162
+ "right_finger5_joint4": 0.0028802156448364258
163
+ },
164
+ "coverage": {
165
+ "hard_limit_source": {
166
+ "available": 25,
167
+ "unavailable": 0,
168
+ "unknown": 0
169
+ },
170
+ "joint_role_manifest": {
171
+ "available": 25,
172
+ "unavailable": 0,
173
+ "unknown": 0
174
+ }
175
+ },
176
+ "hard_limit_sources": {
177
+ "runtime.joint_pos_limits": 25
178
+ },
179
+ "soft_limit_sources": {
180
+ "runtime.soft_joint_pos_limits": 25
181
+ },
182
+ "unknown_runtime_joint_names": [],
183
+ "baseline_checks": {
184
+ "post_home": {
185
+ "checked": 25,
186
+ "hard_violation_episodes": 0
187
+ },
188
+ "post_reset": {
189
+ "checked": 25,
190
+ "hard_violation_episodes": 0
191
+ }
192
+ }
193
+ }
194
+ },
195
+ "robot_motion": {
196
+ "joint_acceleration_rms": 11.951249267736252,
197
+ "joint_effort_rms": 20.44514875153573,
198
+ "joint_jerk_rms": 1189.400386847986,
199
+ "joint_velocity_rms": 0.40440784072931163
200
+ },
201
+ "grasp": {
202
+ "kinematic_grasp_stability_index": 0.8899506768693273,
203
+ "mean_kinematic_grasp_stability": 0.502362379216431,
204
+ "per_object_best_kinematic_grasp_stability": {
205
+ "obj_053_mini_soccer_ball_2": 0.3381775176377101,
206
+ "obj_056_tennis_ball_3": 0.49562905502229043,
207
+ "obj_058_golf_ball_4": 0.7758631385755071,
208
+ "obj_144_table_tennis_5": 0.4537583954899644
209
+ },
210
+ "grasp_gsi_diagnostics": {
211
+ "mean_anti_slip_score": 0.586412002402819,
212
+ "mean_duration_score": 0.479319978896274,
213
+ "mean_hold_score": 0.24796525480377518,
214
+ "mean_kinematic_stability_score": 0.04782750487847628,
215
+ "mean_legacy_gsi": 0.3592507152694987,
216
+ "mean_lift_score": 0.2917478391505431,
217
+ "mean_motion_stability_score": 0.23222736227735297
218
+ }
219
+ },
220
+ "tool": {
221
+ "tool_selection_accuracy": null,
222
+ "tool_switch_success_rate": null,
223
+ "tool_switch_total": 0,
224
+ "tool_switch_successful": 0
225
+ },
226
+ "runtime": {
227
+ "avg_steps": 3293.84,
228
+ "error_count": 0
229
+ }
230
+ }
231
+ }
pyrua-dex/per_task_eval/27_ball_box_loading/Inv/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/27_ball_box_loading/Inv/summary.json ADDED
@@ -0,0 +1,233 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA ring_prefix_thumb_release-f1ac959cf9",
5
+ "base_seed": 102700000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 50,
9
+ "successes": 34,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3570,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.68,
50
+ "success_rate_ci": {
51
+ "mean": 0.68,
52
+ "ci_low": 0.54,
53
+ "ci_high": 0.8
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.905
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 52.49754901960615
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.68,
64
+ "hard_violation_rate": 0.12,
65
+ "drop_rate": 0.1,
66
+ "high_speed_violation_rate": 0.02,
67
+ "safety_violation_step_rate": 6.089577687787352e-05
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.68,
74
+ "at_end_success_rate": 0.68,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.68
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.865,
81
+ "current_chain_depth_progress_score": 0.93,
82
+ "latched_chain_depth_progress_score": 0.97,
83
+ "avg_current_chain_depth": 1.86,
84
+ "avg_latched_chain_depth": 1.94
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.75585014397087,
88
+ "expert_time_s": 39.666666666666664,
89
+ "expert_time_step": 2380,
90
+ "avg_steps_to_success": 3149.8529411764707,
91
+ "avg_policy_steps_to_success": 1050.2941176470588,
92
+ "avg_policy_queries_to_success": 1050.2941176470588
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 4.871662150229881e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_053_mini_soccer_ball_2": 3,
98
+ "drop_to_table_obj_056_tennis_ball_3": 2,
99
+ "drop_to_table_obj_144_table_tennis_5": 2,
100
+ "high_speed_obj_058_golf_ball_4": 1
101
+ },
102
+ "robot_constraint_violation_rate": 0.98,
103
+ "finger_joint_limit_saturation_rate": 0.98,
104
+ "joint_limit_active_step_rate": 0.06154736169046677,
105
+ "finger_joint_limit_saturation_step_rate": 0.0338945894102244,
106
+ "max_joint_limit_excess_rad": 1.2985665798187256,
107
+ "robot_constraint_diagnostics": {
108
+ "joint_limit_soft_counts": {
109
+ "left_finger4_joint1": 526,
110
+ "right_finger1_joint1": 370,
111
+ "right_finger1_joint3": 696,
112
+ "right_finger2_joint1": 2599,
113
+ "right_finger3_joint1": 679,
114
+ "right_finger4_joint1": 1176,
115
+ "right_finger4_joint2": 37,
116
+ "right_finger4_joint3": 2968,
117
+ "right_finger5_joint1": 2113,
118
+ "right_finger5_joint3": 2
119
+ },
120
+ "joint_limit_hard_counts": {
121
+ "left_finger4_joint1": 187,
122
+ "right_finger1_joint1": 15,
123
+ "right_finger1_joint2": 3,
124
+ "right_finger1_joint3": 479,
125
+ "right_finger2_joint1": 67,
126
+ "right_finger3_joint1": 17,
127
+ "right_finger4_joint1": 436,
128
+ "right_finger4_joint2": 39,
129
+ "right_finger4_joint3": 49,
130
+ "right_finger5_joint1": 41,
131
+ "right_finger5_joint3": 9
132
+ },
133
+ "joint_limit_hard_episode_events": 1206,
134
+ "finger_joint_limit_saturation_counts": {
135
+ "right_finger1_joint1": 245,
136
+ "right_finger1_joint3": 93,
137
+ "right_finger2_joint1": 1694,
138
+ "right_finger3_joint1": 240,
139
+ "right_finger4_joint1": 393,
140
+ "right_finger4_joint3": 2671,
141
+ "right_finger5_joint1": 664
142
+ },
143
+ "joint_limit_max_excess_by_joint": {
144
+ "left_finger1_joint1": 0.00020837783813476562,
145
+ "left_finger4_joint1": 0.07155676186084747,
146
+ "left_finger4_joint2": 0.0005711913108825684,
147
+ "right_finger1_joint1": 0.4934276342391968,
148
+ "right_finger1_joint2": 0.008814394474029541,
149
+ "right_finger1_joint3": 1.2985665798187256,
150
+ "right_finger2_joint1": 0.7581371665000916,
151
+ "right_finger2_joint2": 0.0015058517456054688,
152
+ "right_finger2_joint3": 0.0027426481246948242,
153
+ "right_finger2_joint4": 0.0008252859115600586,
154
+ "right_finger3_joint1": 0.6439911723136902,
155
+ "right_finger3_joint2": 0.0025108754634857178,
156
+ "right_finger3_joint3": 0.001021265983581543,
157
+ "right_finger3_joint4": 0.0005322694778442383,
158
+ "right_finger4_joint1": 0.8194855451583862,
159
+ "right_finger4_joint2": 0.07641583681106567,
160
+ "right_finger4_joint3": 0.9702248573303223,
161
+ "right_finger5_joint1": 0.9212390184402466,
162
+ "right_finger5_joint2": 0.0028456151485443115,
163
+ "right_finger5_joint3": 0.017028212547302246,
164
+ "right_finger5_joint4": 0.0030771493911743164
165
+ },
166
+ "coverage": {
167
+ "hard_limit_source": {
168
+ "available": 50,
169
+ "unavailable": 0,
170
+ "unknown": 0
171
+ },
172
+ "joint_role_manifest": {
173
+ "available": 50,
174
+ "unavailable": 0,
175
+ "unknown": 0
176
+ }
177
+ },
178
+ "hard_limit_sources": {
179
+ "runtime.joint_pos_limits": 50
180
+ },
181
+ "soft_limit_sources": {
182
+ "runtime.soft_joint_pos_limits": 50
183
+ },
184
+ "unknown_runtime_joint_names": [],
185
+ "baseline_checks": {
186
+ "post_home": {
187
+ "checked": 50,
188
+ "hard_violation_episodes": 0
189
+ },
190
+ "post_reset": {
191
+ "checked": 50,
192
+ "hard_violation_episodes": 0
193
+ }
194
+ }
195
+ }
196
+ },
197
+ "robot_motion": {
198
+ "joint_acceleration_rms": 11.657780765837785,
199
+ "joint_effort_rms": 18.40281538397183,
200
+ "joint_jerk_rms": 1162.6721515371567,
201
+ "joint_velocity_rms": 0.38106403051999266
202
+ },
203
+ "grasp": {
204
+ "kinematic_grasp_stability_index": 0.849159149274635,
205
+ "mean_kinematic_grasp_stability": 0.43313345993314833,
206
+ "per_object_best_kinematic_grasp_stability": {
207
+ "obj_053_mini_soccer_ball_2": 0.3290501587314744,
208
+ "obj_056_tennis_ball_3": 0.4874963590679336,
209
+ "obj_058_golf_ball_4": 0.6863097372318249,
210
+ "obj_144_table_tennis_5": 0.5181309713288422
211
+ },
212
+ "grasp_gsi_diagnostics": {
213
+ "mean_anti_slip_score": 0.5999450594154654,
214
+ "mean_duration_score": 0.45324354101155817,
215
+ "mean_hold_score": 0.23572901861995169,
216
+ "mean_kinematic_stability_score": 0.0407153719005185,
217
+ "mean_legacy_gsi": 0.34961871890935153,
218
+ "mean_lift_score": 0.2783148345496384,
219
+ "mean_motion_stability_score": 0.22102733997418575
220
+ }
221
+ },
222
+ "tool": {
223
+ "tool_selection_accuracy": null,
224
+ "tool_switch_success_rate": null,
225
+ "tool_switch_total": 0,
226
+ "tool_switch_successful": 0
227
+ },
228
+ "runtime": {
229
+ "avg_steps": 3284.3,
230
+ "error_count": 0
231
+ }
232
+ }
233
+ }
pyrua-dex/per_task_eval/27_ball_box_loading/None/native_shards/000/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/27_ball_box_loading/None/native_shards/000/summary.json ADDED
@@ -0,0 +1,227 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA ring_prefix_thumb_release-f1ac959cf9",
5
+ "base_seed": 102700000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 21,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3570,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.84,
50
+ "success_rate_ci": {
51
+ "mean": 0.84,
52
+ "ci_low": 0.68,
53
+ "ci_high": 0.96
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.92
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 52.80238095237924
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.84,
64
+ "hard_violation_rate": 0.0,
65
+ "drop_rate": 0.0,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 0.0
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.84,
74
+ "at_end_success_rate": 0.84,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.84
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.92,
81
+ "current_chain_depth_progress_score": 0.94,
82
+ "latched_chain_depth_progress_score": 0.94,
83
+ "avg_current_chain_depth": 1.88,
84
+ "avg_latched_chain_depth": 1.88
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7514279244290887,
88
+ "expert_time_s": 39.666666666666664,
89
+ "expert_time_step": 2380,
90
+ "avg_steps_to_success": 3168.1428571428573,
91
+ "avg_policy_steps_to_success": 1056.4761904761904,
92
+ "avg_policy_queries_to_success": 1056.4761904761904
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.0,
96
+ "violation_counts": {},
97
+ "robot_constraint_violation_rate": 0.96,
98
+ "finger_joint_limit_saturation_rate": 0.96,
99
+ "joint_limit_active_step_rate": 0.047147046812933885,
100
+ "finger_joint_limit_saturation_step_rate": 0.03212433950823527,
101
+ "max_joint_limit_excess_rad": 0.9710474610328674,
102
+ "robot_constraint_diagnostics": {
103
+ "joint_limit_soft_counts": {
104
+ "left_finger4_joint1": 248,
105
+ "right_finger1_joint1": 125,
106
+ "right_finger1_joint2": 4,
107
+ "right_finger1_joint3": 266,
108
+ "right_finger2_joint1": 1420,
109
+ "right_finger3_joint1": 638,
110
+ "right_finger4_joint1": 721,
111
+ "right_finger4_joint2": 3,
112
+ "right_finger4_joint3": 767,
113
+ "right_finger5_joint1": 520
114
+ },
115
+ "joint_limit_hard_counts": {
116
+ "left_finger4_joint1": 77,
117
+ "right_finger1_joint1": 3,
118
+ "right_finger1_joint2": 9,
119
+ "right_finger1_joint3": 220,
120
+ "right_finger2_joint1": 35,
121
+ "right_finger3_joint1": 16,
122
+ "right_finger4_joint1": 226,
123
+ "right_finger4_joint2": 15,
124
+ "right_finger4_joint3": 22,
125
+ "right_finger5_joint1": 14
126
+ },
127
+ "joint_limit_hard_episode_events": 575,
128
+ "finger_joint_limit_saturation_counts": {
129
+ "right_finger1_joint1": 59,
130
+ "right_finger1_joint3": 86,
131
+ "right_finger2_joint1": 889,
132
+ "right_finger3_joint1": 302,
133
+ "right_finger4_joint1": 672,
134
+ "right_finger4_joint3": 584,
135
+ "right_finger5_joint1": 455
136
+ },
137
+ "joint_limit_max_excess_by_joint": {
138
+ "left_finger1_joint1": 0.0002200603485107422,
139
+ "left_finger4_joint1": 0.07113239169120789,
140
+ "left_finger4_joint2": 0.0006005764007568359,
141
+ "right_finger1_joint1": 0.4800432622432709,
142
+ "right_finger1_joint2": 0.06551331281661987,
143
+ "right_finger1_joint3": 0.9710474610328674,
144
+ "right_finger2_joint1": 0.33577898144721985,
145
+ "right_finger2_joint2": 0.00034248828887939453,
146
+ "right_finger2_joint3": 0.0011271238327026367,
147
+ "right_finger2_joint4": 0.000924229621887207,
148
+ "right_finger3_joint1": 0.7435428500175476,
149
+ "right_finger3_joint2": 0.001273810863494873,
150
+ "right_finger3_joint3": 0.0010023117065429688,
151
+ "right_finger3_joint4": 0.0005896091461181641,
152
+ "right_finger4_joint1": 0.355190634727478,
153
+ "right_finger4_joint2": 0.013718575239181519,
154
+ "right_finger4_joint3": 0.9593667984008789,
155
+ "right_finger5_joint1": 0.7991753220558167,
156
+ "right_finger5_joint2": 0.003469139337539673,
157
+ "right_finger5_joint3": 0.004376530647277832,
158
+ "right_finger5_joint4": 0.003724336624145508
159
+ },
160
+ "coverage": {
161
+ "hard_limit_source": {
162
+ "available": 25,
163
+ "unavailable": 0,
164
+ "unknown": 0
165
+ },
166
+ "joint_role_manifest": {
167
+ "available": 25,
168
+ "unavailable": 0,
169
+ "unknown": 0
170
+ }
171
+ },
172
+ "hard_limit_sources": {
173
+ "runtime.joint_pos_limits": 25
174
+ },
175
+ "soft_limit_sources": {
176
+ "runtime.soft_joint_pos_limits": 25
177
+ },
178
+ "unknown_runtime_joint_names": [],
179
+ "baseline_checks": {
180
+ "post_home": {
181
+ "checked": 25,
182
+ "hard_violation_episodes": 0
183
+ },
184
+ "post_reset": {
185
+ "checked": 25,
186
+ "hard_violation_episodes": 0
187
+ }
188
+ }
189
+ }
190
+ },
191
+ "robot_motion": {
192
+ "joint_acceleration_rms": 11.465337338179843,
193
+ "joint_effort_rms": 18.967343857277434,
194
+ "joint_jerk_rms": 1144.3812325189824,
195
+ "joint_velocity_rms": 0.36257254403390915
196
+ },
197
+ "grasp": {
198
+ "kinematic_grasp_stability_index": 0.8165457096976999,
199
+ "mean_kinematic_grasp_stability": 0.4881889057372739,
200
+ "per_object_best_kinematic_grasp_stability": {
201
+ "obj_053_mini_soccer_ball_2": 0.3526970423394162,
202
+ "obj_056_tennis_ball_3": 0.4921962231379356,
203
+ "obj_058_golf_ball_4": 0.6986749012120916,
204
+ "obj_144_table_tennis_5": 0.4874299469057438
205
+ },
206
+ "grasp_gsi_diagnostics": {
207
+ "mean_anti_slip_score": 0.6105404075142035,
208
+ "mean_duration_score": 0.4424149540496084,
209
+ "mean_hold_score": 0.23439975663088902,
210
+ "mean_kinematic_stability_score": 0.048359245265091486,
211
+ "mean_legacy_gsi": 0.3472425949765853,
212
+ "mean_lift_score": 0.2745362632025638,
213
+ "mean_motion_stability_score": 0.21711085020854204
214
+ }
215
+ },
216
+ "tool": {
217
+ "tool_selection_accuracy": null,
218
+ "tool_switch_success_rate": null,
219
+ "tool_switch_total": 0,
220
+ "tool_switch_successful": 0
221
+ },
222
+ "runtime": {
223
+ "avg_steps": 3232.44,
224
+ "error_count": 0
225
+ }
226
+ }
227
+ }
pyrua-dex/per_task_eval/27_ball_box_loading/None/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/27_ball_box_loading/None/native_shards/001/summary.json ADDED
@@ -0,0 +1,236 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA ring_prefix_thumb_release-f1ac959cf9",
5
+ "base_seed": 102700000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 20,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3570,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.8,
50
+ "success_rate_ci": {
51
+ "mean": 0.8,
52
+ "ci_low": 0.64,
53
+ "ci_high": 0.96
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.95
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 52.5174999999983
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.8,
64
+ "hard_violation_rate": 0.08,
65
+ "drop_rate": 0.08,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 6.182686006108493e-05
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.8,
74
+ "at_end_success_rate": 0.8,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.8
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.89,
81
+ "current_chain_depth_progress_score": 0.92,
82
+ "latched_chain_depth_progress_score": 1.0,
83
+ "avg_current_chain_depth": 1.84,
84
+ "avg_latched_chain_depth": 2
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7555914956556185,
88
+ "expert_time_s": 39.666666666666664,
89
+ "expert_time_step": 2380,
90
+ "avg_steps_to_success": 3151.05,
91
+ "avg_policy_steps_to_success": 1050.8,
92
+ "avg_policy_queries_to_success": 1050.8
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 6.182686006108493e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_053_mini_soccer_ball_2": 2,
98
+ "drop_to_table_obj_056_tennis_ball_3": 2,
99
+ "drop_to_table_obj_058_golf_ball_4": 1
100
+ },
101
+ "robot_constraint_violation_rate": 1.0,
102
+ "finger_joint_limit_saturation_rate": 0.96,
103
+ "joint_limit_active_step_rate": 0.053146368908508615,
104
+ "finger_joint_limit_saturation_step_rate": 0.02725327991492624,
105
+ "max_joint_limit_excess_rad": 0.9282093048095703,
106
+ "robot_constraint_diagnostics": {
107
+ "joint_limit_soft_counts": {
108
+ "left_finger4_joint1": 293,
109
+ "right_finger1_joint1": 171,
110
+ "right_finger1_joint2": 6,
111
+ "right_finger1_joint3": 550,
112
+ "right_finger2_joint1": 1056,
113
+ "right_finger3_joint1": 349,
114
+ "right_finger3_joint3": 39,
115
+ "right_finger4_joint1": 221,
116
+ "right_finger4_joint2": 32,
117
+ "right_finger4_joint3": 823,
118
+ "right_finger5_joint1": 1269,
119
+ "right_finger5_joint3": 1
120
+ },
121
+ "joint_limit_hard_counts": {
122
+ "left_finger4_joint1": 90,
123
+ "right_finger1_joint1": 5,
124
+ "right_finger1_joint2": 5,
125
+ "right_finger1_joint3": 198,
126
+ "right_finger2_joint1": 35,
127
+ "right_finger3_joint1": 9,
128
+ "right_finger3_joint3": 1,
129
+ "right_finger4_joint1": 254,
130
+ "right_finger4_joint2": 20,
131
+ "right_finger4_joint3": 27,
132
+ "right_finger5_joint1": 11,
133
+ "right_finger5_joint3": 5
134
+ },
135
+ "joint_limit_hard_episode_events": 593,
136
+ "finger_joint_limit_saturation_counts": {
137
+ "right_finger1_joint1": 116,
138
+ "right_finger1_joint3": 72,
139
+ "right_finger2_joint1": 725,
140
+ "right_finger3_joint1": 131,
141
+ "right_finger3_joint3": 30,
142
+ "right_finger4_joint1": 87,
143
+ "right_finger4_joint3": 637,
144
+ "right_finger5_joint1": 549
145
+ },
146
+ "joint_limit_max_excess_by_joint": {
147
+ "left_finger1_joint1": 0.000244140625,
148
+ "left_finger4_joint1": 0.07483582198619843,
149
+ "left_finger4_joint2": 0.0005797445774078369,
150
+ "right_finger1_joint1": 0.7180949449539185,
151
+ "right_finger1_joint2": 0.0955558717250824,
152
+ "right_finger1_joint3": 0.1734459400177002,
153
+ "right_finger2_joint1": 0.4734323024749756,
154
+ "right_finger2_joint2": 0.0008789300918579102,
155
+ "right_finger2_joint3": 0.0013580322265625,
156
+ "right_finger2_joint4": 0.0007679462432861328,
157
+ "right_finger3_joint1": 0.18445196747779846,
158
+ "right_finger3_joint2": 0.0024466514587402344,
159
+ "right_finger3_joint3": 0.35392898321151733,
160
+ "right_finger3_joint4": 0.0006066560745239258,
161
+ "right_finger4_joint1": 0.6638602018356323,
162
+ "right_finger4_joint2": 0.0759023129940033,
163
+ "right_finger4_joint3": 0.9282093048095703,
164
+ "right_finger5_joint1": 0.8390316963195801,
165
+ "right_finger5_joint2": 0.0010249018669128418,
166
+ "right_finger5_joint3": 0.011232733726501465,
167
+ "right_finger5_joint4": 0.0035059452056884766
168
+ },
169
+ "coverage": {
170
+ "hard_limit_source": {
171
+ "available": 25,
172
+ "unavailable": 0,
173
+ "unknown": 0
174
+ },
175
+ "joint_role_manifest": {
176
+ "available": 25,
177
+ "unavailable": 0,
178
+ "unknown": 0
179
+ }
180
+ },
181
+ "hard_limit_sources": {
182
+ "runtime.joint_pos_limits": 25
183
+ },
184
+ "soft_limit_sources": {
185
+ "runtime.soft_joint_pos_limits": 25
186
+ },
187
+ "unknown_runtime_joint_names": [],
188
+ "baseline_checks": {
189
+ "post_home": {
190
+ "checked": 25,
191
+ "hard_violation_episodes": 0
192
+ },
193
+ "post_reset": {
194
+ "checked": 25,
195
+ "hard_violation_episodes": 0
196
+ }
197
+ }
198
+ }
199
+ },
200
+ "robot_motion": {
201
+ "joint_acceleration_rms": 12.466944565192858,
202
+ "joint_effort_rms": 18.460637903613108,
203
+ "joint_jerk_rms": 1247.5486477722427,
204
+ "joint_velocity_rms": 0.3914490032687148
205
+ },
206
+ "grasp": {
207
+ "kinematic_grasp_stability_index": 0.8618019291681619,
208
+ "mean_kinematic_grasp_stability": 0.47988325978017443,
209
+ "per_object_best_kinematic_grasp_stability": {
210
+ "obj_053_mini_soccer_ball_2": 0.331097105454289,
211
+ "obj_056_tennis_ball_3": 0.5360407077004498,
212
+ "obj_058_golf_ball_4": 0.6639441448529636,
213
+ "obj_144_table_tennis_5": 0.5622032632380057
214
+ },
215
+ "grasp_gsi_diagnostics": {
216
+ "mean_anti_slip_score": 0.6007092749008063,
217
+ "mean_duration_score": 0.4690250105017912,
218
+ "mean_hold_score": 0.24844999291759357,
219
+ "mean_kinematic_stability_score": 0.052549827111201756,
220
+ "mean_legacy_gsi": 0.3597703470044603,
221
+ "mean_lift_score": 0.2921464955934068,
222
+ "mean_motion_stability_score": 0.23201467115669444
223
+ }
224
+ },
225
+ "tool": {
226
+ "tool_selection_accuracy": null,
227
+ "tool_switch_success_rate": null,
228
+ "tool_switch_total": 0,
229
+ "tool_switch_successful": 0
230
+ },
231
+ "runtime": {
232
+ "avg_steps": 3234.84,
233
+ "error_count": 0
234
+ }
235
+ }
236
+ }
pyrua-dex/per_task_eval/27_ball_box_loading/None/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/27_ball_box_loading/None/summary.json ADDED
@@ -0,0 +1,236 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA ring_prefix_thumb_release-f1ac959cf9",
5
+ "base_seed": 102700000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 50,
9
+ "successes": 41,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3570,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.82,
50
+ "success_rate_ci": {
51
+ "mean": 0.82,
52
+ "ci_low": 0.72,
53
+ "ci_high": 0.92
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.935
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 52.66341463414464
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.82,
64
+ "hard_violation_rate": 0.04,
65
+ "drop_rate": 0.04,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 3.0924901968060764e-05
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.82,
74
+ "at_end_success_rate": 0.82,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.82
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.905,
81
+ "current_chain_depth_progress_score": 0.93,
82
+ "latched_chain_depth_progress_score": 0.97,
83
+ "avg_current_chain_depth": 1.86,
84
+ "avg_latched_chain_depth": 1.94
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7534589347834935,
88
+ "expert_time_s": 39.666666666666664,
89
+ "expert_time_step": 2380,
90
+ "avg_steps_to_success": 3159.8048780487807,
91
+ "avg_policy_steps_to_success": 1053.7073170731708,
92
+ "avg_policy_queries_to_success": 1053.7073170731708
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 3.0924901968060764e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_053_mini_soccer_ball_2": 2,
98
+ "drop_to_table_obj_056_tennis_ball_3": 2,
99
+ "drop_to_table_obj_058_golf_ball_4": 1
100
+ },
101
+ "robot_constraint_violation_rate": 0.98,
102
+ "finger_joint_limit_saturation_rate": 0.96,
103
+ "joint_limit_active_step_rate": 0.05014782103140733,
104
+ "finger_joint_limit_saturation_step_rate": 0.02968790588933833,
105
+ "max_joint_limit_excess_rad": 0.9710474610328674,
106
+ "robot_constraint_diagnostics": {
107
+ "joint_limit_soft_counts": {
108
+ "left_finger4_joint1": 541,
109
+ "right_finger1_joint1": 296,
110
+ "right_finger1_joint2": 10,
111
+ "right_finger1_joint3": 816,
112
+ "right_finger2_joint1": 2476,
113
+ "right_finger3_joint1": 987,
114
+ "right_finger3_joint3": 39,
115
+ "right_finger4_joint1": 942,
116
+ "right_finger4_joint2": 35,
117
+ "right_finger4_joint3": 1590,
118
+ "right_finger5_joint1": 1789,
119
+ "right_finger5_joint3": 1
120
+ },
121
+ "joint_limit_hard_counts": {
122
+ "left_finger4_joint1": 167,
123
+ "right_finger1_joint1": 8,
124
+ "right_finger1_joint2": 14,
125
+ "right_finger1_joint3": 418,
126
+ "right_finger2_joint1": 70,
127
+ "right_finger3_joint1": 25,
128
+ "right_finger3_joint3": 1,
129
+ "right_finger4_joint1": 480,
130
+ "right_finger4_joint2": 35,
131
+ "right_finger4_joint3": 49,
132
+ "right_finger5_joint1": 25,
133
+ "right_finger5_joint3": 5
134
+ },
135
+ "joint_limit_hard_episode_events": 1168,
136
+ "finger_joint_limit_saturation_counts": {
137
+ "right_finger1_joint1": 175,
138
+ "right_finger1_joint3": 158,
139
+ "right_finger2_joint1": 1614,
140
+ "right_finger3_joint1": 433,
141
+ "right_finger3_joint3": 30,
142
+ "right_finger4_joint1": 759,
143
+ "right_finger4_joint3": 1221,
144
+ "right_finger5_joint1": 1004
145
+ },
146
+ "joint_limit_max_excess_by_joint": {
147
+ "left_finger1_joint1": 0.000244140625,
148
+ "left_finger4_joint1": 0.07483582198619843,
149
+ "left_finger4_joint2": 0.0006005764007568359,
150
+ "right_finger1_joint1": 0.7180949449539185,
151
+ "right_finger1_joint2": 0.0955558717250824,
152
+ "right_finger1_joint3": 0.9710474610328674,
153
+ "right_finger2_joint1": 0.4734323024749756,
154
+ "right_finger2_joint2": 0.0008789300918579102,
155
+ "right_finger2_joint3": 0.0013580322265625,
156
+ "right_finger2_joint4": 0.000924229621887207,
157
+ "right_finger3_joint1": 0.7435428500175476,
158
+ "right_finger3_joint2": 0.0024466514587402344,
159
+ "right_finger3_joint3": 0.35392898321151733,
160
+ "right_finger3_joint4": 0.0006066560745239258,
161
+ "right_finger4_joint1": 0.6638602018356323,
162
+ "right_finger4_joint2": 0.0759023129940033,
163
+ "right_finger4_joint3": 0.9593667984008789,
164
+ "right_finger5_joint1": 0.8390316963195801,
165
+ "right_finger5_joint2": 0.003469139337539673,
166
+ "right_finger5_joint3": 0.011232733726501465,
167
+ "right_finger5_joint4": 0.003724336624145508
168
+ },
169
+ "coverage": {
170
+ "hard_limit_source": {
171
+ "available": 50,
172
+ "unavailable": 0,
173
+ "unknown": 0
174
+ },
175
+ "joint_role_manifest": {
176
+ "available": 50,
177
+ "unavailable": 0,
178
+ "unknown": 0
179
+ }
180
+ },
181
+ "hard_limit_sources": {
182
+ "runtime.joint_pos_limits": 50
183
+ },
184
+ "soft_limit_sources": {
185
+ "runtime.soft_joint_pos_limits": 50
186
+ },
187
+ "unknown_runtime_joint_names": [],
188
+ "baseline_checks": {
189
+ "post_home": {
190
+ "checked": 50,
191
+ "hard_violation_episodes": 0
192
+ },
193
+ "post_reset": {
194
+ "checked": 50,
195
+ "hard_violation_episodes": 0
196
+ }
197
+ }
198
+ }
199
+ },
200
+ "robot_motion": {
201
+ "joint_acceleration_rms": 11.96614095168635,
202
+ "joint_effort_rms": 18.71399088044527,
203
+ "joint_jerk_rms": 1195.9649401456127,
204
+ "joint_velocity_rms": 0.37701077365131197
205
+ },
206
+ "grasp": {
207
+ "kinematic_grasp_stability_index": 0.8391738194329309,
208
+ "mean_kinematic_grasp_stability": 0.483951331269366,
209
+ "per_object_best_kinematic_grasp_stability": {
210
+ "obj_053_mini_soccer_ball_2": 0.3418970738968526,
211
+ "obj_056_tennis_ball_3": 0.5141184654191927,
212
+ "obj_058_golf_ball_4": 0.6813095230325277,
213
+ "obj_144_table_tennis_5": 0.5248166050718748
214
+ },
215
+ "grasp_gsi_diagnostics": {
216
+ "mean_anti_slip_score": 0.6056248412075048,
217
+ "mean_duration_score": 0.45571998227569976,
218
+ "mean_hold_score": 0.2414248747742413,
219
+ "mean_kinematic_stability_score": 0.05045453618814662,
220
+ "mean_legacy_gsi": 0.3535064709905228,
221
+ "mean_lift_score": 0.2833413793979853,
222
+ "mean_motion_stability_score": 0.22456276068261824
223
+ }
224
+ },
225
+ "tool": {
226
+ "tool_selection_accuracy": null,
227
+ "tool_switch_success_rate": null,
228
+ "tool_switch_total": 0,
229
+ "tool_switch_successful": 0
230
+ },
231
+ "runtime": {
232
+ "avg_steps": 3233.64,
233
+ "error_count": 0
234
+ }
235
+ }
236
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/native_shards/000/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/native_shards/000/summary.json ADDED
@@ -0,0 +1,230 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 7,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.28,
50
+ "success_rate_ci": {
51
+ "mean": 0.28,
52
+ "ci_low": 0.12,
53
+ "ci_high": 0.44
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.71
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 38.78809523809432
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.28,
64
+ "hard_violation_rate": 0.08,
65
+ "drop_rate": 0.08,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 3.210324403280952e-05
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.32,
74
+ "at_end_success_rate": 0.32,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.28
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.69,
81
+ "current_chain_depth_progress_score": 0.82,
82
+ "latched_chain_depth_progress_score": 0.82,
83
+ "avg_current_chain_depth": 1.64,
84
+ "avg_latched_chain_depth": 1.64
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7326364172908444,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2327.285714285714,
91
+ "avg_policy_steps_to_success": 776.2857142857143,
92
+ "avg_policy_queries_to_success": 776.2857142857143
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 3.210324403280952e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_189_brush_5": 2
98
+ },
99
+ "robot_constraint_violation_rate": 0.48,
100
+ "finger_joint_limit_saturation_rate": 0.16,
101
+ "joint_limit_active_step_rate": 0.0036918730637730945,
102
+ "finger_joint_limit_saturation_step_rate": 0.00040129055041011896,
103
+ "max_joint_limit_excess_rad": 0.18280619382858276,
104
+ "robot_constraint_diagnostics": {
105
+ "joint_limit_soft_counts": {
106
+ "left_thumb_CMC_AA": 35,
107
+ "left_thumb_CMC_FE": 30,
108
+ "left_thumb_IP": 1,
109
+ "left_thumb_MCP_FE": 205,
110
+ "multi_right_index_MCP_FE": 13,
111
+ "multi_right_ring_MCP_FE": 8
112
+ },
113
+ "joint_limit_hard_counts": {
114
+ "left_thumb_CMC_AA": 12,
115
+ "left_thumb_CMC_FE": 6,
116
+ "left_thumb_IP": 1,
117
+ "left_thumb_MCP_FE": 8,
118
+ "multi_right_index_MCP_FE": 1,
119
+ "multi_right_middle_MCP_FE": 1,
120
+ "multi_right_ring_MCP_FE": 2,
121
+ "multi_right_thumb_MCP_FE": 4
122
+ },
123
+ "joint_limit_hard_episode_events": 23,
124
+ "finger_joint_limit_saturation_counts": {
125
+ "left_thumb_MCP_FE": 25
126
+ },
127
+ "joint_limit_max_excess_by_joint": {
128
+ "left_index_DIP": 3.204543099855073e-05,
129
+ "left_index_MCP_AA": 9.554624557495117e-05,
130
+ "left_index_PIP": 0.0001601288968231529,
131
+ "left_middle_DIP": 6.649220449617133e-05,
132
+ "left_middle_MCP_AA": 0.00026673078536987305,
133
+ "left_middle_PIP": 0.002455418696627021,
134
+ "left_pinky_CMC": 0.0009240270592272282,
135
+ "left_pinky_DIP": 0.00023516813234891742,
136
+ "left_pinky_PIP": 0.002236607251688838,
137
+ "left_ring_DIP": 0.00023639347637072206,
138
+ "left_ring_MCP_AA": 0.0002575814723968506,
139
+ "left_ring_PIP": 0.0025726533494889736,
140
+ "left_thumb_CMC_AA": 0.049435049295425415,
141
+ "left_thumb_CMC_FE": 0.0477139949798584,
142
+ "left_thumb_IP": 0.010958608239889145,
143
+ "left_thumb_MCP_FE": 0.18280619382858276,
144
+ "multi_right_index_DIP": 1.91191679732583e-06,
145
+ "multi_right_index_MCP_AA": 0.0004904270172119141,
146
+ "multi_right_index_MCP_FE": 0.03127044439315796,
147
+ "multi_right_index_PIP": 4.583731424645521e-05,
148
+ "multi_right_middle_DIP": 7.657383060966083e-11,
149
+ "multi_right_middle_MCP_AA": 0.0008530318737030029,
150
+ "multi_right_middle_MCP_FE": 0.006068050861358643,
151
+ "multi_right_middle_PIP": 1.8229566467464764e-10,
152
+ "multi_right_pinky_CMC": 0.00015467866614926606,
153
+ "multi_right_pinky_DIP": 0.00022075380547903478,
154
+ "multi_right_pinky_PIP": 0.0010524645913392305,
155
+ "multi_right_ring_DIP": 6.262936221901327e-05,
156
+ "multi_right_ring_MCP_AA": 0.001186370849609375,
157
+ "multi_right_ring_MCP_FE": 0.02263711392879486,
158
+ "multi_right_ring_PIP": 0.0003152181743644178,
159
+ "multi_right_thumb_CMC_FE": 0.0009893178939819336,
160
+ "multi_right_thumb_IP": 0.0031181832309812307,
161
+ "multi_right_thumb_MCP_FE": 0.0067231059074401855
162
+ },
163
+ "coverage": {
164
+ "hard_limit_source": {
165
+ "available": 25,
166
+ "unavailable": 0,
167
+ "unknown": 0
168
+ },
169
+ "joint_role_manifest": {
170
+ "available": 25,
171
+ "unavailable": 0,
172
+ "unknown": 0
173
+ }
174
+ },
175
+ "hard_limit_sources": {
176
+ "runtime.joint_pos_limits": 25
177
+ },
178
+ "soft_limit_sources": {
179
+ "runtime.soft_joint_pos_limits": 25
180
+ },
181
+ "unknown_runtime_joint_names": [],
182
+ "baseline_checks": {
183
+ "post_home": {
184
+ "checked": 25,
185
+ "hard_violation_episodes": 0
186
+ },
187
+ "post_reset": {
188
+ "checked": 25,
189
+ "hard_violation_episodes": 0
190
+ }
191
+ }
192
+ }
193
+ },
194
+ "robot_motion": {
195
+ "joint_acceleration_rms": 10.393687048121446,
196
+ "joint_effort_rms": 24.512865785162916,
197
+ "joint_jerk_rms": 1047.377248090862,
198
+ "joint_velocity_rms": 0.4173831000948758
199
+ },
200
+ "grasp": {
201
+ "kinematic_grasp_stability_index": 0.7143138348633663,
202
+ "mean_kinematic_grasp_stability": 0.4198849900956888,
203
+ "per_object_best_kinematic_grasp_stability": {
204
+ "obj_008_pudding_box_2": 0.2937304231270293,
205
+ "obj_009_gelatin_box_3": 0.39424965472258344,
206
+ "obj_033_spatula_4": 0.6021628529418348,
207
+ "obj_189_brush_5": 0.2873377125865381
208
+ },
209
+ "grasp_gsi_diagnostics": {
210
+ "mean_anti_slip_score": 0.5783507411450234,
211
+ "mean_duration_score": 0.4571239011665827,
212
+ "mean_hold_score": 0.4139043581581523,
213
+ "mean_kinematic_stability_score": 0.03741622546371758,
214
+ "mean_legacy_gsi": 0.4469774911538093,
215
+ "mean_lift_score": 0.4357513156445447,
216
+ "mean_motion_stability_score": 0.3894202043457795
217
+ }
218
+ },
219
+ "tool": {
220
+ "tool_selection_accuracy": null,
221
+ "tool_switch_success_rate": null,
222
+ "tool_switch_total": 0,
223
+ "tool_switch_successful": 0
224
+ },
225
+ "runtime": {
226
+ "avg_steps": 2491.96,
227
+ "error_count": 0
228
+ }
229
+ }
230
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/native_shards/001/summary.json ADDED
@@ -0,0 +1,230 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 2,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.08,
50
+ "success_rate_ci": {
51
+ "mean": 0.08,
52
+ "ci_low": 0.0,
53
+ "ci_high": 0.2
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.54
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 39.73333333333236
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.08,
64
+ "hard_violation_rate": 0.08,
65
+ "drop_rate": 0.08,
66
+ "high_speed_violation_rate": 0.04,
67
+ "safety_violation_step_rate": 0.02426206809742589
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.08,
74
+ "at_end_success_rate": 0.08,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.08
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.49,
81
+ "current_chain_depth_progress_score": 0.62,
82
+ "latched_chain_depth_progress_score": 0.66,
83
+ "avg_current_chain_depth": 1.24,
84
+ "avg_latched_chain_depth": 1.32
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7145467959765435,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2384,
91
+ "avg_policy_steps_to_success": 795,
92
+ "avg_policy_queries_to_success": 795
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 9.440493423122915e-05,
96
+ "violation_counts": {
97
+ "drop_obj_008_pudding_box_2": 1,
98
+ "drop_to_table_obj_008_pudding_box_2": 1,
99
+ "drop_to_table_obj_009_gelatin_box_3": 2,
100
+ "high_speed_obj_008_pudding_box_2": 1,
101
+ "high_speed_obj_130_tray_1": 1
102
+ },
103
+ "robot_constraint_violation_rate": 0.6,
104
+ "finger_joint_limit_saturation_rate": 0.0,
105
+ "joint_limit_active_step_rate": 0.0012902007678267985,
106
+ "finger_joint_limit_saturation_step_rate": 0.0,
107
+ "max_joint_limit_excess_rad": 0.21823371946811676,
108
+ "robot_constraint_diagnostics": {
109
+ "joint_limit_soft_counts": {
110
+ "left_thumb_CMC_AA": 11,
111
+ "left_thumb_IP": 40,
112
+ "left_thumb_MCP_FE": 16,
113
+ "multi_right_ring_MCP_FE": 26
114
+ },
115
+ "joint_limit_hard_counts": {
116
+ "left_thumb_CMC_AA": 3,
117
+ "left_thumb_CMC_FE": 9,
118
+ "left_thumb_IP": 5,
119
+ "left_thumb_MCP_FE": 2,
120
+ "multi_right_middle_MCP_FE": 1,
121
+ "multi_right_pinky_CMC": 6,
122
+ "multi_right_ring_MCP_FE": 4,
123
+ "multi_right_thumb_MCP_FE": 7
124
+ },
125
+ "joint_limit_hard_episode_events": 33,
126
+ "finger_joint_limit_saturation_counts": {},
127
+ "joint_limit_max_excess_by_joint": {
128
+ "left_index_DIP": 1.1528239156177733e-05,
129
+ "left_index_MCP_AA": 9.962916374206543e-05,
130
+ "left_index_PIP": 5.894372225156985e-05,
131
+ "left_middle_DIP": 0.00010698544792830944,
132
+ "left_middle_MCP_AA": 0.0006248354911804199,
133
+ "left_middle_PIP": 0.0021947845816612244,
134
+ "left_pinky_CMC": 0.0026631131768226624,
135
+ "left_pinky_DIP": 0.0002770415158011019,
136
+ "left_pinky_PIP": 0.0033714687451720238,
137
+ "left_ring_DIP": 0.0002563894377090037,
138
+ "left_ring_MCP_AA": 0.0014854073524475098,
139
+ "left_ring_PIP": 0.001857868628576398,
140
+ "left_thumb_CMC_AA": 0.027220696210861206,
141
+ "left_thumb_CMC_FE": 0.00819551944732666,
142
+ "left_thumb_IP": 0.21823371946811676,
143
+ "left_thumb_MCP_FE": 0.11968910694122314,
144
+ "multi_right_index_DIP": 1.644961571400927e-06,
145
+ "multi_right_index_MCP_AA": 0.00024199485778808594,
146
+ "multi_right_index_MCP_FE": 0.0002470463514328003,
147
+ "multi_right_index_PIP": 4.5257038436830044e-05,
148
+ "multi_right_middle_DIP": 7.657401795979624e-11,
149
+ "multi_right_middle_MCP_AA": 0.0014006197452545166,
150
+ "multi_right_middle_MCP_FE": 0.005063697695732117,
151
+ "multi_right_middle_PIP": 1.8036007409794053e-10,
152
+ "multi_right_pinky_CMC": 0.00726401386782527,
153
+ "multi_right_pinky_DIP": 0.00017510003817733377,
154
+ "multi_right_pinky_PIP": 0.0038751193787902594,
155
+ "multi_right_ring_DIP": 5.110819256515242e-05,
156
+ "multi_right_ring_MCP_AA": 0.0011217594146728516,
157
+ "multi_right_ring_MCP_FE": 0.04040217399597168,
158
+ "multi_right_ring_PIP": 0.0002557076804805547,
159
+ "multi_right_thumb_CMC_FE": 7.677078247070312e-05,
160
+ "multi_right_thumb_IP": 9.155570296570659e-06,
161
+ "multi_right_thumb_MCP_FE": 0.0057648420333862305
162
+ },
163
+ "coverage": {
164
+ "hard_limit_source": {
165
+ "available": 25,
166
+ "unavailable": 0,
167
+ "unknown": 0
168
+ },
169
+ "joint_role_manifest": {
170
+ "available": 25,
171
+ "unavailable": 0,
172
+ "unknown": 0
173
+ }
174
+ },
175
+ "hard_limit_sources": {
176
+ "runtime.joint_pos_limits": 25
177
+ },
178
+ "soft_limit_sources": {
179
+ "runtime.soft_joint_pos_limits": 25
180
+ },
181
+ "unknown_runtime_joint_names": [],
182
+ "baseline_checks": {
183
+ "post_home": {
184
+ "checked": 25,
185
+ "hard_violation_episodes": 0
186
+ },
187
+ "post_reset": {
188
+ "checked": 25,
189
+ "hard_violation_episodes": 0
190
+ }
191
+ }
192
+ }
193
+ },
194
+ "robot_motion": {
195
+ "joint_acceleration_rms": 9.67006188940952,
196
+ "joint_effort_rms": 21.403726672053665,
197
+ "joint_jerk_rms": 977.3215968140937,
198
+ "joint_velocity_rms": 0.34180720254094377
199
+ },
200
+ "grasp": {
201
+ "kinematic_grasp_stability_index": 0.6426634478416016,
202
+ "mean_kinematic_grasp_stability": 0.5007471230047726,
203
+ "per_object_best_kinematic_grasp_stability": {
204
+ "obj_008_pudding_box_2": 0.2874181310198293,
205
+ "obj_009_gelatin_box_3": 0.37468166132574443,
206
+ "obj_033_spatula_4": 0.5757445302709522,
207
+ "obj_189_brush_5": 0.10840220400826517
208
+ },
209
+ "grasp_gsi_diagnostics": {
210
+ "mean_anti_slip_score": 0.6029577429093188,
211
+ "mean_duration_score": 0.442036991355562,
212
+ "mean_hold_score": 0.3763447797418784,
213
+ "mean_kinematic_stability_score": 0.05107085391616294,
214
+ "mean_legacy_gsi": 0.4248665353206085,
215
+ "mean_lift_score": 0.392041877263725,
216
+ "mean_motion_stability_score": 0.3494558041962596
217
+ }
218
+ },
219
+ "tool": {
220
+ "tool_selection_accuracy": null,
221
+ "tool_switch_success_rate": null,
222
+ "tool_switch_total": 0,
223
+ "tool_switch_successful": 0
224
+ },
225
+ "runtime": {
226
+ "avg_steps": 2542.24,
227
+ "error_count": 0
228
+ }
229
+ }
230
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Equi/summary.json ADDED
@@ -0,0 +1,236 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 50,
9
+ "successes": 9,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.18,
50
+ "success_rate_ci": {
51
+ "mean": 0.18,
52
+ "ci_low": 0.08,
53
+ "ci_high": 0.28049999999999276
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.625
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 38.998148148147216
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.18,
64
+ "hard_violation_rate": 0.08,
65
+ "drop_rate": 0.08,
66
+ "high_speed_violation_rate": 0.02,
67
+ "safety_violation_step_rate": 0.01226808628977792
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.2,
74
+ "at_end_success_rate": 0.2,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.18
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.59,
81
+ "current_chain_depth_progress_score": 0.72,
82
+ "latched_chain_depth_progress_score": 0.74,
83
+ "avg_current_chain_depth": 1.44,
84
+ "avg_latched_chain_depth": 1.48
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.728616501443222,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2339.8888888888887,
91
+ "avg_policy_steps_to_success": 780.4444444444445,
92
+ "avg_policy_queries_to_success": 780.4444444444445
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 6.356521393667315e-05,
96
+ "violation_counts": {
97
+ "drop_obj_008_pudding_box_2": 1,
98
+ "drop_to_table_obj_008_pudding_box_2": 1,
99
+ "drop_to_table_obj_009_gelatin_box_3": 2,
100
+ "drop_to_table_obj_189_brush_5": 2,
101
+ "high_speed_obj_008_pudding_box_2": 1,
102
+ "high_speed_obj_130_tray_1": 1
103
+ },
104
+ "robot_constraint_violation_rate": 0.54,
105
+ "finger_joint_limit_saturation_rate": 0.08,
106
+ "joint_limit_active_step_rate": 0.002479043343530253,
107
+ "finger_joint_limit_saturation_step_rate": 0.0001986412935521036,
108
+ "max_joint_limit_excess_rad": 0.21823371946811676,
109
+ "robot_constraint_diagnostics": {
110
+ "joint_limit_soft_counts": {
111
+ "left_thumb_CMC_AA": 46,
112
+ "left_thumb_CMC_FE": 30,
113
+ "left_thumb_IP": 41,
114
+ "left_thumb_MCP_FE": 221,
115
+ "multi_right_index_MCP_FE": 13,
116
+ "multi_right_ring_MCP_FE": 34
117
+ },
118
+ "joint_limit_hard_counts": {
119
+ "left_thumb_CMC_AA": 15,
120
+ "left_thumb_CMC_FE": 15,
121
+ "left_thumb_IP": 6,
122
+ "left_thumb_MCP_FE": 10,
123
+ "multi_right_index_MCP_FE": 1,
124
+ "multi_right_middle_MCP_FE": 2,
125
+ "multi_right_pinky_CMC": 6,
126
+ "multi_right_ring_MCP_FE": 6,
127
+ "multi_right_thumb_MCP_FE": 11
128
+ },
129
+ "joint_limit_hard_episode_events": 56,
130
+ "finger_joint_limit_saturation_counts": {
131
+ "left_thumb_MCP_FE": 25
132
+ },
133
+ "joint_limit_max_excess_by_joint": {
134
+ "left_index_DIP": 3.204543099855073e-05,
135
+ "left_index_MCP_AA": 9.962916374206543e-05,
136
+ "left_index_PIP": 0.0001601288968231529,
137
+ "left_middle_DIP": 0.00010698544792830944,
138
+ "left_middle_MCP_AA": 0.0006248354911804199,
139
+ "left_middle_PIP": 0.002455418696627021,
140
+ "left_pinky_CMC": 0.0026631131768226624,
141
+ "left_pinky_DIP": 0.0002770415158011019,
142
+ "left_pinky_PIP": 0.0033714687451720238,
143
+ "left_ring_DIP": 0.0002563894377090037,
144
+ "left_ring_MCP_AA": 0.0014854073524475098,
145
+ "left_ring_PIP": 0.0025726533494889736,
146
+ "left_thumb_CMC_AA": 0.049435049295425415,
147
+ "left_thumb_CMC_FE": 0.0477139949798584,
148
+ "left_thumb_IP": 0.21823371946811676,
149
+ "left_thumb_MCP_FE": 0.18280619382858276,
150
+ "multi_right_index_DIP": 1.91191679732583e-06,
151
+ "multi_right_index_MCP_AA": 0.0004904270172119141,
152
+ "multi_right_index_MCP_FE": 0.03127044439315796,
153
+ "multi_right_index_PIP": 4.583731424645521e-05,
154
+ "multi_right_middle_DIP": 7.657401795979624e-11,
155
+ "multi_right_middle_MCP_AA": 0.0014006197452545166,
156
+ "multi_right_middle_MCP_FE": 0.006068050861358643,
157
+ "multi_right_middle_PIP": 1.8229566467464764e-10,
158
+ "multi_right_pinky_CMC": 0.00726401386782527,
159
+ "multi_right_pinky_DIP": 0.00022075380547903478,
160
+ "multi_right_pinky_PIP": 0.0038751193787902594,
161
+ "multi_right_ring_DIP": 6.262936221901327e-05,
162
+ "multi_right_ring_MCP_AA": 0.001186370849609375,
163
+ "multi_right_ring_MCP_FE": 0.04040217399597168,
164
+ "multi_right_ring_PIP": 0.0003152181743644178,
165
+ "multi_right_thumb_CMC_FE": 0.0009893178939819336,
166
+ "multi_right_thumb_IP": 0.0031181832309812307,
167
+ "multi_right_thumb_MCP_FE": 0.0067231059074401855
168
+ },
169
+ "coverage": {
170
+ "hard_limit_source": {
171
+ "available": 50,
172
+ "unavailable": 0,
173
+ "unknown": 0
174
+ },
175
+ "joint_role_manifest": {
176
+ "available": 50,
177
+ "unavailable": 0,
178
+ "unknown": 0
179
+ }
180
+ },
181
+ "hard_limit_sources": {
182
+ "runtime.joint_pos_limits": 50
183
+ },
184
+ "soft_limit_sources": {
185
+ "runtime.soft_joint_pos_limits": 50
186
+ },
187
+ "unknown_runtime_joint_names": [],
188
+ "baseline_checks": {
189
+ "post_home": {
190
+ "checked": 50,
191
+ "hard_violation_episodes": 0
192
+ },
193
+ "post_reset": {
194
+ "checked": 50,
195
+ "hard_violation_episodes": 0
196
+ }
197
+ }
198
+ }
199
+ },
200
+ "robot_motion": {
201
+ "joint_acceleration_rms": 10.031874468765483,
202
+ "joint_effort_rms": 22.95829622860829,
203
+ "joint_jerk_rms": 1012.3494224524778,
204
+ "joint_velocity_rms": 0.3795951513179098
205
+ },
206
+ "grasp": {
207
+ "kinematic_grasp_stability_index": 0.678488641352484,
208
+ "mean_kinematic_grasp_stability": 0.4603160565502308,
209
+ "per_object_best_kinematic_grasp_stability": {
210
+ "obj_008_pudding_box_2": 0.2905742770734293,
211
+ "obj_009_gelatin_box_3": 0.38446565802416394,
212
+ "obj_033_spatula_4": 0.5889536916063934,
213
+ "obj_189_brush_5": 0.19786995829740164
214
+ },
215
+ "grasp_gsi_diagnostics": {
216
+ "mean_anti_slip_score": 0.5906542420271711,
217
+ "mean_duration_score": 0.44958044626107235,
218
+ "mean_hold_score": 0.39512456895001535,
219
+ "mean_kinematic_stability_score": 0.044243539689940256,
220
+ "mean_legacy_gsi": 0.4359220132372089,
221
+ "mean_lift_score": 0.4138965964541349,
222
+ "mean_motion_stability_score": 0.3694380042710195
223
+ }
224
+ },
225
+ "tool": {
226
+ "tool_selection_accuracy": null,
227
+ "tool_switch_success_rate": null,
228
+ "tool_switch_total": 0,
229
+ "tool_switch_successful": 0
230
+ },
231
+ "runtime": {
232
+ "avg_steps": 2517.1,
233
+ "error_count": 0
234
+ }
235
+ }
236
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/native_shards/000/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/native_shards/000/summary.json ADDED
@@ -0,0 +1,238 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 203200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 7,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.28,
50
+ "success_rate_ci": {
51
+ "mean": 0.28,
52
+ "ci_low": 0.12,
53
+ "ci_high": 0.48
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.77
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 38.090476190475314
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.28,
64
+ "hard_violation_rate": 0.08,
65
+ "drop_rate": 0.08,
66
+ "high_speed_violation_rate": 0.04,
67
+ "safety_violation_step_rate": 0.0231590491242783
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.32,
74
+ "at_end_success_rate": 0.28,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.28
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.74,
81
+ "current_chain_depth_progress_score": 0.88,
82
+ "latched_chain_depth_progress_score": 0.92,
83
+ "avg_current_chain_depth": 1.76,
84
+ "avg_latched_chain_depth": 1.84
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.746113582842515,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2285.4285714285716,
91
+ "avg_policy_steps_to_success": 762.1428571428571,
92
+ "avg_policy_queries_to_success": 762.1428571428571
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 9.676482921007644e-05,
96
+ "violation_counts": {
97
+ "drop_obj_008_pudding_box_2": 1,
98
+ "drop_to_table_obj_008_pudding_box_2": 1,
99
+ "drop_to_table_obj_009_gelatin_box_3": 1,
100
+ "drop_to_table_obj_189_brush_5": 1,
101
+ "high_speed_obj_008_pudding_box_2": 1,
102
+ "high_speed_obj_130_tray_1": 1
103
+ },
104
+ "robot_constraint_violation_rate": 0.52,
105
+ "finger_joint_limit_saturation_rate": 0.04,
106
+ "joint_limit_active_step_rate": 0.0022901009579718093,
107
+ "finger_joint_limit_saturation_step_rate": 4.838241460503822e-05,
108
+ "max_joint_limit_excess_rad": 0.14987415075302124,
109
+ "robot_constraint_diagnostics": {
110
+ "joint_limit_soft_counts": {
111
+ "left_pinky_MCP_FE": 7,
112
+ "left_thumb_CMC_AA": 26,
113
+ "left_thumb_CMC_FE": 20,
114
+ "left_thumb_IP": 7,
115
+ "left_thumb_MCP_FE": 113
116
+ },
117
+ "joint_limit_hard_counts": {
118
+ "left_pinky_CMC": 3,
119
+ "left_pinky_MCP_FE": 1,
120
+ "left_pinky_PIP": 1,
121
+ "left_thumb_CMC_AA": 7,
122
+ "left_thumb_CMC_FE": 17,
123
+ "left_thumb_IP": 5,
124
+ "left_thumb_MCP_FE": 7,
125
+ "multi_right_pinky_CMC": 1,
126
+ "multi_right_thumb_IP": 1,
127
+ "multi_right_thumb_MCP_FE": 1
128
+ },
129
+ "joint_limit_hard_episode_events": 35,
130
+ "finger_joint_limit_saturation_counts": {
131
+ "left_thumb_MCP_FE": 3
132
+ },
133
+ "joint_limit_max_excess_by_joint": {
134
+ "left_index_DIP": 3.0825594876660034e-05,
135
+ "left_index_MCP_AA": 0.0001315772533416748,
136
+ "left_index_PIP": 0.0017800049390643835,
137
+ "left_middle_DIP": 0.00014326437667477876,
138
+ "left_middle_MCP_AA": 0.0004926919937133789,
139
+ "left_middle_PIP": 0.00488057266920805,
140
+ "left_pinky_CMC": 0.006102359853684902,
141
+ "left_pinky_DIP": 0.00021328224102035165,
142
+ "left_pinky_MCP_FE": 0.03453095257282257,
143
+ "left_pinky_PIP": 0.005239320453256369,
144
+ "left_ring_DIP": 0.0001839229662436992,
145
+ "left_ring_MCP_AA": 0.0005010366439819336,
146
+ "left_ring_PIP": 0.0023643861059099436,
147
+ "left_thumb_CMC_AA": 0.06530192494392395,
148
+ "left_thumb_CMC_FE": 0.025467753410339355,
149
+ "left_thumb_IP": 0.03799530118703842,
150
+ "left_thumb_MCP_FE": 0.14987415075302124,
151
+ "multi_A3": 1.5735626220703125e-05,
152
+ "multi_right_index_DIP": 1.52639574935165e-06,
153
+ "multi_right_index_MCP_AA": 0.0002525448799133301,
154
+ "multi_right_index_MCP_FE": 0.0008076131343841553,
155
+ "multi_right_index_PIP": 4.5763288653688505e-05,
156
+ "multi_right_middle_DIP": 7.657401795979624e-11,
157
+ "multi_right_middle_MCP_AA": 0.00035515427589416504,
158
+ "multi_right_middle_MCP_FE": 0.0028501152992248535,
159
+ "multi_right_middle_PIP": 1.2028865403745925e-10,
160
+ "multi_right_pinky_CMC": 0.005388373974710703,
161
+ "multi_right_pinky_DIP": 0.0002515497035346925,
162
+ "multi_right_pinky_PIP": 0.003020935459062457,
163
+ "multi_right_ring_DIP": 0.00024057368864305317,
164
+ "multi_right_ring_MCP_AA": 0.0008302032947540283,
165
+ "multi_right_ring_MCP_FE": 0.0033588409423828125,
166
+ "multi_right_ring_PIP": 0.0011946060694754124,
167
+ "multi_right_thumb_CMC_FE": 0.00118255615234375,
168
+ "multi_right_thumb_IP": 0.006918040104210377,
169
+ "multi_right_thumb_MCP_FE": 0.005062222480773926
170
+ },
171
+ "coverage": {
172
+ "hard_limit_source": {
173
+ "available": 25,
174
+ "unavailable": 0,
175
+ "unknown": 0
176
+ },
177
+ "joint_role_manifest": {
178
+ "available": 25,
179
+ "unavailable": 0,
180
+ "unknown": 0
181
+ }
182
+ },
183
+ "hard_limit_sources": {
184
+ "runtime.joint_pos_limits": 25
185
+ },
186
+ "soft_limit_sources": {
187
+ "runtime.soft_joint_pos_limits": 25
188
+ },
189
+ "unknown_runtime_joint_names": [],
190
+ "baseline_checks": {
191
+ "post_home": {
192
+ "checked": 25,
193
+ "hard_violation_episodes": 0
194
+ },
195
+ "post_reset": {
196
+ "checked": 25,
197
+ "hard_violation_episodes": 0
198
+ }
199
+ }
200
+ }
201
+ },
202
+ "robot_motion": {
203
+ "joint_acceleration_rms": 11.70999806465987,
204
+ "joint_effort_rms": 26.459411562609887,
205
+ "joint_jerk_rms": 1184.7980099697734,
206
+ "joint_velocity_rms": 0.38476936188150274
207
+ },
208
+ "grasp": {
209
+ "kinematic_grasp_stability_index": 0.7169885640156849,
210
+ "mean_kinematic_grasp_stability": 0.4167027293286003,
211
+ "per_object_best_kinematic_grasp_stability": {
212
+ "obj_008_pudding_box_2": 0.29892474345950126,
213
+ "obj_009_gelatin_box_3": 0.4441370547604748,
214
+ "obj_033_spatula_4": 0.6310077707229708,
215
+ "obj_189_brush_5": 0.5006086195275509
216
+ },
217
+ "grasp_gsi_diagnostics": {
218
+ "mean_anti_slip_score": 0.5356992416235303,
219
+ "mean_duration_score": 0.49267816526645775,
220
+ "mean_hold_score": 0.41942185443637287,
221
+ "mean_kinematic_stability_score": 0.029227359058599363,
222
+ "mean_legacy_gsi": 0.4466148216685,
223
+ "mean_lift_score": 0.4400854770209935,
224
+ "mean_motion_stability_score": 0.3898218364465351
225
+ }
226
+ },
227
+ "tool": {
228
+ "tool_selection_accuracy": null,
229
+ "tool_switch_success_rate": null,
230
+ "tool_switch_total": 0,
231
+ "tool_switch_successful": 0
232
+ },
233
+ "runtime": {
234
+ "avg_steps": 2480.24,
235
+ "error_count": 0
236
+ }
237
+ }
238
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/native_shards/001/summary.json ADDED
@@ -0,0 +1,230 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 203200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 4,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.16,
50
+ "success_rate_ci": {
51
+ "mean": 0.16,
52
+ "ci_low": 0.04,
53
+ "ci_high": 0.32
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.64
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 37.412499999999156
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.16,
64
+ "hard_violation_rate": 0.08,
65
+ "drop_rate": 0.08,
66
+ "high_speed_violation_rate": 0.04,
67
+ "safety_violation_step_rate": 0.022935120900167586
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.16,
74
+ "at_end_success_rate": 0.16,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.16
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.63,
81
+ "current_chain_depth_progress_score": 0.78,
82
+ "latched_chain_depth_progress_score": 0.8,
83
+ "avg_current_chain_depth": 1.56,
84
+ "avg_latched_chain_depth": 1.6
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7587705641771512,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2244.75,
91
+ "avg_policy_steps_to_success": 748.5,
92
+ "avg_policy_queries_to_success": 748.5
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 9.576250897773521e-05,
96
+ "violation_counts": {
97
+ "drop_obj_008_pudding_box_2": 1,
98
+ "drop_to_table_obj_008_pudding_box_2": 1,
99
+ "drop_to_table_obj_009_gelatin_box_3": 2,
100
+ "high_speed_obj_008_pudding_box_2": 1,
101
+ "high_speed_obj_130_tray_1": 1
102
+ },
103
+ "robot_constraint_violation_rate": 0.44,
104
+ "finger_joint_limit_saturation_rate": 0.12,
105
+ "joint_limit_active_step_rate": 0.0031282419599393506,
106
+ "finger_joint_limit_saturation_step_rate": 0.0013087542893623814,
107
+ "max_joint_limit_excess_rad": 0.24029040336608887,
108
+ "robot_constraint_diagnostics": {
109
+ "joint_limit_soft_counts": {
110
+ "left_thumb_CMC_AA": 24,
111
+ "left_thumb_CMC_FE": 18,
112
+ "left_thumb_MCP_FE": 184
113
+ },
114
+ "joint_limit_hard_counts": {
115
+ "left_thumb_CMC_AA": 10,
116
+ "left_thumb_CMC_FE": 10,
117
+ "left_thumb_MCP_FE": 7,
118
+ "multi_right_middle_MCP_FE": 1,
119
+ "multi_right_pinky_CMC": 1,
120
+ "multi_right_ring_MCP_FE": 2,
121
+ "multi_right_thumb_MCP_FE": 3
122
+ },
123
+ "joint_limit_hard_episode_events": 22,
124
+ "finger_joint_limit_saturation_counts": {
125
+ "left_thumb_MCP_FE": 82
126
+ },
127
+ "joint_limit_max_excess_by_joint": {
128
+ "left_index_DIP": 3.342629861435853e-05,
129
+ "left_index_MCP_AA": 0.00010222196578979492,
130
+ "left_index_PIP": 0.0008091848576441407,
131
+ "left_middle_DIP": 0.00013017075252719223,
132
+ "left_middle_MCP_AA": 0.0005394816398620605,
133
+ "left_middle_PIP": 0.002738075563684106,
134
+ "left_pinky_CMC": 0.0022819603327661753,
135
+ "left_pinky_DIP": 0.00018430969794280827,
136
+ "left_pinky_PIP": 0.002066470682621002,
137
+ "left_ring_DIP": 0.0001628250174690038,
138
+ "left_ring_MCP_AA": 0.0005986690521240234,
139
+ "left_ring_PIP": 0.0030039905104786158,
140
+ "left_thumb_CMC_AA": 0.03819581866264343,
141
+ "left_thumb_CMC_FE": 0.056396484375,
142
+ "left_thumb_IP": 4.4074749894207343e-05,
143
+ "left_thumb_MCP_FE": 0.24029040336608887,
144
+ "multi_right_index_DIP": 1.0285930329700932e-06,
145
+ "multi_right_index_MCP_AA": 0.00030609965324401855,
146
+ "multi_right_index_MCP_FE": 0.003085196018218994,
147
+ "multi_right_index_PIP": 0.004464675672352314,
148
+ "multi_right_middle_DIP": 7.662199347224785e-11,
149
+ "multi_right_middle_MCP_AA": 0.00040149688720703125,
150
+ "multi_right_middle_MCP_FE": 0.005505815148353577,
151
+ "multi_right_middle_PIP": 7.468960866452434e-11,
152
+ "multi_right_pinky_CMC": 0.005548879038542509,
153
+ "multi_right_pinky_DIP": 0.00014481331163551658,
154
+ "multi_right_pinky_PIP": 0.0036197053268551826,
155
+ "multi_right_ring_DIP": 5.122147558722645e-05,
156
+ "multi_right_ring_MCP_AA": 0.0009909272193908691,
157
+ "multi_right_ring_MCP_FE": 0.006244748830795288,
158
+ "multi_right_ring_PIP": 0.00025599816581234336,
159
+ "multi_right_thumb_CMC_FE": 0.0001087188720703125,
160
+ "multi_right_thumb_IP": 0.0014656420098617673,
161
+ "multi_right_thumb_MCP_FE": 0.005576014518737793
162
+ },
163
+ "coverage": {
164
+ "hard_limit_source": {
165
+ "available": 25,
166
+ "unavailable": 0,
167
+ "unknown": 0
168
+ },
169
+ "joint_role_manifest": {
170
+ "available": 25,
171
+ "unavailable": 0,
172
+ "unknown": 0
173
+ }
174
+ },
175
+ "hard_limit_sources": {
176
+ "runtime.joint_pos_limits": 25
177
+ },
178
+ "soft_limit_sources": {
179
+ "runtime.soft_joint_pos_limits": 25
180
+ },
181
+ "unknown_runtime_joint_names": [],
182
+ "baseline_checks": {
183
+ "post_home": {
184
+ "checked": 25,
185
+ "hard_violation_episodes": 0
186
+ },
187
+ "post_reset": {
188
+ "checked": 25,
189
+ "hard_violation_episodes": 0
190
+ }
191
+ }
192
+ }
193
+ },
194
+ "robot_motion": {
195
+ "joint_acceleration_rms": 10.370723459949021,
196
+ "joint_effort_rms": 22.67024490029966,
197
+ "joint_jerk_rms": 1045.4285504760649,
198
+ "joint_velocity_rms": 0.34842602684067286
199
+ },
200
+ "grasp": {
201
+ "kinematic_grasp_stability_index": 0.6573163418859193,
202
+ "mean_kinematic_grasp_stability": 0.44478550041426507,
203
+ "per_object_best_kinematic_grasp_stability": {
204
+ "obj_008_pudding_box_2": 0.2960619052085125,
205
+ "obj_009_gelatin_box_3": 0.3151907525816141,
206
+ "obj_033_spatula_4": 0.5479037480942103,
207
+ "obj_189_brush_5": 0.2730420166176859
208
+ },
209
+ "grasp_gsi_diagnostics": {
210
+ "mean_anti_slip_score": 0.5485868847474353,
211
+ "mean_duration_score": 0.4790187946418155,
212
+ "mean_hold_score": 0.41315996554531975,
213
+ "mean_kinematic_stability_score": 0.029906534455991823,
214
+ "mean_legacy_gsi": 0.4439624097271413,
215
+ "mean_lift_score": 0.4308587194222873,
216
+ "mean_motion_stability_score": 0.3861783123673564
217
+ }
218
+ },
219
+ "tool": {
220
+ "tool_selection_accuracy": null,
221
+ "tool_switch_success_rate": null,
222
+ "tool_switch_total": 0,
223
+ "tool_switch_successful": 0
224
+ },
225
+ "runtime": {
226
+ "avg_steps": 2506.2,
227
+ "error_count": 0
228
+ }
229
+ }
230
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Full/summary.json ADDED
@@ -0,0 +1,240 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 203200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 50,
9
+ "successes": 11,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.22,
50
+ "success_rate_ci": {
51
+ "mean": 0.22,
52
+ "ci_low": 0.12,
53
+ "ci_high": 0.34
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.705
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 37.84393939393853
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.22,
64
+ "hard_violation_rate": 0.08,
65
+ "drop_rate": 0.08,
66
+ "high_speed_violation_rate": 0.04,
67
+ "safety_violation_step_rate": 0.023046502113732444
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.24,
74
+ "at_end_success_rate": 0.22,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.22
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.685,
81
+ "current_chain_depth_progress_score": 0.83,
82
+ "latched_chain_depth_progress_score": 0.86,
83
+ "avg_current_chain_depth": 1.66,
84
+ "avg_latched_chain_depth": 1.72
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7507161215096555,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2270.6363636363635,
91
+ "avg_policy_steps_to_success": 757.1818181818181,
92
+ "avg_policy_queries_to_success": 757.1818181818181
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 9.626105999470565e-05,
96
+ "violation_counts": {
97
+ "drop_obj_008_pudding_box_2": 2,
98
+ "drop_to_table_obj_008_pudding_box_2": 2,
99
+ "drop_to_table_obj_009_gelatin_box_3": 3,
100
+ "drop_to_table_obj_189_brush_5": 1,
101
+ "high_speed_obj_008_pudding_box_2": 2,
102
+ "high_speed_obj_130_tray_1": 2
103
+ },
104
+ "robot_constraint_violation_rate": 0.48,
105
+ "finger_joint_limit_saturation_rate": 0.08,
106
+ "joint_limit_active_step_rate": 0.0027113531898508756,
107
+ "finger_joint_limit_saturation_step_rate": 0.0006818491749624983,
108
+ "max_joint_limit_excess_rad": 0.24029040336608887,
109
+ "robot_constraint_diagnostics": {
110
+ "joint_limit_soft_counts": {
111
+ "left_pinky_MCP_FE": 7,
112
+ "left_thumb_CMC_AA": 50,
113
+ "left_thumb_CMC_FE": 38,
114
+ "left_thumb_IP": 7,
115
+ "left_thumb_MCP_FE": 297
116
+ },
117
+ "joint_limit_hard_counts": {
118
+ "left_pinky_CMC": 3,
119
+ "left_pinky_MCP_FE": 1,
120
+ "left_pinky_PIP": 1,
121
+ "left_thumb_CMC_AA": 17,
122
+ "left_thumb_CMC_FE": 27,
123
+ "left_thumb_IP": 5,
124
+ "left_thumb_MCP_FE": 14,
125
+ "multi_right_middle_MCP_FE": 1,
126
+ "multi_right_pinky_CMC": 2,
127
+ "multi_right_ring_MCP_FE": 2,
128
+ "multi_right_thumb_IP": 1,
129
+ "multi_right_thumb_MCP_FE": 4
130
+ },
131
+ "joint_limit_hard_episode_events": 57,
132
+ "finger_joint_limit_saturation_counts": {
133
+ "left_thumb_MCP_FE": 85
134
+ },
135
+ "joint_limit_max_excess_by_joint": {
136
+ "left_index_DIP": 3.342629861435853e-05,
137
+ "left_index_MCP_AA": 0.0001315772533416748,
138
+ "left_index_PIP": 0.0017800049390643835,
139
+ "left_middle_DIP": 0.00014326437667477876,
140
+ "left_middle_MCP_AA": 0.0005394816398620605,
141
+ "left_middle_PIP": 0.00488057266920805,
142
+ "left_pinky_CMC": 0.006102359853684902,
143
+ "left_pinky_DIP": 0.00021328224102035165,
144
+ "left_pinky_MCP_FE": 0.03453095257282257,
145
+ "left_pinky_PIP": 0.005239320453256369,
146
+ "left_ring_DIP": 0.0001839229662436992,
147
+ "left_ring_MCP_AA": 0.0005986690521240234,
148
+ "left_ring_PIP": 0.0030039905104786158,
149
+ "left_thumb_CMC_AA": 0.06530192494392395,
150
+ "left_thumb_CMC_FE": 0.056396484375,
151
+ "left_thumb_IP": 0.03799530118703842,
152
+ "left_thumb_MCP_FE": 0.24029040336608887,
153
+ "multi_A3": 1.5735626220703125e-05,
154
+ "multi_right_index_DIP": 1.52639574935165e-06,
155
+ "multi_right_index_MCP_AA": 0.00030609965324401855,
156
+ "multi_right_index_MCP_FE": 0.003085196018218994,
157
+ "multi_right_index_PIP": 0.004464675672352314,
158
+ "multi_right_middle_DIP": 7.662199347224785e-11,
159
+ "multi_right_middle_MCP_AA": 0.00040149688720703125,
160
+ "multi_right_middle_MCP_FE": 0.005505815148353577,
161
+ "multi_right_middle_PIP": 1.2028865403745925e-10,
162
+ "multi_right_pinky_CMC": 0.005548879038542509,
163
+ "multi_right_pinky_DIP": 0.0002515497035346925,
164
+ "multi_right_pinky_PIP": 0.0036197053268551826,
165
+ "multi_right_ring_DIP": 0.00024057368864305317,
166
+ "multi_right_ring_MCP_AA": 0.0009909272193908691,
167
+ "multi_right_ring_MCP_FE": 0.006244748830795288,
168
+ "multi_right_ring_PIP": 0.0011946060694754124,
169
+ "multi_right_thumb_CMC_FE": 0.00118255615234375,
170
+ "multi_right_thumb_IP": 0.006918040104210377,
171
+ "multi_right_thumb_MCP_FE": 0.005576014518737793
172
+ },
173
+ "coverage": {
174
+ "hard_limit_source": {
175
+ "available": 50,
176
+ "unavailable": 0,
177
+ "unknown": 0
178
+ },
179
+ "joint_role_manifest": {
180
+ "available": 50,
181
+ "unavailable": 0,
182
+ "unknown": 0
183
+ }
184
+ },
185
+ "hard_limit_sources": {
186
+ "runtime.joint_pos_limits": 50
187
+ },
188
+ "soft_limit_sources": {
189
+ "runtime.soft_joint_pos_limits": 50
190
+ },
191
+ "unknown_runtime_joint_names": [],
192
+ "baseline_checks": {
193
+ "post_home": {
194
+ "checked": 50,
195
+ "hard_violation_episodes": 0
196
+ },
197
+ "post_reset": {
198
+ "checked": 50,
199
+ "hard_violation_episodes": 0
200
+ }
201
+ }
202
+ }
203
+ },
204
+ "robot_motion": {
205
+ "joint_acceleration_rms": 11.040360762304445,
206
+ "joint_effort_rms": 24.564828231454772,
207
+ "joint_jerk_rms": 1115.113280222919,
208
+ "joint_velocity_rms": 0.3665976943610878
209
+ },
210
+ "grasp": {
211
+ "kinematic_grasp_stability_index": 0.6871524529508021,
212
+ "mean_kinematic_grasp_stability": 0.4307441148714327,
213
+ "per_object_best_kinematic_grasp_stability": {
214
+ "obj_008_pudding_box_2": 0.2974933243340069,
215
+ "obj_009_gelatin_box_3": 0.37966390367104447,
216
+ "obj_033_spatula_4": 0.5894557594085905,
217
+ "obj_189_brush_5": 0.3868253180726184
218
+ },
219
+ "grasp_gsi_diagnostics": {
220
+ "mean_anti_slip_score": 0.5421430631854829,
221
+ "mean_duration_score": 0.4858484799541366,
222
+ "mean_hold_score": 0.4162909099908463,
223
+ "mean_kinematic_stability_score": 0.02956694675729559,
224
+ "mean_legacy_gsi": 0.44528861569782063,
225
+ "mean_lift_score": 0.4354720982216404,
226
+ "mean_motion_stability_score": 0.3880000744069458
227
+ }
228
+ },
229
+ "tool": {
230
+ "tool_selection_accuracy": null,
231
+ "tool_switch_success_rate": null,
232
+ "tool_switch_total": 0,
233
+ "tool_switch_successful": 0
234
+ },
235
+ "runtime": {
236
+ "avg_steps": 2493.22,
237
+ "error_count": 0
238
+ }
239
+ }
240
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/native_shards/000/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/native_shards/000/summary.json ADDED
@@ -0,0 +1,227 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 26,
9
+ "successes": 11,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.4230769230769231,
50
+ "success_rate_ci": {
51
+ "mean": 0.4230769230769231,
52
+ "ci_low": 0.23076923076923078,
53
+ "ci_high": 0.6153846153846154
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.8269230769230769
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 38.95606060605968
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.4230769230769231,
64
+ "hard_violation_rate": 0.038461538461538464,
65
+ "drop_rate": 0.038461538461538464,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 1.5612558742252267e-05
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.46153846153846156,
74
+ "at_end_success_rate": 0.4230769230769231,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.4230769230769231
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.8173076923076923,
81
+ "current_chain_depth_progress_score": 0.9423076923076923,
82
+ "latched_chain_depth_progress_score": 0.9423076923076923,
83
+ "avg_current_chain_depth": 1.8846153846153846,
84
+ "avg_latched_chain_depth": 1.8846153846153846
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7296912523478297,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2337.3636363636365,
91
+ "avg_policy_steps_to_success": 779.5454545454545,
92
+ "avg_policy_queries_to_success": 779.5454545454545
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 1.5612558742252267e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_189_brush_5": 1
98
+ },
99
+ "robot_constraint_violation_rate": 0.34615384615384615,
100
+ "finger_joint_limit_saturation_rate": 0.038461538461538464,
101
+ "joint_limit_active_step_rate": 0.0016549312266787403,
102
+ "finger_joint_limit_saturation_step_rate": 0.0002810260573605408,
103
+ "max_joint_limit_excess_rad": 0.2964831590652466,
104
+ "robot_constraint_diagnostics": {
105
+ "joint_limit_soft_counts": {
106
+ "left_thumb_CMC_AA": 48,
107
+ "left_thumb_IP": 33,
108
+ "left_thumb_MCP_FE": 26
109
+ },
110
+ "joint_limit_hard_counts": {
111
+ "left_thumb_CMC_AA": 2,
112
+ "left_thumb_CMC_FE": 11,
113
+ "left_thumb_IP": 2,
114
+ "left_thumb_MCP_FE": 2,
115
+ "multi_right_ring_MCP_FE": 3,
116
+ "multi_right_thumb_IP": 1,
117
+ "multi_right_thumb_MCP_FE": 3
118
+ },
119
+ "joint_limit_hard_episode_events": 21,
120
+ "finger_joint_limit_saturation_counts": {
121
+ "left_thumb_IP": 18
122
+ },
123
+ "joint_limit_max_excess_by_joint": {
124
+ "left_index_DIP": 2.5854900741251186e-05,
125
+ "left_index_MCP_AA": 9.816884994506836e-05,
126
+ "left_index_PIP": 0.0012298865476623178,
127
+ "left_middle_DIP": 0.0002539681445341557,
128
+ "left_middle_MCP_AA": 0.0007359981536865234,
129
+ "left_middle_PIP": 0.0024956411216408014,
130
+ "left_pinky_CMC": 0.003886774880811572,
131
+ "left_pinky_DIP": 0.0002827503194566816,
132
+ "left_pinky_PIP": 0.0020380360074341297,
133
+ "left_ring_DIP": 0.0002530738420318812,
134
+ "left_ring_MCP_AA": 0.0008116960525512695,
135
+ "left_ring_PIP": 0.0026729372330009937,
136
+ "left_thumb_CMC_AA": 0.0694333016872406,
137
+ "left_thumb_CMC_FE": 0.009229421615600586,
138
+ "left_thumb_IP": 0.2964831590652466,
139
+ "left_thumb_MCP_FE": 0.14769291877746582,
140
+ "multi_A3": 3.5762786865234375e-06,
141
+ "multi_right_index_DIP": 1.6727758520573843e-06,
142
+ "multi_right_index_MCP_AA": 0.00015816092491149902,
143
+ "multi_right_index_MCP_FE": 0.001903831958770752,
144
+ "multi_right_index_PIP": 4.599662497639656e-05,
145
+ "multi_right_middle_DIP": 7.657938172478396e-11,
146
+ "multi_right_middle_MCP_AA": 0.00026035308837890625,
147
+ "multi_right_middle_MCP_FE": 0.004621222615242004,
148
+ "multi_right_middle_PIP": 1.0033789771268786e-10,
149
+ "multi_right_pinky_CMC": 0.0022513186559081078,
150
+ "multi_right_pinky_DIP": 5.5989366956055164e-05,
151
+ "multi_right_pinky_PIP": 0.0024412593338638544,
152
+ "multi_right_ring_DIP": 5.390810838434845e-05,
153
+ "multi_right_ring_MCP_AA": 0.00031045079231262207,
154
+ "multi_right_ring_MCP_FE": 0.007723212242126465,
155
+ "multi_right_ring_PIP": 0.00026808975962921977,
156
+ "multi_right_thumb_CMC_FE": 0.00015926361083984375,
157
+ "multi_right_thumb_IP": 0.005177497863769531,
158
+ "multi_right_thumb_MCP_FE": 0.00662684440612793
159
+ },
160
+ "coverage": {
161
+ "hard_limit_source": {
162
+ "available": 26,
163
+ "unavailable": 0,
164
+ "unknown": 0
165
+ },
166
+ "joint_role_manifest": {
167
+ "available": 26,
168
+ "unavailable": 0,
169
+ "unknown": 0
170
+ }
171
+ },
172
+ "hard_limit_sources": {
173
+ "runtime.joint_pos_limits": 26
174
+ },
175
+ "soft_limit_sources": {
176
+ "runtime.soft_joint_pos_limits": 26
177
+ },
178
+ "unknown_runtime_joint_names": [],
179
+ "baseline_checks": {
180
+ "post_home": {
181
+ "checked": 26,
182
+ "hard_violation_episodes": 0
183
+ },
184
+ "post_reset": {
185
+ "checked": 26,
186
+ "hard_violation_episodes": 0
187
+ }
188
+ }
189
+ }
190
+ },
191
+ "robot_motion": {
192
+ "joint_acceleration_rms": 11.641767794755419,
193
+ "joint_effort_rms": 25.089669182652727,
194
+ "joint_jerk_rms": 1177.97041089466,
195
+ "joint_velocity_rms": 0.3772248099506435
196
+ },
197
+ "grasp": {
198
+ "kinematic_grasp_stability_index": 0.6858413194230126,
199
+ "mean_kinematic_grasp_stability": 0.37640299987866477,
200
+ "per_object_best_kinematic_grasp_stability": {
201
+ "obj_008_pudding_box_2": 0.32140694447708706,
202
+ "obj_009_gelatin_box_3": 0.42388075114990054,
203
+ "obj_033_spatula_4": 0.567802413636381,
204
+ "obj_189_brush_5": 0.41247511524757335
205
+ },
206
+ "grasp_gsi_diagnostics": {
207
+ "mean_anti_slip_score": 0.5130337679626906,
208
+ "mean_duration_score": 0.5059150991730927,
209
+ "mean_hold_score": 0.44708705560149753,
210
+ "mean_kinematic_stability_score": 0.02342685310970755,
211
+ "mean_legacy_gsi": 0.4621576963605788,
212
+ "mean_lift_score": 0.4687475548081832,
213
+ "mean_motion_stability_score": 0.41885437764813477
214
+ }
215
+ },
216
+ "tool": {
217
+ "tool_selection_accuracy": null,
218
+ "tool_switch_success_rate": null,
219
+ "tool_switch_total": 0,
220
+ "tool_switch_successful": 0
221
+ },
222
+ "runtime": {
223
+ "avg_steps": 2463.5,
224
+ "error_count": 0
225
+ }
226
+ }
227
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/native_shards/001/summary.json ADDED
@@ -0,0 +1,225 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 26,
9
+ "successes": 8,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.3076923076923077,
50
+ "success_rate_ci": {
51
+ "mean": 0.3076923076923077,
52
+ "ci_low": 0.15384615384615385,
53
+ "ci_high": 0.5
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.7211538461538461
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 37.58333333333248
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.3076923076923077,
64
+ "hard_violation_rate": 0.0,
65
+ "drop_rate": 0.0,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 0.0
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.3076923076923077,
74
+ "at_end_success_rate": 0.3076923076923077,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.3076923076923077
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.7211538461538461,
81
+ "current_chain_depth_progress_score": 0.8269230769230769,
82
+ "latched_chain_depth_progress_score": 0.8269230769230769,
83
+ "avg_current_chain_depth": 1.6538461538461537,
84
+ "avg_latched_chain_depth": 1.6538461538461537
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7559436930112637,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2255,
91
+ "avg_policy_steps_to_success": 752,
92
+ "avg_policy_queries_to_success": 752
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.0,
96
+ "violation_counts": {},
97
+ "robot_constraint_violation_rate": 0.23076923076923078,
98
+ "finger_joint_limit_saturation_rate": 0.038461538461538464,
99
+ "joint_limit_active_step_rate": 0.000874344241818636,
100
+ "finger_joint_limit_saturation_step_rate": 3.122658006495129e-05,
101
+ "max_joint_limit_excess_rad": 0.1354694962501526,
102
+ "robot_constraint_diagnostics": {
103
+ "joint_limit_soft_counts": {
104
+ "left_thumb_CMC_AA": 19,
105
+ "left_thumb_CMC_FE": 2,
106
+ "left_thumb_IP": 1,
107
+ "left_thumb_MCP_FE": 55
108
+ },
109
+ "joint_limit_hard_counts": {
110
+ "left_thumb_CMC_AA": 3,
111
+ "left_thumb_CMC_FE": 4,
112
+ "left_thumb_IP": 2,
113
+ "left_thumb_MCP_FE": 1,
114
+ "multi_right_middle_MCP_FE": 1,
115
+ "multi_right_thumb_MCP_FE": 4
116
+ },
117
+ "joint_limit_hard_episode_events": 13,
118
+ "finger_joint_limit_saturation_counts": {
119
+ "left_thumb_MCP_FE": 2
120
+ },
121
+ "joint_limit_max_excess_by_joint": {
122
+ "left_index_DIP": 2.0355373635538854e-05,
123
+ "left_index_MCP_AA": 9.509921073913574e-05,
124
+ "left_index_PIP": 0.00010115083568962291,
125
+ "left_middle_DIP": 0.00010241794370813295,
126
+ "left_middle_MCP_AA": 0.0005443394184112549,
127
+ "left_middle_PIP": 0.0026255114935338497,
128
+ "left_pinky_CMC": 0.0023345130030065775,
129
+ "left_pinky_DIP": 0.00022600566444452852,
130
+ "left_pinky_PIP": 0.002117196097970009,
131
+ "left_ring_DIP": 0.00023119417892303318,
132
+ "left_ring_MCP_AA": 0.0005798935890197754,
133
+ "left_ring_PIP": 0.0027670767158269882,
134
+ "left_thumb_CMC_AA": 0.04382333159446716,
135
+ "left_thumb_CMC_FE": 0.011059880256652832,
136
+ "left_thumb_IP": 0.013164320029318333,
137
+ "left_thumb_MCP_FE": 0.1354694962501526,
138
+ "multi_A3": 0.00010704994201660156,
139
+ "multi_right_index_DIP": 1.8943131863125018e-06,
140
+ "multi_right_index_MCP_AA": 0.00030288100242614746,
141
+ "multi_right_index_MCP_FE": 0.002075761556625366,
142
+ "multi_right_index_PIP": 4.6002100134501234e-05,
143
+ "multi_right_middle_DIP": 1.0761639496204012e-10,
144
+ "multi_right_middle_MCP_AA": 0.0008371174335479736,
145
+ "multi_right_middle_MCP_FE": 0.0058893561363220215,
146
+ "multi_right_middle_PIP": 9.854752430760172e-11,
147
+ "multi_right_pinky_CMC": 0.002974559087306261,
148
+ "multi_right_pinky_DIP": 0.00018321051902603358,
149
+ "multi_right_pinky_PIP": 0.002482728334143758,
150
+ "multi_right_ring_DIP": 7.821453618817031e-05,
151
+ "multi_right_ring_MCP_AA": 0.0014412999153137207,
152
+ "multi_right_ring_MCP_FE": 0.003663763403892517,
153
+ "multi_right_ring_PIP": 0.000383607140975073,
154
+ "multi_right_thumb_CMC_FE": 0.00037682056427001953,
155
+ "multi_right_thumb_IP": 9.508449693385046e-06,
156
+ "multi_right_thumb_MCP_FE": 0.007659554481506348
157
+ },
158
+ "coverage": {
159
+ "hard_limit_source": {
160
+ "available": 26,
161
+ "unavailable": 0,
162
+ "unknown": 0
163
+ },
164
+ "joint_role_manifest": {
165
+ "available": 26,
166
+ "unavailable": 0,
167
+ "unknown": 0
168
+ }
169
+ },
170
+ "hard_limit_sources": {
171
+ "runtime.joint_pos_limits": 26
172
+ },
173
+ "soft_limit_sources": {
174
+ "runtime.soft_joint_pos_limits": 26
175
+ },
176
+ "unknown_runtime_joint_names": [],
177
+ "baseline_checks": {
178
+ "post_home": {
179
+ "checked": 26,
180
+ "hard_violation_episodes": 0
181
+ },
182
+ "post_reset": {
183
+ "checked": 26,
184
+ "hard_violation_episodes": 0
185
+ }
186
+ }
187
+ }
188
+ },
189
+ "robot_motion": {
190
+ "joint_acceleration_rms": 10.550250557524324,
191
+ "joint_effort_rms": 24.27680371254765,
192
+ "joint_jerk_rms": 1067.5140615339633,
193
+ "joint_velocity_rms": 0.38743886865103017
194
+ },
195
+ "grasp": {
196
+ "kinematic_grasp_stability_index": 0.6625200148069456,
197
+ "mean_kinematic_grasp_stability": 0.3831520687970661,
198
+ "per_object_best_kinematic_grasp_stability": {
199
+ "obj_008_pudding_box_2": 0.28320735370930916,
200
+ "obj_009_gelatin_box_3": 0.3488996105500065,
201
+ "obj_033_spatula_4": 0.5745682775464469,
202
+ "obj_189_brush_5": 0.2873969342619228
203
+ },
204
+ "grasp_gsi_diagnostics": {
205
+ "mean_anti_slip_score": 0.5443230475198425,
206
+ "mean_duration_score": 0.46472665966727095,
207
+ "mean_hold_score": 0.4189878810492634,
208
+ "mean_kinematic_stability_score": 0.01810367149507761,
209
+ "mean_legacy_gsi": 0.4405549046097361,
210
+ "mean_lift_score": 0.4350976593247544,
211
+ "mean_motion_stability_score": 0.38596718466309754
212
+ }
213
+ },
214
+ "tool": {
215
+ "tool_selection_accuracy": null,
216
+ "tool_switch_success_rate": null,
217
+ "tool_switch_total": 0,
218
+ "tool_switch_successful": 0
219
+ },
220
+ "runtime": {
221
+ "avg_steps": 2463.3846153846152,
222
+ "error_count": 0
223
+ }
224
+ }
225
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/Inv/summary.json ADDED
@@ -0,0 +1,230 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 50,
9
+ "successes": 19,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.38,
50
+ "success_rate_ci": {
51
+ "mean": 0.38,
52
+ "ci_low": 0.24,
53
+ "ci_high": 0.52
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.78
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 38.3780701754377
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.38,
64
+ "hard_violation_rate": 0.02,
65
+ "drop_rate": 0.02,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 8.130940668525942e-06
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.4,
74
+ "at_end_success_rate": 0.38,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.38
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.775,
81
+ "current_chain_depth_progress_score": 0.89,
82
+ "latched_chain_depth_progress_score": 0.89,
83
+ "avg_current_chain_depth": 1.78,
84
+ "avg_latched_chain_depth": 1.78
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7407449115745388,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2302.684210526316,
91
+ "avg_policy_steps_to_success": 767.9473684210526,
92
+ "avg_policy_queries_to_success": 767.9473684210526
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 8.130940668525942e-06,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_189_brush_5": 1
98
+ },
99
+ "robot_constraint_violation_rate": 0.3,
100
+ "finger_joint_limit_saturation_rate": 0.04,
101
+ "joint_limit_active_step_rate": 0.0013172123883012025,
102
+ "finger_joint_limit_saturation_step_rate": 0.00016261881337051884,
103
+ "max_joint_limit_excess_rad": 0.2964831590652466,
104
+ "robot_constraint_diagnostics": {
105
+ "joint_limit_soft_counts": {
106
+ "left_thumb_CMC_AA": 67,
107
+ "left_thumb_CMC_FE": 2,
108
+ "left_thumb_IP": 34,
109
+ "left_thumb_MCP_FE": 81
110
+ },
111
+ "joint_limit_hard_counts": {
112
+ "left_thumb_CMC_AA": 5,
113
+ "left_thumb_CMC_FE": 15,
114
+ "left_thumb_IP": 4,
115
+ "left_thumb_MCP_FE": 3,
116
+ "multi_right_middle_MCP_FE": 1,
117
+ "multi_right_ring_MCP_FE": 3,
118
+ "multi_right_thumb_IP": 1,
119
+ "multi_right_thumb_MCP_FE": 7
120
+ },
121
+ "joint_limit_hard_episode_events": 34,
122
+ "finger_joint_limit_saturation_counts": {
123
+ "left_thumb_IP": 18,
124
+ "left_thumb_MCP_FE": 2
125
+ },
126
+ "joint_limit_max_excess_by_joint": {
127
+ "left_index_DIP": 2.5854900741251186e-05,
128
+ "left_index_MCP_AA": 9.816884994506836e-05,
129
+ "left_index_PIP": 0.0012298865476623178,
130
+ "left_middle_DIP": 0.0002539681445341557,
131
+ "left_middle_MCP_AA": 0.0007359981536865234,
132
+ "left_middle_PIP": 0.0026255114935338497,
133
+ "left_pinky_CMC": 0.003886774880811572,
134
+ "left_pinky_DIP": 0.0002827503194566816,
135
+ "left_pinky_PIP": 0.002117196097970009,
136
+ "left_ring_DIP": 0.0002530738420318812,
137
+ "left_ring_MCP_AA": 0.0008116960525512695,
138
+ "left_ring_PIP": 0.0027670767158269882,
139
+ "left_thumb_CMC_AA": 0.0694333016872406,
140
+ "left_thumb_CMC_FE": 0.011059880256652832,
141
+ "left_thumb_IP": 0.2964831590652466,
142
+ "left_thumb_MCP_FE": 0.14769291877746582,
143
+ "multi_A3": 0.00010704994201660156,
144
+ "multi_right_index_DIP": 1.8943131863125018e-06,
145
+ "multi_right_index_MCP_AA": 0.00030288100242614746,
146
+ "multi_right_index_MCP_FE": 0.002075761556625366,
147
+ "multi_right_index_PIP": 4.6002100134501234e-05,
148
+ "multi_right_middle_DIP": 1.0761639496204012e-10,
149
+ "multi_right_middle_MCP_AA": 0.0008371174335479736,
150
+ "multi_right_middle_MCP_FE": 0.0058893561363220215,
151
+ "multi_right_middle_PIP": 1.0033789771268786e-10,
152
+ "multi_right_pinky_CMC": 0.002974559087306261,
153
+ "multi_right_pinky_DIP": 0.00018321051902603358,
154
+ "multi_right_pinky_PIP": 0.002482728334143758,
155
+ "multi_right_ring_DIP": 7.821453618817031e-05,
156
+ "multi_right_ring_MCP_AA": 0.0014412999153137207,
157
+ "multi_right_ring_MCP_FE": 0.007723212242126465,
158
+ "multi_right_ring_PIP": 0.000383607140975073,
159
+ "multi_right_thumb_CMC_FE": 0.00037682056427001953,
160
+ "multi_right_thumb_IP": 0.005177497863769531,
161
+ "multi_right_thumb_MCP_FE": 0.007659554481506348
162
+ },
163
+ "coverage": {
164
+ "hard_limit_source": {
165
+ "available": 50,
166
+ "unavailable": 0,
167
+ "unknown": 0
168
+ },
169
+ "joint_role_manifest": {
170
+ "available": 50,
171
+ "unavailable": 0,
172
+ "unknown": 0
173
+ }
174
+ },
175
+ "hard_limit_sources": {
176
+ "runtime.joint_pos_limits": 50
177
+ },
178
+ "soft_limit_sources": {
179
+ "runtime.soft_joint_pos_limits": 50
180
+ },
181
+ "unknown_runtime_joint_names": [],
182
+ "baseline_checks": {
183
+ "post_home": {
184
+ "checked": 50,
185
+ "hard_violation_episodes": 0
186
+ },
187
+ "post_reset": {
188
+ "checked": 50,
189
+ "hard_violation_episodes": 0
190
+ }
191
+ }
192
+ }
193
+ },
194
+ "robot_motion": {
195
+ "joint_acceleration_rms": 11.176679190280542,
196
+ "joint_effort_rms": 24.82223570858676,
197
+ "joint_jerk_rms": 1130.6009068638746,
198
+ "joint_velocity_rms": 0.38620972698055034
199
+ },
200
+ "grasp": {
201
+ "kinematic_grasp_stability_index": 0.6750571658010093,
202
+ "mean_kinematic_grasp_stability": 0.37224378429517097,
203
+ "per_object_best_kinematic_grasp_stability": {
204
+ "obj_008_pudding_box_2": 0.3010144219608551,
205
+ "obj_009_gelatin_box_3": 0.391765716725148,
206
+ "obj_033_spatula_4": 0.5803263511517472,
207
+ "obj_189_brush_5": 0.3439609023773695
208
+ },
209
+ "grasp_gsi_diagnostics": {
210
+ "mean_anti_slip_score": 0.5282338643430403,
211
+ "mean_duration_score": 0.48464959879104236,
212
+ "mean_hold_score": 0.431496780047441,
213
+ "mean_kinematic_stability_score": 0.019809914988323647,
214
+ "mean_legacy_gsi": 0.4502591983227179,
215
+ "mean_lift_score": 0.450854362679331,
216
+ "mean_motion_stability_score": 0.40090075113824586
217
+ }
218
+ },
219
+ "tool": {
220
+ "tool_selection_accuracy": null,
221
+ "tool_switch_success_rate": null,
222
+ "tool_switch_total": 0,
223
+ "tool_switch_successful": 0
224
+ },
225
+ "runtime": {
226
+ "avg_steps": 2459.74,
227
+ "error_count": 0
228
+ }
229
+ }
230
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/native_shards/000/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/native_shards/000/summary.json ADDED
@@ -0,0 +1,228 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 9,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.36,
50
+ "success_rate_ci": {
51
+ "mean": 0.36,
52
+ "ci_low": 0.16,
53
+ "ci_high": 0.56
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.8
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 38.15925925925838
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.36,
64
+ "hard_violation_rate": 0.0,
65
+ "drop_rate": 0.0,
66
+ "high_speed_violation_rate": 0.0,
67
+ "safety_violation_step_rate": 0.0
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.36,
74
+ "at_end_success_rate": 0.36,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.36
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.8,
81
+ "current_chain_depth_progress_score": 0.94,
82
+ "latched_chain_depth_progress_score": 0.94,
83
+ "avg_current_chain_depth": 1.88,
84
+ "avg_latched_chain_depth": 1.88
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7449428655506063,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2289.5555555555557,
91
+ "avg_policy_steps_to_success": 763.6666666666666,
92
+ "avg_policy_queries_to_success": 763.6666666666666
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.0,
96
+ "violation_counts": {},
97
+ "robot_constraint_violation_rate": 0.32,
98
+ "finger_joint_limit_saturation_rate": 0.0,
99
+ "joint_limit_active_step_rate": 0.0018698578908002991,
100
+ "finger_joint_limit_saturation_step_rate": 0.0,
101
+ "max_joint_limit_excess_rad": 0.1085924506187439,
102
+ "robot_constraint_diagnostics": {
103
+ "joint_limit_soft_counts": {
104
+ "left_pinky_MCP_FE": 12,
105
+ "left_thumb_CMC_AA": 47,
106
+ "left_thumb_CMC_FE": 13,
107
+ "left_thumb_IP": 6,
108
+ "left_thumb_MCP_FE": 39
109
+ },
110
+ "joint_limit_hard_counts": {
111
+ "left_pinky_CMC": 3,
112
+ "left_pinky_MCP_FE": 1,
113
+ "left_thumb_CMC_AA": 9,
114
+ "left_thumb_CMC_FE": 5,
115
+ "left_thumb_IP": 3,
116
+ "left_thumb_MCP_FE": 3,
117
+ "multi_right_index_PIP": 1,
118
+ "multi_right_ring_MCP_FE": 1
119
+ },
120
+ "joint_limit_hard_episode_events": 22,
121
+ "finger_joint_limit_saturation_counts": {},
122
+ "joint_limit_max_excess_by_joint": {
123
+ "left_index_DIP": 2.6806912501342595e-05,
124
+ "left_index_MCP_AA": 0.0001448988914489746,
125
+ "left_index_PIP": 0.0005326985265128314,
126
+ "left_middle_DIP": 0.00010738193668657914,
127
+ "left_middle_MCP_AA": 0.0005458295345306396,
128
+ "left_middle_PIP": 0.002447026316076517,
129
+ "left_pinky_CMC": 0.006223288830369711,
130
+ "left_pinky_DIP": 0.00022771456860937178,
131
+ "left_pinky_MCP_FE": 0.03192277252674103,
132
+ "left_pinky_PIP": 0.004318539518862963,
133
+ "left_ring_DIP": 0.00022146120318211615,
134
+ "left_ring_MCP_AA": 0.0005952417850494385,
135
+ "left_ring_PIP": 0.0025520522613078356,
136
+ "left_thumb_CMC_AA": 0.1085924506187439,
137
+ "left_thumb_CMC_FE": 0.0460895299911499,
138
+ "left_thumb_IP": 0.05418488755822182,
139
+ "left_thumb_MCP_FE": 0.09735137224197388,
140
+ "multi_A3": 4.0531158447265625e-06,
141
+ "multi_A6": 4.76837158203125e-06,
142
+ "multi_right_index_DIP": 1.6824434396767174e-06,
143
+ "multi_right_index_MCP_AA": 0.00023305416107177734,
144
+ "multi_right_index_MCP_FE": 0.00010530650615692139,
145
+ "multi_right_index_PIP": 0.005310348700731993,
146
+ "multi_right_middle_DIP": 7.657401102090233e-11,
147
+ "multi_right_middle_MCP_AA": 0.00031834840774536133,
148
+ "multi_right_middle_MCP_FE": 0.002497851848602295,
149
+ "multi_right_middle_PIP": 1.039135930080981e-10,
150
+ "multi_right_pinky_CMC": 0.0013424770440906286,
151
+ "multi_right_pinky_DIP": 6.447032501455396e-05,
152
+ "multi_right_pinky_PIP": 0.002023750450462103,
153
+ "multi_right_ring_DIP": 5.641915777232498e-05,
154
+ "multi_right_ring_MCP_AA": 0.0005308389663696289,
155
+ "multi_right_ring_MCP_FE": 0.006735458970069885,
156
+ "multi_right_ring_PIP": 0.0002812324673868716,
157
+ "multi_right_thumb_CMC_FE": 0.0003038644790649414,
158
+ "multi_right_thumb_IP": 0.0017924016574397683,
159
+ "multi_right_thumb_MCP_FE": 0.00470966100692749
160
+ },
161
+ "coverage": {
162
+ "hard_limit_source": {
163
+ "available": 25,
164
+ "unavailable": 0,
165
+ "unknown": 0
166
+ },
167
+ "joint_role_manifest": {
168
+ "available": 25,
169
+ "unavailable": 0,
170
+ "unknown": 0
171
+ }
172
+ },
173
+ "hard_limit_sources": {
174
+ "runtime.joint_pos_limits": 25
175
+ },
176
+ "soft_limit_sources": {
177
+ "runtime.soft_joint_pos_limits": 25
178
+ },
179
+ "unknown_runtime_joint_names": [],
180
+ "baseline_checks": {
181
+ "post_home": {
182
+ "checked": 25,
183
+ "hard_violation_episodes": 0
184
+ },
185
+ "post_reset": {
186
+ "checked": 25,
187
+ "hard_violation_episodes": 0
188
+ }
189
+ }
190
+ }
191
+ },
192
+ "robot_motion": {
193
+ "joint_acceleration_rms": 11.61807060009224,
194
+ "joint_effort_rms": 25.08402132566482,
195
+ "joint_jerk_rms": 1175.5482843173716,
196
+ "joint_velocity_rms": 0.376413447839903
197
+ },
198
+ "grasp": {
199
+ "kinematic_grasp_stability_index": 0.7150835261526663,
200
+ "mean_kinematic_grasp_stability": 0.4272437689715131,
201
+ "per_object_best_kinematic_grasp_stability": {
202
+ "obj_008_pudding_box_2": 0.3038320876686362,
203
+ "obj_009_gelatin_box_3": 0.3469398721487977,
204
+ "obj_033_spatula_4": 0.6060175282018089,
205
+ "obj_189_brush_5": 0.35749551075900865
206
+ },
207
+ "grasp_gsi_diagnostics": {
208
+ "mean_anti_slip_score": 0.5294419547706882,
209
+ "mean_duration_score": 0.5029950826960542,
210
+ "mean_hold_score": 0.4520846378152176,
211
+ "mean_kinematic_stability_score": 0.03503681384363399,
212
+ "mean_legacy_gsi": 0.4668992984931837,
213
+ "mean_lift_score": 0.4710089959727912,
214
+ "mean_motion_stability_score": 0.42239528738184146
215
+ }
216
+ },
217
+ "tool": {
218
+ "tool_selection_accuracy": null,
219
+ "tool_switch_success_rate": null,
220
+ "tool_switch_total": 0,
221
+ "tool_switch_successful": 0
222
+ },
223
+ "runtime": {
224
+ "avg_steps": 2460.08,
225
+ "error_count": 0
226
+ }
227
+ }
228
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/native_shards/001/summary.json ADDED
@@ -0,0 +1,234 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 4,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.16,
50
+ "success_rate_ci": {
51
+ "mean": 0.16,
52
+ "ci_low": 0.04,
53
+ "ci_high": 0.32
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.65
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 38.21249999999911
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.16,
64
+ "hard_violation_rate": 0.12,
65
+ "drop_rate": 0.12,
66
+ "high_speed_violation_rate": 0.04,
67
+ "safety_violation_step_rate": 0.022339968494916224
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.16,
74
+ "at_end_success_rate": 0.16,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.16
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.63,
81
+ "current_chain_depth_progress_score": 0.78,
82
+ "latched_chain_depth_progress_score": 0.8,
83
+ "avg_current_chain_depth": 1.56,
84
+ "avg_latched_chain_depth": 1.6
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7452524728163944,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2292.75,
91
+ "avg_policy_steps_to_success": 764.75,
92
+ "avg_policy_queries_to_success": 764.75
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.00011138160930513788,
96
+ "violation_counts": {
97
+ "drop_obj_008_pudding_box_2": 1,
98
+ "drop_to_table_obj_008_pudding_box_2": 1,
99
+ "drop_to_table_obj_009_gelatin_box_3": 1,
100
+ "drop_to_table_obj_033_spatula_4": 1,
101
+ "drop_to_table_obj_189_brush_5": 1,
102
+ "high_speed_obj_008_pudding_box_2": 1,
103
+ "high_speed_obj_130_tray_1": 1
104
+ },
105
+ "robot_constraint_violation_rate": 0.24,
106
+ "finger_joint_limit_saturation_rate": 0.04,
107
+ "joint_limit_active_step_rate": 0.0028004518911006093,
108
+ "finger_joint_limit_saturation_step_rate": 4.773497541648766e-05,
109
+ "max_joint_limit_excess_rad": 0.15280389785766602,
110
+ "robot_constraint_diagnostics": {
111
+ "joint_limit_soft_counts": {
112
+ "left_thumb_CMC_AA": 30,
113
+ "left_thumb_CMC_FE": 36,
114
+ "left_thumb_IP": 6,
115
+ "left_thumb_MCP_FE": 109,
116
+ "multi_right_ring_MCP_FE": 22
117
+ },
118
+ "joint_limit_hard_counts": {
119
+ "left_thumb_CMC_AA": 7,
120
+ "left_thumb_CMC_FE": 4,
121
+ "left_thumb_IP": 1,
122
+ "left_thumb_MCP_FE": 2,
123
+ "multi_right_ring_MCP_FE": 1,
124
+ "multi_right_thumb_MCP_FE": 1
125
+ },
126
+ "joint_limit_hard_episode_events": 9,
127
+ "finger_joint_limit_saturation_counts": {
128
+ "left_thumb_MCP_FE": 3
129
+ },
130
+ "joint_limit_max_excess_by_joint": {
131
+ "left_index_DIP": 2.5519006157992408e-05,
132
+ "left_index_MCP_AA": 0.00012895464897155762,
133
+ "left_index_PIP": 0.00011405559052946046,
134
+ "left_middle_DIP": 0.00023273831175174564,
135
+ "left_middle_MCP_AA": 0.0008727312088012695,
136
+ "left_middle_PIP": 0.0025665750727057457,
137
+ "left_pinky_CMC": 0.0032248259522020817,
138
+ "left_pinky_DIP": 0.0002365727414144203,
139
+ "left_pinky_PIP": 0.0028719084803014994,
140
+ "left_ring_DIP": 0.0001914222229970619,
141
+ "left_ring_MCP_AA": 0.0006642043590545654,
142
+ "left_ring_PIP": 0.0025066195521503687,
143
+ "left_thumb_CMC_AA": 0.0490325391292572,
144
+ "left_thumb_CMC_FE": 0.11739826202392578,
145
+ "left_thumb_IP": 0.0580114983022213,
146
+ "left_thumb_MCP_FE": 0.15280389785766602,
147
+ "multi_A3": 2.86102294921875e-06,
148
+ "multi_right_index_DIP": 4.28590965384501e-06,
149
+ "multi_right_index_MCP_AA": 0.00026422739028930664,
150
+ "multi_right_index_MCP_FE": 8.675456047058105e-05,
151
+ "multi_right_index_PIP": 4.624227585736662e-05,
152
+ "multi_right_middle_DIP": 7.662253470597236e-11,
153
+ "multi_right_middle_MCP_AA": 0.000336378812789917,
154
+ "multi_right_middle_MCP_FE": 0.0039657652378082275,
155
+ "multi_right_middle_PIP": 9.456432858989672e-11,
156
+ "multi_right_pinky_CMC": 0.0023716522846370935,
157
+ "multi_right_pinky_DIP": 0.0002255796134704724,
158
+ "multi_right_pinky_PIP": 0.002576685044914484,
159
+ "multi_right_ring_DIP": 0.00024087121710181236,
160
+ "multi_right_ring_MCP_AA": 0.000679314136505127,
161
+ "multi_right_ring_MCP_FE": 0.09894850850105286,
162
+ "multi_right_ring_PIP": 0.0011888236040249467,
163
+ "multi_right_thumb_CMC_FE": 9.09566879272461e-05,
164
+ "multi_right_thumb_IP": 1.855799200711772e-05,
165
+ "multi_right_thumb_MCP_FE": 0.0056116580963134766
166
+ },
167
+ "coverage": {
168
+ "hard_limit_source": {
169
+ "available": 25,
170
+ "unavailable": 0,
171
+ "unknown": 0
172
+ },
173
+ "joint_role_manifest": {
174
+ "available": 25,
175
+ "unavailable": 0,
176
+ "unknown": 0
177
+ }
178
+ },
179
+ "hard_limit_sources": {
180
+ "runtime.joint_pos_limits": 25
181
+ },
182
+ "soft_limit_sources": {
183
+ "runtime.soft_joint_pos_limits": 25
184
+ },
185
+ "unknown_runtime_joint_names": [],
186
+ "baseline_checks": {
187
+ "post_home": {
188
+ "checked": 25,
189
+ "hard_violation_episodes": 0
190
+ },
191
+ "post_reset": {
192
+ "checked": 25,
193
+ "hard_violation_episodes": 0
194
+ }
195
+ }
196
+ }
197
+ },
198
+ "robot_motion": {
199
+ "joint_acceleration_rms": 10.55690455175215,
200
+ "joint_effort_rms": 23.214092702594876,
201
+ "joint_jerk_rms": 1067.626287839437,
202
+ "joint_velocity_rms": 0.3481468811676283
203
+ },
204
+ "grasp": {
205
+ "kinematic_grasp_stability_index": 0.6525180141471666,
206
+ "mean_kinematic_grasp_stability": 0.4087552424435135,
207
+ "per_object_best_kinematic_grasp_stability": {
208
+ "obj_008_pudding_box_2": 0.2909439179201037,
209
+ "obj_009_gelatin_box_3": 0.3544227570808365,
210
+ "obj_033_spatula_4": 0.5792121972785838,
211
+ "obj_189_brush_5": 0.24903882055778992
212
+ },
213
+ "grasp_gsi_diagnostics": {
214
+ "mean_anti_slip_score": 0.5770118461511101,
215
+ "mean_duration_score": 0.4618306066572913,
216
+ "mean_hold_score": 0.3952495936979196,
217
+ "mean_kinematic_stability_score": 0.0401991319271516,
218
+ "mean_legacy_gsi": 0.4341518718645797,
219
+ "mean_lift_score": 0.4119837339362711,
220
+ "mean_motion_stability_score": 0.3667403865158515
221
+ }
222
+ },
223
+ "tool": {
224
+ "tool_selection_accuracy": null,
225
+ "tool_switch_success_rate": null,
226
+ "tool_switch_total": 0,
227
+ "tool_switch_successful": 0
228
+ },
229
+ "runtime": {
230
+ "avg_steps": 2513.88,
231
+ "error_count": 0
232
+ }
233
+ }
234
+ }
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/32_baking_tray_prep_with_tools/None/summary.json ADDED
@@ -0,0 +1,240 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA tray_hybrid-c636840313",
5
+ "base_seed": 103200000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 50,
9
+ "successes": 13,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 2556,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.26,
50
+ "success_rate_ci": {
51
+ "mean": 0.26,
52
+ "ci_low": 0.14,
53
+ "ci_high": 0.38
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.725
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": 38.17564102564014
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.26,
64
+ "hard_violation_rate": 0.06,
65
+ "drop_rate": 0.06,
66
+ "high_speed_violation_rate": 0.02,
67
+ "safety_violation_step_rate": 0.011290802499416963
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.26,
74
+ "at_end_success_rate": 0.26,
75
+ "task_family_success_rate": {
76
+ "rigid_pick_place": 0.26
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.715,
81
+ "current_chain_depth_progress_score": 0.86,
82
+ "latched_chain_depth_progress_score": 0.87,
83
+ "avg_current_chain_depth": 1.72,
84
+ "avg_latched_chain_depth": 1.74
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": 0.7450381293246949,
88
+ "expert_time_s": 28.383333333333333,
89
+ "expert_time_step": 1703,
90
+ "avg_steps_to_success": 2290.5384615384614,
91
+ "avg_policy_steps_to_success": 764,
92
+ "avg_policy_queries_to_success": 764
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 5.629317485464298e-05,
96
+ "violation_counts": {
97
+ "drop_obj_008_pudding_box_2": 1,
98
+ "drop_to_table_obj_008_pudding_box_2": 1,
99
+ "drop_to_table_obj_009_gelatin_box_3": 1,
100
+ "drop_to_table_obj_033_spatula_4": 1,
101
+ "drop_to_table_obj_189_brush_5": 1,
102
+ "high_speed_obj_008_pudding_box_2": 1,
103
+ "high_speed_obj_130_tray_1": 1
104
+ },
105
+ "robot_constraint_violation_rate": 0.28,
106
+ "finger_joint_limit_saturation_rate": 0.02,
107
+ "joint_limit_active_step_rate": 0.0023401876975287297,
108
+ "finger_joint_limit_saturation_step_rate": 2.4125646366275563e-05,
109
+ "max_joint_limit_excess_rad": 0.15280389785766602,
110
+ "robot_constraint_diagnostics": {
111
+ "joint_limit_soft_counts": {
112
+ "left_pinky_MCP_FE": 12,
113
+ "left_thumb_CMC_AA": 77,
114
+ "left_thumb_CMC_FE": 49,
115
+ "left_thumb_IP": 12,
116
+ "left_thumb_MCP_FE": 148,
117
+ "multi_right_ring_MCP_FE": 22
118
+ },
119
+ "joint_limit_hard_counts": {
120
+ "left_pinky_CMC": 3,
121
+ "left_pinky_MCP_FE": 1,
122
+ "left_thumb_CMC_AA": 16,
123
+ "left_thumb_CMC_FE": 9,
124
+ "left_thumb_IP": 4,
125
+ "left_thumb_MCP_FE": 5,
126
+ "multi_right_index_PIP": 1,
127
+ "multi_right_ring_MCP_FE": 2,
128
+ "multi_right_thumb_MCP_FE": 1
129
+ },
130
+ "joint_limit_hard_episode_events": 31,
131
+ "finger_joint_limit_saturation_counts": {
132
+ "left_thumb_MCP_FE": 3
133
+ },
134
+ "joint_limit_max_excess_by_joint": {
135
+ "left_index_DIP": 2.6806912501342595e-05,
136
+ "left_index_MCP_AA": 0.0001448988914489746,
137
+ "left_index_PIP": 0.0005326985265128314,
138
+ "left_middle_DIP": 0.00023273831175174564,
139
+ "left_middle_MCP_AA": 0.0008727312088012695,
140
+ "left_middle_PIP": 0.0025665750727057457,
141
+ "left_pinky_CMC": 0.006223288830369711,
142
+ "left_pinky_DIP": 0.0002365727414144203,
143
+ "left_pinky_MCP_FE": 0.03192277252674103,
144
+ "left_pinky_PIP": 0.004318539518862963,
145
+ "left_ring_DIP": 0.00022146120318211615,
146
+ "left_ring_MCP_AA": 0.0006642043590545654,
147
+ "left_ring_PIP": 0.0025520522613078356,
148
+ "left_thumb_CMC_AA": 0.1085924506187439,
149
+ "left_thumb_CMC_FE": 0.11739826202392578,
150
+ "left_thumb_IP": 0.0580114983022213,
151
+ "left_thumb_MCP_FE": 0.15280389785766602,
152
+ "multi_A3": 4.0531158447265625e-06,
153
+ "multi_A6": 4.76837158203125e-06,
154
+ "multi_right_index_DIP": 4.28590965384501e-06,
155
+ "multi_right_index_MCP_AA": 0.00026422739028930664,
156
+ "multi_right_index_MCP_FE": 0.00010530650615692139,
157
+ "multi_right_index_PIP": 0.005310348700731993,
158
+ "multi_right_middle_DIP": 7.662253470597236e-11,
159
+ "multi_right_middle_MCP_AA": 0.000336378812789917,
160
+ "multi_right_middle_MCP_FE": 0.0039657652378082275,
161
+ "multi_right_middle_PIP": 1.039135930080981e-10,
162
+ "multi_right_pinky_CMC": 0.0023716522846370935,
163
+ "multi_right_pinky_DIP": 0.0002255796134704724,
164
+ "multi_right_pinky_PIP": 0.002576685044914484,
165
+ "multi_right_ring_DIP": 0.00024087121710181236,
166
+ "multi_right_ring_MCP_AA": 0.000679314136505127,
167
+ "multi_right_ring_MCP_FE": 0.09894850850105286,
168
+ "multi_right_ring_PIP": 0.0011888236040249467,
169
+ "multi_right_thumb_CMC_FE": 0.0003038644790649414,
170
+ "multi_right_thumb_IP": 0.0017924016574397683,
171
+ "multi_right_thumb_MCP_FE": 0.0056116580963134766
172
+ },
173
+ "coverage": {
174
+ "hard_limit_source": {
175
+ "available": 50,
176
+ "unavailable": 0,
177
+ "unknown": 0
178
+ },
179
+ "joint_role_manifest": {
180
+ "available": 50,
181
+ "unavailable": 0,
182
+ "unknown": 0
183
+ }
184
+ },
185
+ "hard_limit_sources": {
186
+ "runtime.joint_pos_limits": 50
187
+ },
188
+ "soft_limit_sources": {
189
+ "runtime.soft_joint_pos_limits": 50
190
+ },
191
+ "unknown_runtime_joint_names": [],
192
+ "baseline_checks": {
193
+ "post_home": {
194
+ "checked": 50,
195
+ "hard_violation_episodes": 0
196
+ },
197
+ "post_reset": {
198
+ "checked": 50,
199
+ "hard_violation_episodes": 0
200
+ }
201
+ }
202
+ }
203
+ },
204
+ "robot_motion": {
205
+ "joint_acceleration_rms": 11.087487575922195,
206
+ "joint_effort_rms": 24.14905701412985,
207
+ "joint_jerk_rms": 1121.5872860784043,
208
+ "joint_velocity_rms": 0.36228016450376566
209
+ },
210
+ "grasp": {
211
+ "kinematic_grasp_stability_index": 0.6838007701499165,
212
+ "mean_kinematic_grasp_stability": 0.41818816414147253,
213
+ "per_object_best_kinematic_grasp_stability": {
214
+ "obj_008_pudding_box_2": 0.29738800279436994,
215
+ "obj_009_gelatin_box_3": 0.3506813146148171,
216
+ "obj_033_spatula_4": 0.5926148627401963,
217
+ "obj_189_brush_5": 0.3032671656583993
218
+ },
219
+ "grasp_gsi_diagnostics": {
220
+ "mean_anti_slip_score": 0.5532269004608992,
221
+ "mean_duration_score": 0.4824128446766728,
222
+ "mean_hold_score": 0.4236671157565686,
223
+ "mean_kinematic_stability_score": 0.03761797288539279,
224
+ "mean_legacy_gsi": 0.4505255851788817,
225
+ "mean_lift_score": 0.44149636495453115,
226
+ "mean_motion_stability_score": 0.39456783694884645
227
+ }
228
+ },
229
+ "tool": {
230
+ "tool_selection_accuracy": null,
231
+ "tool_switch_success_rate": null,
232
+ "tool_switch_total": 0,
233
+ "tool_switch_successful": 0
234
+ },
235
+ "runtime": {
236
+ "avg_steps": 2486.98,
237
+ "error_count": 0
238
+ }
239
+ }
240
+ }
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/native_shards/000/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/native_shards/000/summary.json ADDED
@@ -0,0 +1,247 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA retensioned_service-3cc32133ca",
5
+ "base_seed": 103400000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 0,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3240,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.0,
50
+ "success_rate_ci": {
51
+ "mean": 0.0,
52
+ "ci_low": 0.0,
53
+ "ci_high": 0.0
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.448
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": null
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.0,
64
+ "hard_violation_rate": 0.32,
65
+ "drop_rate": 0.32,
66
+ "high_speed_violation_rate": 0.16,
67
+ "safety_violation_step_rate": 0.00016049382716049382
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.0,
74
+ "at_end_success_rate": 0.0,
75
+ "task_family_success_rate": {
76
+ "articulation_pour": 0.0
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.376,
81
+ "current_chain_depth_progress_score": 0.408,
82
+ "latched_chain_depth_progress_score": 0.448,
83
+ "avg_current_chain_depth": 2.04,
84
+ "avg_latched_chain_depth": 2.24
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": null,
88
+ "expert_time_s": 36.0,
89
+ "expert_time_step": 2160,
90
+ "avg_steps_to_success": null,
91
+ "avg_policy_steps_to_success": null,
92
+ "avg_policy_queries_to_success": null
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.00014814814814814815,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_224_wine_bottle_1": 8,
98
+ "high_speed_obj_224_wine_bottle_1": 4
99
+ },
100
+ "robot_constraint_violation_rate": 0.96,
101
+ "finger_joint_limit_saturation_rate": 0.96,
102
+ "joint_limit_active_step_rate": 0.46470370370370373,
103
+ "finger_joint_limit_saturation_step_rate": 0.4076543209876543,
104
+ "max_joint_limit_excess_rad": 1.476585865020752,
105
+ "robot_constraint_diagnostics": {
106
+ "joint_limit_soft_counts": {
107
+ "left_index_pip_joint": 79,
108
+ "left_index_pitch_joint": 34,
109
+ "left_middle_pip_joint": 62,
110
+ "left_pinky_mcp_joint": 955,
111
+ "left_pinky_pip_joint": 391,
112
+ "left_ring_mcp_joint": 640,
113
+ "left_ring_pip_joint": 399,
114
+ "left_thumb_mcp_joint": 26280,
115
+ "left_thumb_pip_joint": 10,
116
+ "right_index_pip_joint": 11,
117
+ "right_index_pitch_joint": 2,
118
+ "right_middle_pip_joint": 7,
119
+ "right_thumb_mcp_joint": 10660
120
+ },
121
+ "joint_limit_hard_counts": {
122
+ "left_index_pip_joint": 520,
123
+ "left_index_pitch_joint": 6,
124
+ "left_middle_pip_joint": 687,
125
+ "left_pinky_mcp_joint": 33,
126
+ "left_pinky_pip_joint": 52,
127
+ "left_ring_mcp_joint": 33,
128
+ "left_ring_pip_joint": 40,
129
+ "left_thumb_mcp_joint": 30,
130
+ "left_thumb_pip_joint": 3,
131
+ "right_index_pip_joint": 280,
132
+ "right_index_pitch_joint": 1,
133
+ "right_middle_pip_joint": 234,
134
+ "right_thumb_mcp_joint": 14
135
+ },
136
+ "joint_limit_hard_episode_events": 129,
137
+ "finger_joint_limit_saturation_counts": {
138
+ "left_pinky_mcp_joint": 422,
139
+ "left_pinky_pip_joint": 30,
140
+ "left_ring_mcp_joint": 268,
141
+ "left_ring_pip_joint": 67,
142
+ "left_thumb_mcp_joint": 22311,
143
+ "right_thumb_mcp_joint": 10245
144
+ },
145
+ "joint_limit_max_excess_by_joint": {
146
+ "left_index_dip_joint": 4.974401335289258e-08,
147
+ "left_index_mcp_joint": 0.0007352232933044434,
148
+ "left_index_pip_joint": 0.2337973266839981,
149
+ "left_index_pitch_joint": 0.08670616149902344,
150
+ "left_middle_dip_joint": 0.00018115094280801713,
151
+ "left_middle_mcp_joint": 0.0037761330604553223,
152
+ "left_middle_pip_joint": 0.23067748546600342,
153
+ "left_middle_pitch_joint": 1.1920928955078125e-05,
154
+ "left_pinky_dip_joint": 0.0015446245670318604,
155
+ "left_pinky_mcp_joint": 1.2026245594024658,
156
+ "left_pinky_pip_joint": 0.29773014783859253,
157
+ "left_ring_dip_joint": 0.001460728351958096,
158
+ "left_ring_mcp_joint": 1.476585865020752,
159
+ "left_ring_pip_joint": 0.3912563920021057,
160
+ "left_thumb_dip_joint": 0.0004205558798275888,
161
+ "left_thumb_mcp_joint": 0.3710569739341736,
162
+ "left_thumb_pip_joint": 0.09045249223709106,
163
+ "left_thumb_yaw_joint": 8.321014884071776e-12,
164
+ "right_index_dip_joint": 5.298456926539075e-07,
165
+ "right_index_mcp_joint": 1.1801719665527344e-05,
166
+ "right_index_pip_joint": 0.24766361713409424,
167
+ "right_index_pitch_joint": 0.0121982516720891,
168
+ "right_middle_dip_joint": 0.00010408458911115304,
169
+ "right_middle_mcp_joint": 0.0011547207832336426,
170
+ "right_middle_pip_joint": 0.017163483425974846,
171
+ "right_middle_pitch_joint": 0.0001785755157470703,
172
+ "right_pinky_dip_joint": 0.00030521073495037854,
173
+ "right_pinky_mcp_joint": 0.0018028930062428117,
174
+ "right_pinky_pip_joint": 0.0007014996954239905,
175
+ "right_ring_dip_joint": 0.0002502551069483161,
176
+ "right_ring_mcp_joint": 0.0007603414705954492,
177
+ "right_ring_pip_joint": 0.0005853768088854849,
178
+ "right_thumb_dip_joint": 1.9294868252472952e-05,
179
+ "right_thumb_mcp_joint": 0.34224799275398254,
180
+ "right_thumb_pip_joint": 2.7984886630250116e-10,
181
+ "right_thumb_yaw_joint": 3.7822961429512247e-10
182
+ },
183
+ "coverage": {
184
+ "hard_limit_source": {
185
+ "available": 25,
186
+ "unavailable": 0,
187
+ "unknown": 0
188
+ },
189
+ "joint_role_manifest": {
190
+ "available": 25,
191
+ "unavailable": 0,
192
+ "unknown": 0
193
+ }
194
+ },
195
+ "hard_limit_sources": {
196
+ "runtime.joint_pos_limits": 25
197
+ },
198
+ "soft_limit_sources": {
199
+ "runtime.soft_joint_pos_limits": 25
200
+ },
201
+ "unknown_runtime_joint_names": [],
202
+ "baseline_checks": {
203
+ "post_home": {
204
+ "checked": 25,
205
+ "hard_violation_episodes": 0
206
+ },
207
+ "post_reset": {
208
+ "checked": 25,
209
+ "hard_violation_episodes": 0
210
+ }
211
+ }
212
+ }
213
+ },
214
+ "robot_motion": {
215
+ "joint_acceleration_rms": 12.598426308495142,
216
+ "joint_effort_rms": 25.073755888586632,
217
+ "joint_jerk_rms": 1275.7403163402143,
218
+ "joint_velocity_rms": 0.4212569709572984
219
+ },
220
+ "grasp": {
221
+ "kinematic_grasp_stability_index": 0.6704500175396516,
222
+ "mean_kinematic_grasp_stability": 0.6390847309690668,
223
+ "per_object_best_kinematic_grasp_stability": {
224
+ "obj_224_wine_bottle_1": 0.6704500175396516
225
+ },
226
+ "grasp_gsi_diagnostics": {
227
+ "mean_anti_slip_score": 0.5484567901234573,
228
+ "mean_duration_score": 0.5775493827160494,
229
+ "mean_hold_score": 0.3669728395061731,
230
+ "mean_kinematic_stability_score": 0.02211458226295279,
231
+ "mean_legacy_gsi": 0.4253328866104136,
232
+ "mean_lift_score": 0.3741604938271605,
233
+ "mean_motion_stability_score": 0.3068352985747989
234
+ }
235
+ },
236
+ "tool": {
237
+ "tool_selection_accuracy": null,
238
+ "tool_switch_success_rate": null,
239
+ "tool_switch_total": 0,
240
+ "tool_switch_successful": 0
241
+ },
242
+ "runtime": {
243
+ "avg_steps": 3240,
244
+ "error_count": 0
245
+ }
246
+ }
247
+ }
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/native_shards/001/summary.json ADDED
@@ -0,0 +1,249 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA retensioned_service-3cc32133ca",
5
+ "base_seed": 103400000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 0,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3240,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.0,
50
+ "success_rate_ci": {
51
+ "mean": 0.0,
52
+ "ci_low": 0.0,
53
+ "ci_high": 0.0
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.368
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": null
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.0,
64
+ "hard_violation_rate": 0.24,
65
+ "drop_rate": 0.24,
66
+ "high_speed_violation_rate": 0.16,
67
+ "safety_violation_step_rate": 0.0001728395061728395
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.0,
74
+ "at_end_success_rate": 0.0,
75
+ "task_family_success_rate": {
76
+ "articulation_pour": 0.0
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.336,
81
+ "current_chain_depth_progress_score": 0.36,
82
+ "latched_chain_depth_progress_score": 0.368,
83
+ "avg_current_chain_depth": 1.8,
84
+ "avg_latched_chain_depth": 1.84
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": null,
88
+ "expert_time_s": 36.0,
89
+ "expert_time_step": 2160,
90
+ "avg_steps_to_success": null,
91
+ "avg_policy_steps_to_success": null,
92
+ "avg_policy_queries_to_success": null
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.0001234567901234568,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_224_wine_bottle_1": 6,
98
+ "high_speed_obj_224_wine_bottle_1": 4
99
+ },
100
+ "robot_constraint_violation_rate": 1.0,
101
+ "finger_joint_limit_saturation_rate": 1.0,
102
+ "joint_limit_active_step_rate": 0.32298765432098764,
103
+ "finger_joint_limit_saturation_step_rate": 0.29271604938271606,
104
+ "max_joint_limit_excess_rad": 1.5886521339416504,
105
+ "robot_constraint_diagnostics": {
106
+ "joint_limit_soft_counts": {
107
+ "left_index_pip_joint": 229,
108
+ "left_index_pitch_joint": 72,
109
+ "left_middle_pip_joint": 1093,
110
+ "left_middle_pitch_joint": 43,
111
+ "left_pinky_mcp_joint": 1794,
112
+ "left_pinky_pip_joint": 479,
113
+ "left_ring_mcp_joint": 818,
114
+ "left_ring_pip_joint": 510,
115
+ "left_thumb_mcp_joint": 17988,
116
+ "left_thumb_pip_joint": 32,
117
+ "right_middle_pip_joint": 4,
118
+ "right_thumb_mcp_joint": 6635
119
+ },
120
+ "joint_limit_hard_counts": {
121
+ "left_index_pip_joint": 368,
122
+ "left_index_pitch_joint": 6,
123
+ "left_middle_pip_joint": 289,
124
+ "left_middle_pitch_joint": 1,
125
+ "left_pinky_mcp_joint": 30,
126
+ "left_pinky_pip_joint": 108,
127
+ "left_ring_mcp_joint": 38,
128
+ "left_ring_pip_joint": 59,
129
+ "left_thumb_mcp_joint": 32,
130
+ "left_thumb_pip_joint": 3,
131
+ "right_index_pip_joint": 146,
132
+ "right_middle_pip_joint": 244,
133
+ "right_thumb_mcp_joint": 8
134
+ },
135
+ "joint_limit_hard_episode_events": 131,
136
+ "finger_joint_limit_saturation_counts": {
137
+ "left_index_pip_joint": 42,
138
+ "left_index_pitch_joint": 51,
139
+ "left_middle_pip_joint": 676,
140
+ "left_middle_pitch_joint": 37,
141
+ "left_pinky_mcp_joint": 938,
142
+ "left_pinky_pip_joint": 56,
143
+ "left_ring_mcp_joint": 304,
144
+ "left_ring_pip_joint": 153,
145
+ "left_thumb_mcp_joint": 16413,
146
+ "right_thumb_mcp_joint": 6330
147
+ },
148
+ "joint_limit_max_excess_by_joint": {
149
+ "left_index_dip_joint": 3.433893525084386e-08,
150
+ "left_index_mcp_joint": 0.0008804500102996826,
151
+ "left_index_pip_joint": 0.6620429754257202,
152
+ "left_index_pitch_joint": 1.0985686779022217,
153
+ "left_middle_dip_joint": 0.00013008569658268243,
154
+ "left_middle_mcp_joint": 0.0031856298446655273,
155
+ "left_middle_pip_joint": 1.1455157995224,
156
+ "left_middle_pitch_joint": 0.8504464030265808,
157
+ "left_pinky_dip_joint": 0.002340366831049323,
158
+ "left_pinky_mcp_joint": 1.4706686735153198,
159
+ "left_pinky_pip_joint": 1.1555039882659912,
160
+ "left_ring_dip_joint": 0.0021921969018876553,
161
+ "left_ring_mcp_joint": 1.366401195526123,
162
+ "left_ring_pip_joint": 1.5886521339416504,
163
+ "left_thumb_dip_joint": 0.0003835729730781168,
164
+ "left_thumb_mcp_joint": 0.49498772621154785,
165
+ "left_thumb_pip_joint": 0.22042275965213776,
166
+ "right_index_dip_joint": 2.8910730520692596e-07,
167
+ "right_index_mcp_joint": 1.823902130126953e-05,
168
+ "right_index_pip_joint": 0.008097084239125252,
169
+ "right_index_pitch_joint": 0.0001628398895263672,
170
+ "right_middle_dip_joint": 0.0001278225681744516,
171
+ "right_middle_mcp_joint": 0.0007988214492797852,
172
+ "right_middle_pip_joint": 0.012908851727843285,
173
+ "right_middle_pitch_joint": 0.00017368793487548828,
174
+ "right_pinky_dip_joint": 0.00028790676151402295,
175
+ "right_pinky_mcp_joint": 0.0019511537393555045,
176
+ "right_pinky_pip_joint": 0.0006625486421398818,
177
+ "right_ring_dip_joint": 0.0002571576333139092,
178
+ "right_ring_mcp_joint": 0.0007245147135108709,
179
+ "right_ring_pip_joint": 0.0005985461175441742,
180
+ "right_thumb_dip_joint": 3.949963502236642e-05,
181
+ "right_thumb_mcp_joint": 0.3139530420303345,
182
+ "right_thumb_pip_joint": 1.0328363858613798e-10,
183
+ "right_thumb_yaw_joint": 3.1474323147762107e-11
184
+ },
185
+ "coverage": {
186
+ "hard_limit_source": {
187
+ "available": 25,
188
+ "unavailable": 0,
189
+ "unknown": 0
190
+ },
191
+ "joint_role_manifest": {
192
+ "available": 25,
193
+ "unavailable": 0,
194
+ "unknown": 0
195
+ }
196
+ },
197
+ "hard_limit_sources": {
198
+ "runtime.joint_pos_limits": 25
199
+ },
200
+ "soft_limit_sources": {
201
+ "runtime.soft_joint_pos_limits": 25
202
+ },
203
+ "unknown_runtime_joint_names": [],
204
+ "baseline_checks": {
205
+ "post_home": {
206
+ "checked": 25,
207
+ "hard_violation_episodes": 0
208
+ },
209
+ "post_reset": {
210
+ "checked": 25,
211
+ "hard_violation_episodes": 0
212
+ }
213
+ }
214
+ }
215
+ },
216
+ "robot_motion": {
217
+ "joint_acceleration_rms": 13.245178791773396,
218
+ "joint_effort_rms": 23.35393582416703,
219
+ "joint_jerk_rms": 1366.6838832977176,
220
+ "joint_velocity_rms": 0.40660764817041156
221
+ },
222
+ "grasp": {
223
+ "kinematic_grasp_stability_index": 0.5066161817339326,
224
+ "mean_kinematic_grasp_stability": 0.6897640207748883,
225
+ "per_object_best_kinematic_grasp_stability": {
226
+ "obj_224_wine_bottle_1": 0.5066161817339326
227
+ },
228
+ "grasp_gsi_diagnostics": {
229
+ "mean_anti_slip_score": 0.6537325102880658,
230
+ "mean_duration_score": 0.42923456790123454,
231
+ "mean_hold_score": 0.32748950617283956,
232
+ "mean_kinematic_stability_score": 0.07519069371242543,
233
+ "mean_legacy_gsi": 0.3973492859298234,
234
+ "mean_lift_score": 0.3324814814814815,
235
+ "mean_motion_stability_score": 0.28223680259101513
236
+ }
237
+ },
238
+ "tool": {
239
+ "tool_selection_accuracy": null,
240
+ "tool_switch_success_rate": null,
241
+ "tool_switch_total": 0,
242
+ "tool_switch_successful": 0
243
+ },
244
+ "runtime": {
245
+ "avg_steps": 3240,
246
+ "error_count": 0
247
+ }
248
+ }
249
+ }
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Equi/summary.json ADDED
@@ -0,0 +1,253 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA retensioned_service-3cc32133ca",
5
+ "base_seed": 103400000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 50,
9
+ "successes": 0,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3240,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.0,
50
+ "success_rate_ci": {
51
+ "mean": 0.0,
52
+ "ci_low": 0.0,
53
+ "ci_high": 0.0
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.40800000000000003
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": null
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.0,
64
+ "hard_violation_rate": 0.28,
65
+ "drop_rate": 0.28,
66
+ "high_speed_violation_rate": 0.16,
67
+ "safety_violation_step_rate": 0.00016666666666666666
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.0,
74
+ "at_end_success_rate": 0.0,
75
+ "task_family_success_rate": {
76
+ "articulation_pour": 0.0
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.356,
81
+ "current_chain_depth_progress_score": 0.384,
82
+ "latched_chain_depth_progress_score": 0.40800000000000003,
83
+ "avg_current_chain_depth": 1.92,
84
+ "avg_latched_chain_depth": 2.04
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": null,
88
+ "expert_time_s": 36.0,
89
+ "expert_time_step": 2160,
90
+ "avg_steps_to_success": null,
91
+ "avg_policy_steps_to_success": null,
92
+ "avg_policy_queries_to_success": null
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.00013580246913580247,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_224_wine_bottle_1": 14,
98
+ "high_speed_obj_224_wine_bottle_1": 8
99
+ },
100
+ "robot_constraint_violation_rate": 0.98,
101
+ "finger_joint_limit_saturation_rate": 0.98,
102
+ "joint_limit_active_step_rate": 0.3938456790123457,
103
+ "finger_joint_limit_saturation_step_rate": 0.3501851851851852,
104
+ "max_joint_limit_excess_rad": 1.5886521339416504,
105
+ "robot_constraint_diagnostics": {
106
+ "joint_limit_soft_counts": {
107
+ "left_index_pip_joint": 308,
108
+ "left_index_pitch_joint": 106,
109
+ "left_middle_pip_joint": 1155,
110
+ "left_middle_pitch_joint": 43,
111
+ "left_pinky_mcp_joint": 2749,
112
+ "left_pinky_pip_joint": 870,
113
+ "left_ring_mcp_joint": 1458,
114
+ "left_ring_pip_joint": 909,
115
+ "left_thumb_mcp_joint": 44268,
116
+ "left_thumb_pip_joint": 42,
117
+ "right_index_pip_joint": 11,
118
+ "right_index_pitch_joint": 2,
119
+ "right_middle_pip_joint": 11,
120
+ "right_thumb_mcp_joint": 17295
121
+ },
122
+ "joint_limit_hard_counts": {
123
+ "left_index_pip_joint": 888,
124
+ "left_index_pitch_joint": 12,
125
+ "left_middle_pip_joint": 976,
126
+ "left_middle_pitch_joint": 1,
127
+ "left_pinky_mcp_joint": 63,
128
+ "left_pinky_pip_joint": 160,
129
+ "left_ring_mcp_joint": 71,
130
+ "left_ring_pip_joint": 99,
131
+ "left_thumb_mcp_joint": 62,
132
+ "left_thumb_pip_joint": 6,
133
+ "right_index_pip_joint": 426,
134
+ "right_index_pitch_joint": 1,
135
+ "right_middle_pip_joint": 478,
136
+ "right_thumb_mcp_joint": 22
137
+ },
138
+ "joint_limit_hard_episode_events": 260,
139
+ "finger_joint_limit_saturation_counts": {
140
+ "left_index_pip_joint": 42,
141
+ "left_index_pitch_joint": 51,
142
+ "left_middle_pip_joint": 676,
143
+ "left_middle_pitch_joint": 37,
144
+ "left_pinky_mcp_joint": 1360,
145
+ "left_pinky_pip_joint": 86,
146
+ "left_ring_mcp_joint": 572,
147
+ "left_ring_pip_joint": 220,
148
+ "left_thumb_mcp_joint": 38724,
149
+ "right_thumb_mcp_joint": 16575
150
+ },
151
+ "joint_limit_max_excess_by_joint": {
152
+ "left_index_dip_joint": 4.974401335289258e-08,
153
+ "left_index_mcp_joint": 0.0008804500102996826,
154
+ "left_index_pip_joint": 0.6620429754257202,
155
+ "left_index_pitch_joint": 1.0985686779022217,
156
+ "left_middle_dip_joint": 0.00018115094280801713,
157
+ "left_middle_mcp_joint": 0.0037761330604553223,
158
+ "left_middle_pip_joint": 1.1455157995224,
159
+ "left_middle_pitch_joint": 0.8504464030265808,
160
+ "left_pinky_dip_joint": 0.002340366831049323,
161
+ "left_pinky_mcp_joint": 1.4706686735153198,
162
+ "left_pinky_pip_joint": 1.1555039882659912,
163
+ "left_ring_dip_joint": 0.0021921969018876553,
164
+ "left_ring_mcp_joint": 1.476585865020752,
165
+ "left_ring_pip_joint": 1.5886521339416504,
166
+ "left_thumb_dip_joint": 0.0004205558798275888,
167
+ "left_thumb_mcp_joint": 0.49498772621154785,
168
+ "left_thumb_pip_joint": 0.22042275965213776,
169
+ "left_thumb_yaw_joint": 8.321014884071776e-12,
170
+ "right_index_dip_joint": 5.298456926539075e-07,
171
+ "right_index_mcp_joint": 1.823902130126953e-05,
172
+ "right_index_pip_joint": 0.24766361713409424,
173
+ "right_index_pitch_joint": 0.0121982516720891,
174
+ "right_middle_dip_joint": 0.0001278225681744516,
175
+ "right_middle_mcp_joint": 0.0011547207832336426,
176
+ "right_middle_pip_joint": 0.017163483425974846,
177
+ "right_middle_pitch_joint": 0.0001785755157470703,
178
+ "right_pinky_dip_joint": 0.00030521073495037854,
179
+ "right_pinky_mcp_joint": 0.0019511537393555045,
180
+ "right_pinky_pip_joint": 0.0007014996954239905,
181
+ "right_ring_dip_joint": 0.0002571576333139092,
182
+ "right_ring_mcp_joint": 0.0007603414705954492,
183
+ "right_ring_pip_joint": 0.0005985461175441742,
184
+ "right_thumb_dip_joint": 3.949963502236642e-05,
185
+ "right_thumb_mcp_joint": 0.34224799275398254,
186
+ "right_thumb_pip_joint": 2.7984886630250116e-10,
187
+ "right_thumb_yaw_joint": 3.7822961429512247e-10
188
+ },
189
+ "coverage": {
190
+ "hard_limit_source": {
191
+ "available": 50,
192
+ "unavailable": 0,
193
+ "unknown": 0
194
+ },
195
+ "joint_role_manifest": {
196
+ "available": 50,
197
+ "unavailable": 0,
198
+ "unknown": 0
199
+ }
200
+ },
201
+ "hard_limit_sources": {
202
+ "runtime.joint_pos_limits": 50
203
+ },
204
+ "soft_limit_sources": {
205
+ "runtime.soft_joint_pos_limits": 50
206
+ },
207
+ "unknown_runtime_joint_names": [],
208
+ "baseline_checks": {
209
+ "post_home": {
210
+ "checked": 50,
211
+ "hard_violation_episodes": 0
212
+ },
213
+ "post_reset": {
214
+ "checked": 50,
215
+ "hard_violation_episodes": 0
216
+ }
217
+ }
218
+ }
219
+ },
220
+ "robot_motion": {
221
+ "joint_acceleration_rms": 12.921802550134268,
222
+ "joint_effort_rms": 24.21384585637683,
223
+ "joint_jerk_rms": 1321.212099818966,
224
+ "joint_velocity_rms": 0.413932309563855
225
+ },
226
+ "grasp": {
227
+ "kinematic_grasp_stability_index": 0.5885330996367921,
228
+ "mean_kinematic_grasp_stability": 0.6605850357351729,
229
+ "per_object_best_kinematic_grasp_stability": {
230
+ "obj_224_wine_bottle_1": 0.5885330996367921
231
+ },
232
+ "grasp_gsi_diagnostics": {
233
+ "mean_anti_slip_score": 0.6010946502057615,
234
+ "mean_duration_score": 0.503391975308642,
235
+ "mean_hold_score": 0.34723117283950633,
236
+ "mean_kinematic_stability_score": 0.04865263798768911,
237
+ "mean_legacy_gsi": 0.4113410862701185,
238
+ "mean_lift_score": 0.35332098765432096,
239
+ "mean_motion_stability_score": 0.29453605058290705
240
+ }
241
+ },
242
+ "tool": {
243
+ "tool_selection_accuracy": null,
244
+ "tool_switch_success_rate": null,
245
+ "tool_switch_total": 0,
246
+ "tool_switch_successful": 0
247
+ },
248
+ "runtime": {
249
+ "avg_steps": 3240,
250
+ "error_count": 0
251
+ }
252
+ }
253
+ }
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Full/native_shards/000/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Full/native_shards/000/summary.json ADDED
@@ -0,0 +1,246 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA retensioned_service-3cc32133ca",
5
+ "base_seed": 203400000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 0,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3240,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.0,
50
+ "success_rate_ci": {
51
+ "mean": 0.0,
52
+ "ci_low": 0.0,
53
+ "ci_high": 0.0
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.424
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": null
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.0,
64
+ "hard_violation_rate": 0.36,
65
+ "drop_rate": 0.36,
66
+ "high_speed_violation_rate": 0.16,
67
+ "safety_violation_step_rate": 0.0002345679012345679
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.0,
74
+ "at_end_success_rate": 0.0,
75
+ "task_family_success_rate": {
76
+ "articulation_pour": 0.0
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.328,
81
+ "current_chain_depth_progress_score": 0.376,
82
+ "latched_chain_depth_progress_score": 0.424,
83
+ "avg_current_chain_depth": 1.88,
84
+ "avg_latched_chain_depth": 2.12
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": null,
88
+ "expert_time_s": 36.0,
89
+ "expert_time_step": 2160,
90
+ "avg_steps_to_success": null,
91
+ "avg_policy_steps_to_success": null,
92
+ "avg_policy_queries_to_success": null
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 0.0001728395061728395,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_224_wine_bottle_1": 9,
98
+ "high_speed_obj_029_plate_2": 1,
99
+ "high_speed_obj_224_wine_bottle_1": 4
100
+ },
101
+ "robot_constraint_violation_rate": 1.0,
102
+ "finger_joint_limit_saturation_rate": 1.0,
103
+ "joint_limit_active_step_rate": 0.4278148148148148,
104
+ "finger_joint_limit_saturation_step_rate": 0.2837654320987654,
105
+ "max_joint_limit_excess_rad": 1.5045123100280762,
106
+ "robot_constraint_diagnostics": {
107
+ "joint_limit_soft_counts": {
108
+ "left_index_pip_joint": 93,
109
+ "left_index_pitch_joint": 38,
110
+ "left_middle_pip_joint": 985,
111
+ "left_pinky_mcp_joint": 6330,
112
+ "left_pinky_pip_joint": 1795,
113
+ "left_ring_mcp_joint": 3180,
114
+ "left_ring_pip_joint": 652,
115
+ "left_thumb_mcp_joint": 24515,
116
+ "left_thumb_pip_joint": 27,
117
+ "right_middle_pip_joint": 5,
118
+ "right_thumb_mcp_joint": 3947
119
+ },
120
+ "joint_limit_hard_counts": {
121
+ "left_index_pip_joint": 394,
122
+ "left_index_pitch_joint": 8,
123
+ "left_middle_pip_joint": 1521,
124
+ "left_pinky_mcp_joint": 40,
125
+ "left_pinky_pip_joint": 122,
126
+ "left_ring_mcp_joint": 38,
127
+ "left_ring_pip_joint": 378,
128
+ "left_thumb_mcp_joint": 36,
129
+ "left_thumb_pip_joint": 2,
130
+ "right_index_pip_joint": 166,
131
+ "right_middle_pip_joint": 178,
132
+ "right_thumb_mcp_joint": 7
133
+ },
134
+ "joint_limit_hard_episode_events": 117,
135
+ "finger_joint_limit_saturation_counts": {
136
+ "left_index_pip_joint": 2,
137
+ "left_middle_pip_joint": 15,
138
+ "left_pinky_mcp_joint": 4032,
139
+ "left_pinky_pip_joint": 31,
140
+ "left_ring_mcp_joint": 2848,
141
+ "left_ring_pip_joint": 60,
142
+ "left_thumb_mcp_joint": 15138,
143
+ "right_thumb_mcp_joint": 3724
144
+ },
145
+ "joint_limit_max_excess_by_joint": {
146
+ "left_index_dip_joint": 3.0073383783246754e-08,
147
+ "left_index_mcp_joint": 0.00043001770973205566,
148
+ "left_index_pip_joint": 0.272444486618042,
149
+ "left_index_pitch_joint": 0.12401819974184036,
150
+ "left_middle_dip_joint": 0.00014909535821061581,
151
+ "left_middle_mcp_joint": 0.0028853416442871094,
152
+ "left_middle_pip_joint": 0.3207648992538452,
153
+ "left_middle_pitch_joint": 8.571147918701172e-05,
154
+ "left_pinky_dip_joint": 0.0017986830789595842,
155
+ "left_pinky_mcp_joint": 1.4388014078140259,
156
+ "left_pinky_pip_joint": 0.35989105701446533,
157
+ "left_ring_dip_joint": 0.0016349328216165304,
158
+ "left_ring_mcp_joint": 1.5045123100280762,
159
+ "left_ring_pip_joint": 0.4254114329814911,
160
+ "left_thumb_dip_joint": 0.0003955458232667297,
161
+ "left_thumb_mcp_joint": 0.32935643196105957,
162
+ "left_thumb_pip_joint": 0.1844843178987503,
163
+ "right_index_dip_joint": 5.465116714731266e-08,
164
+ "right_index_mcp_joint": 1.5407800674438477e-05,
165
+ "right_index_pip_joint": 0.008035627193748951,
166
+ "right_index_pitch_joint": 9.34600830078125e-05,
167
+ "right_middle_dip_joint": 9.460913861403242e-05,
168
+ "right_middle_mcp_joint": 0.0008990764617919922,
169
+ "right_middle_pip_joint": 0.013511516153812408,
170
+ "right_middle_pitch_joint": 0.00010406970977783203,
171
+ "right_pinky_dip_joint": 0.00023830337158869952,
172
+ "right_pinky_mcp_joint": 0.0014220057055354118,
173
+ "right_pinky_pip_joint": 0.0005826842971146107,
174
+ "right_ring_dip_joint": 0.0002042955457000062,
175
+ "right_ring_mcp_joint": 0.0006972637493163347,
176
+ "right_ring_pip_joint": 0.0005319320480339229,
177
+ "right_thumb_dip_joint": 4.43896533397492e-06,
178
+ "right_thumb_mcp_joint": 0.3152830898761749,
179
+ "right_thumb_pip_joint": 1.0334607475348534e-10,
180
+ "right_thumb_yaw_joint": 3.263915179796406e-11
181
+ },
182
+ "coverage": {
183
+ "hard_limit_source": {
184
+ "available": 25,
185
+ "unavailable": 0,
186
+ "unknown": 0
187
+ },
188
+ "joint_role_manifest": {
189
+ "available": 25,
190
+ "unavailable": 0,
191
+ "unknown": 0
192
+ }
193
+ },
194
+ "hard_limit_sources": {
195
+ "runtime.joint_pos_limits": 25
196
+ },
197
+ "soft_limit_sources": {
198
+ "runtime.soft_joint_pos_limits": 25
199
+ },
200
+ "unknown_runtime_joint_names": [],
201
+ "baseline_checks": {
202
+ "post_home": {
203
+ "checked": 25,
204
+ "hard_violation_episodes": 0
205
+ },
206
+ "post_reset": {
207
+ "checked": 25,
208
+ "hard_violation_episodes": 0
209
+ }
210
+ }
211
+ }
212
+ },
213
+ "robot_motion": {
214
+ "joint_acceleration_rms": 15.388080491007328,
215
+ "joint_effort_rms": 23.013178175867317,
216
+ "joint_jerk_rms": 1597.1392733862529,
217
+ "joint_velocity_rms": 0.4521382388475974
218
+ },
219
+ "grasp": {
220
+ "kinematic_grasp_stability_index": 0.6172920728205088,
221
+ "mean_kinematic_grasp_stability": 0.7114876756629162,
222
+ "per_object_best_kinematic_grasp_stability": {
223
+ "obj_224_wine_bottle_1": 0.6172920728205088
224
+ },
225
+ "grasp_gsi_diagnostics": {
226
+ "mean_anti_slip_score": 0.7028271604938273,
227
+ "mean_duration_score": 0.5170308641975309,
228
+ "mean_hold_score": 0.29813271604938285,
229
+ "mean_kinematic_stability_score": 0.13009440965474467,
230
+ "mean_legacy_gsi": 0.4121830249475945,
231
+ "mean_lift_score": 0.30392592592592593,
232
+ "mean_motion_stability_score": 0.26459052017908086
233
+ }
234
+ },
235
+ "tool": {
236
+ "tool_selection_accuracy": null,
237
+ "tool_switch_success_rate": null,
238
+ "tool_switch_total": 0,
239
+ "tool_switch_successful": 0
240
+ },
241
+ "runtime": {
242
+ "avg_steps": 3240,
243
+ "error_count": 0
244
+ }
245
+ }
246
+ }
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Full/native_shards/001/per_episode.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
pyrua-dex/per_task_eval/34_fridge_wine_interhand_pour/Full/native_shards/001/summary.json ADDED
@@ -0,0 +1,247 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metric_schema_version": "core-v1.8-policy-step-query-split",
3
+ "metadata": {
4
+ "policy_name": "PyRUA retensioned_service-3cc32133ca",
5
+ "base_seed": 203400000,
6
+ "seed_policy": "episode_seed = base_seed + episode_index - 1",
7
+ "randomization_recorded": true,
8
+ "n_episodes": 25,
9
+ "successes": 0,
10
+ "error_count": 0,
11
+ "evaluation_protocol": "reach_and_stop",
12
+ "early_stop": true,
13
+ "success_definition": "terminal.raw_condition holds for dwell_time_s",
14
+ "measurement_window": "episode_start_to_stable_success_or_failure_budget",
15
+ "max_episode_steps": 3240,
16
+ "dwell_time_s": 0.5,
17
+ "field_semantics": {
18
+ "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
19
+ "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
20
+ "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
21
+ "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
22
+ "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
23
+ "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
24
+ "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
25
+ "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
26
+ "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
27
+ "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
28
+ "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
29
+ "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
30
+ "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
31
+ "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
32
+ "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
33
+ "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
34
+ "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
35
+ "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
36
+ "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
37
+ "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
38
+ "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
39
+ "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
40
+ "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
41
+ "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
42
+ "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
43
+ "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
44
+ "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
45
+ }
46
+ },
47
+ "core_metrics": {
48
+ "completion": {
49
+ "success_rate": 0.0,
50
+ "success_rate_ci": {
51
+ "mean": 0.0,
52
+ "ci_low": 0.0,
53
+ "ci_high": 0.0
54
+ }
55
+ },
56
+ "progress": {
57
+ "latched_stage_completion_rate": 0.368
58
+ },
59
+ "efficiency": {
60
+ "avg_time_to_success_s": null
61
+ },
62
+ "safety": {
63
+ "safe_success_rate": 0.0,
64
+ "hard_violation_rate": 0.16,
65
+ "drop_rate": 0.16,
66
+ "high_speed_violation_rate": 0.08,
67
+ "safety_violation_step_rate": 0.00018518518518518518
68
+ },
69
+ "robustness": {}
70
+ },
71
+ "diagnostic_metrics": {
72
+ "completion": {
73
+ "ever_instant_success_rate": 0.0,
74
+ "at_end_success_rate": 0.0,
75
+ "task_family_success_rate": {
76
+ "articulation_pour": 0.0
77
+ }
78
+ },
79
+ "progress": {
80
+ "current_stage_completion_rate": 0.34400000000000003,
81
+ "current_chain_depth_progress_score": 0.36,
82
+ "latched_chain_depth_progress_score": 0.368,
83
+ "avg_current_chain_depth": 1.8,
84
+ "avg_latched_chain_depth": 1.84
85
+ },
86
+ "efficiency": {
87
+ "task_efficiency": null,
88
+ "expert_time_s": 36.0,
89
+ "expert_time_step": 2160,
90
+ "avg_steps_to_success": null,
91
+ "avg_policy_steps_to_success": null,
92
+ "avg_policy_queries_to_success": null
93
+ },
94
+ "safety": {
95
+ "safety_violation_events_per_step": 8.641975308641975e-05,
96
+ "violation_counts": {
97
+ "drop_to_table_obj_224_wine_bottle_1": 4,
98
+ "high_speed_obj_194_wineglass_2": 1,
99
+ "high_speed_obj_224_wine_bottle_1": 2
100
+ },
101
+ "robot_constraint_violation_rate": 1.0,
102
+ "finger_joint_limit_saturation_rate": 1.0,
103
+ "joint_limit_active_step_rate": 0.4138148148148148,
104
+ "finger_joint_limit_saturation_step_rate": 0.36318518518518517,
105
+ "max_joint_limit_excess_rad": 1.6308687925338745,
106
+ "robot_constraint_diagnostics": {
107
+ "joint_limit_soft_counts": {
108
+ "left_index_pip_joint": 154,
109
+ "left_index_pitch_joint": 25,
110
+ "left_middle_pip_joint": 991,
111
+ "left_pinky_mcp_joint": 3861,
112
+ "left_pinky_pip_joint": 473,
113
+ "left_ring_mcp_joint": 8849,
114
+ "left_ring_pip_joint": 568,
115
+ "left_thumb_mcp_joint": 20626,
116
+ "left_thumb_pip_joint": 40,
117
+ "right_middle_pip_joint": 24,
118
+ "right_thumb_mcp_joint": 4744
119
+ },
120
+ "joint_limit_hard_counts": {
121
+ "left_index_pip_joint": 223,
122
+ "left_index_pitch_joint": 7,
123
+ "left_middle_mcp_joint": 1,
124
+ "left_middle_pip_joint": 1366,
125
+ "left_pinky_mcp_joint": 42,
126
+ "left_pinky_pip_joint": 1166,
127
+ "left_ring_mcp_joint": 41,
128
+ "left_ring_pip_joint": 1753,
129
+ "left_thumb_mcp_joint": 321,
130
+ "left_thumb_pip_joint": 6,
131
+ "right_index_pip_joint": 551,
132
+ "right_middle_pip_joint": 110,
133
+ "right_thumb_mcp_joint": 5
134
+ },
135
+ "joint_limit_hard_episode_events": 127,
136
+ "finger_joint_limit_saturation_counts": {
137
+ "left_index_pip_joint": 10,
138
+ "left_middle_pip_joint": 6,
139
+ "left_pinky_mcp_joint": 3398,
140
+ "left_pinky_pip_joint": 43,
141
+ "left_ring_mcp_joint": 8420,
142
+ "left_ring_pip_joint": 113,
143
+ "left_thumb_mcp_joint": 16130,
144
+ "right_thumb_mcp_joint": 4601
145
+ },
146
+ "joint_limit_max_excess_by_joint": {
147
+ "left_index_dip_joint": 3.408552018413502e-08,
148
+ "left_index_mcp_joint": 0.0010075569152832031,
149
+ "left_index_pip_joint": 0.4152810573577881,
150
+ "left_index_pitch_joint": 0.12891508638858795,
151
+ "left_middle_dip_joint": 0.0001399490429321304,
152
+ "left_middle_mcp_joint": 0.005136996507644653,
153
+ "left_middle_pip_joint": 0.32733118534088135,
154
+ "left_middle_pitch_joint": 0.0001316070556640625,
155
+ "left_pinky_dip_joint": 0.001334750559180975,
156
+ "left_pinky_mcp_joint": 1.3784568309783936,
157
+ "left_pinky_pip_joint": 0.32053330540657043,
158
+ "left_ring_dip_joint": 0.00130355020519346,
159
+ "left_ring_mcp_joint": 1.6308687925338745,
160
+ "left_ring_pip_joint": 0.4477488696575165,
161
+ "left_thumb_dip_joint": 0.0004473574517760426,
162
+ "left_thumb_mcp_joint": 0.5096300840377808,
163
+ "left_thumb_pip_joint": 0.17832864820957184,
164
+ "right_index_dip_joint": 2.3801575294024602e-11,
165
+ "right_index_mcp_joint": 1.2367963790893555e-05,
166
+ "right_index_pip_joint": 0.009393181651830673,
167
+ "right_index_pitch_joint": 7.522106170654297e-05,
168
+ "right_middle_dip_joint": 0.00012741322279907763,
169
+ "right_middle_mcp_joint": 0.0007479190826416016,
170
+ "right_middle_pip_joint": 0.02005964331328869,
171
+ "right_middle_pitch_joint": 8.845329284667969e-05,
172
+ "right_pinky_dip_joint": 0.0003100998292211443,
173
+ "right_pinky_mcp_joint": 0.0014221431920304894,
174
+ "right_pinky_pip_joint": 0.0007375688874162734,
175
+ "right_ring_dip_joint": 0.000250101467827335,
176
+ "right_ring_mcp_joint": 0.0006604689406231046,
177
+ "right_ring_pip_joint": 0.000629979360383004,
178
+ "right_thumb_dip_joint": 1.6887446690816432e-05,
179
+ "right_thumb_mcp_joint": 0.32737788558006287,
180
+ "right_thumb_pip_joint": 1.0332932426360131e-10,
181
+ "right_thumb_yaw_joint": 3.161778477922539e-11
182
+ },
183
+ "coverage": {
184
+ "hard_limit_source": {
185
+ "available": 25,
186
+ "unavailable": 0,
187
+ "unknown": 0
188
+ },
189
+ "joint_role_manifest": {
190
+ "available": 25,
191
+ "unavailable": 0,
192
+ "unknown": 0
193
+ }
194
+ },
195
+ "hard_limit_sources": {
196
+ "runtime.joint_pos_limits": 25
197
+ },
198
+ "soft_limit_sources": {
199
+ "runtime.soft_joint_pos_limits": 25
200
+ },
201
+ "unknown_runtime_joint_names": [],
202
+ "baseline_checks": {
203
+ "post_home": {
204
+ "checked": 25,
205
+ "hard_violation_episodes": 0
206
+ },
207
+ "post_reset": {
208
+ "checked": 25,
209
+ "hard_violation_episodes": 0
210
+ }
211
+ }
212
+ }
213
+ },
214
+ "robot_motion": {
215
+ "joint_acceleration_rms": 18.907304211549793,
216
+ "joint_effort_rms": 27.593549039037274,
217
+ "joint_jerk_rms": 1998.9176453651467,
218
+ "joint_velocity_rms": 0.5624379589891351
219
+ },
220
+ "grasp": {
221
+ "kinematic_grasp_stability_index": 0.41793228762026896,
222
+ "mean_kinematic_grasp_stability": 0.5780066250732315,
223
+ "per_object_best_kinematic_grasp_stability": {
224
+ "obj_224_wine_bottle_1": 0.41793228762026896
225
+ },
226
+ "grasp_gsi_diagnostics": {
227
+ "mean_anti_slip_score": 0.7095596707818931,
228
+ "mean_duration_score": 0.3380283950617284,
229
+ "mean_hold_score": 0.2532012345679013,
230
+ "mean_kinematic_stability_score": 0.043144712659870174,
231
+ "mean_legacy_gsi": 0.3479237221554561,
232
+ "mean_lift_score": 0.2572716049382716,
233
+ "mean_motion_stability_score": 0.21545191555397694
234
+ }
235
+ },
236
+ "tool": {
237
+ "tool_selection_accuracy": null,
238
+ "tool_switch_success_rate": null,
239
+ "tool_switch_total": 0,
240
+ "tool_switch_successful": 0
241
+ },
242
+ "runtime": {
243
+ "avg_steps": 3240,
244
+ "error_count": 0
245
+ }
246
+ }
247
+ }