anhtld commited on
Commit
866675d
·
verified ·
1 Parent(s): bcbfc2d

auto-sync 2026-07-02T16:56:12Z workspace (part 5)

Browse files
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025_summary.json ADDED
@@ -0,0 +1,577 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
3
+ "objective": "transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1",
4
+ "out_name": "policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.3884057971014493,
9
+ "std_success": 0.01759303293390554,
10
+ "mean_progress": 0.5993607256448139,
11
+ "mean_action_mse_to_best": 0.5128867500585814,
12
+ "gain_vs_h4": 0.09170579710144927,
13
+ "gain_vs_h16_rank_checkpoint": 0.09101449275362322,
14
+ "rows": [
15
+ {
16
+ "seed": 0,
17
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_0/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 48,
21
+ "candidate_sigma": 0.0,
22
+ "proposal_types": [],
23
+ "proposal_lattice_types": [],
24
+ "selection_margin": 0.0,
25
+ "prepend_policy_candidate": false,
26
+ "field_optim_steps": 0,
27
+ "field_optim_step_size": 0.0,
28
+ "field_optim_trust_radius": 0.0,
29
+ "field_optim_l2_penalty": 0.0,
30
+ "retrieval_neighbors": 6,
31
+ "retrieval_metric": "raw",
32
+ "retrieval_type_min_success": 0.0,
33
+ "retrieval_type_success_bonus_scale": 0.0,
34
+ "retrieval_residual_consensus_penalty_scale": 0.0,
35
+ "retrieval_residual_min_source_progress": 0.0,
36
+ "retrieval_residual_min_source_advantage": -1000000000.0,
37
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
38
+ "retrieval_residual_source_score_bonus_scale": 0.0,
39
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
40
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
41
+ "retrieval_residual_scale": 1.0,
42
+ "retrieval_residual_scales": [
43
+ 0.35,
44
+ 0.4,
45
+ 0.45
46
+ ],
47
+ "retrieval_residual_anchor": "expert",
48
+ "retrieval_residual_direction": "candidate_minus_anchor",
49
+ "retrieval_residual_reduce": "compose_mean_by_type",
50
+ "retrieval_residual_challenger_types": [],
51
+ "retrieval_residual_challenger_scales": [],
52
+ "retrieval_residual_challenger_margin": 0.0,
53
+ "retrieval_residual_challenger_type_margins": {},
54
+ "retrieval_residual_challenger_tasks": [],
55
+ "retrieval_residual_challenger_type_tasks": {},
56
+ "lattice_exclude_types": [
57
+ "residual_random_negative",
58
+ "residual_wrong_direction",
59
+ "residual_near_miss+residual_no_op",
60
+ "residual_no_op+residual_wrong_gripper"
61
+ ],
62
+ "lattice_exclude_type_tasks": {},
63
+ "candidate_type_bonuses": {},
64
+ "candidate_type_bonuses_by_task": {},
65
+ "residual_scale_bonuses_by_task": {
66
+ "*": {
67
+ "0.35": -9.674979510607974e-05,
68
+ "0.4": 0.0003106990953088251,
69
+ "0.45": -0.00016780893426614763
70
+ },
71
+ "LiftPegUpright-v1": {
72
+ "0.35": -8.968927557415074e-05,
73
+ "0.4": -0.0003124815588572361,
74
+ "0.45": 0.0005748239586551157
75
+ },
76
+ "PickCube-v1": {
77
+ "0.35": -0.00016158133519031948,
78
+ "0.4": 0.00011917837628256934,
79
+ "0.45": 0.00015914294978351952
80
+ },
81
+ "PullCube-v1": {
82
+ "0.35": -0.0005339817564724847,
83
+ "0.4": 0.0008820353746935294,
84
+ "0.45": -1.8324470630295964e-05
85
+ },
86
+ "PushCube-v1": {
87
+ "0.35": -0.0005731775627019931,
88
+ "0.4": 0.0009944043024089578,
89
+ "0.45": -6.036984813302393e-05
90
+ },
91
+ "StackCube-v1": {
92
+ "0.35": 0.00026410051799642675,
93
+ "0.4": -0.00037601939216213854,
94
+ "0.45": -9.074650458396062e-06
95
+ }
96
+ },
97
+ "candidate_type_bonus_components": false,
98
+ "candidate_oracle_rollouts": 0,
99
+ "candidate_oracle_unique_tolerance": 1e-06,
100
+ "candidate_oracle_success_rate": null,
101
+ "candidate_oracle_progress": null,
102
+ "candidate_oracle_score": null,
103
+ "candidate_oracle_regret": null,
104
+ "candidate_oracle_score_gain_over_selected": null,
105
+ "candidate_oracle_unique_count": null,
106
+ "candidate_oracle_improvement_rate": null,
107
+ "candidate_oracle_selected_branch_success_rate": null,
108
+ "candidate_oracle_selected_branch_progress": null,
109
+ "candidate_oracle_type_counts": {},
110
+ "candidate_oracle_best_branch_rank": null,
111
+ "candidate_oracle_best_branch_rank_counts": {},
112
+ "candidate_oracle_branch_success_rates": [],
113
+ "candidate_oracle_branch_progress": [],
114
+ "candidate_oracle_branch_score_gains_over_selected": [],
115
+ "selected_residual_scale_counts": {
116
+ "0.45": 276,
117
+ "0.35": 148,
118
+ "0.4": 151
119
+ },
120
+ "policy_rollout_success_rate": 0.3773913043478261,
121
+ "policy_rollout_progress": 0.5870517010922017,
122
+ "oracle_success_rate": 0.8573913043478261,
123
+ "action_mse_to_best": 0.49790886099085857,
124
+ "best_policy_val": {
125
+ "bc_loss": 0.3445456624031067,
126
+ "field_effect_loss": 0.009637858977334367,
127
+ "field_potential_loss": 0.28480341533819836,
128
+ "field_preference_loss": 0.5244891477955712,
129
+ "lattice_edges": 3833.3333333333335,
130
+ "progress_mae": 0.19576937291357252,
131
+ "proposal_loss": 0.0,
132
+ "rank_acc": 0.8192069795396593,
133
+ "rank_loss": 0.524484740363227,
134
+ "regret_mae": 0.42191632754272884,
135
+ "success_accuracy": 0.8693191342883639,
136
+ "total_loss": 0.261704473859734,
137
+ "transport_field_candidates": 0.0,
138
+ "transport_field_edges": 0.0,
139
+ "transport_field_loss": 0.0,
140
+ "transport_field_potential_loss": 0.0,
141
+ "transport_field_preference_loss": 0.0,
142
+ "transport_field_rank_acc": 0.0
143
+ },
144
+ "per_task": {
145
+ "LiftPegUpright-v1": {
146
+ "action_mse_to_best": 0.33983143654258285,
147
+ "expert_success_rate": 0.8865979381443299,
148
+ "num_groups": 97,
149
+ "oracle_success_rate": 0.9278350515463918,
150
+ "policy_expert_regret": 1.07819144181984,
151
+ "policy_oracle_regret": 1.1046069195282828,
152
+ "policy_rollout_progress": 0.5716054517890989,
153
+ "policy_rollout_success_rate": 0.2268041237113402,
154
+ "restore_max_error": 2.384185791015625e-07
155
+ },
156
+ "PickCube-v1": {
157
+ "action_mse_to_best": 0.4634422076006348,
158
+ "expert_success_rate": 0.9375,
159
+ "num_groups": 208,
160
+ "oracle_success_rate": 0.9471153846153846,
161
+ "policy_expert_regret": 0.9422957511135162,
162
+ "policy_oracle_regret": 0.949073279213805,
163
+ "policy_rollout_progress": 0.6223854628087093,
164
+ "policy_rollout_success_rate": 0.3605769230769231,
165
+ "restore_max_error": 2.384185791015625e-07
166
+ },
167
+ "PullCube-v1": {
168
+ "action_mse_to_best": 0.6120887859301134,
169
+ "expert_success_rate": 0.19480519480519481,
170
+ "num_groups": 77,
171
+ "oracle_success_rate": 0.36363636363636365,
172
+ "policy_expert_regret": 0.21602793449124733,
173
+ "policy_oracle_regret": 0.4055431050478251,
174
+ "policy_rollout_progress": 0.363066815831638,
175
+ "policy_rollout_success_rate": 0.2597402597402597,
176
+ "restore_max_error": 2.3096799850463867e-07
177
+ },
178
+ "PushCube-v1": {
179
+ "action_mse_to_best": 0.5037185225556505,
180
+ "expert_success_rate": 0.8617021276595744,
181
+ "num_groups": 94,
182
+ "oracle_success_rate": 0.9893617021276596,
183
+ "policy_expert_regret": 0.37636212211974124,
184
+ "policy_oracle_regret": 0.4349256801478406,
185
+ "policy_rollout_progress": 0.7794892613558059,
186
+ "policy_rollout_success_rate": 0.7659574468085106,
187
+ "restore_max_error": 2.384185791015625e-07
188
+ },
189
+ "StackCube-v1": {
190
+ "action_mse_to_best": 0.6308847364376892,
191
+ "expert_success_rate": 0.6767676767676768,
192
+ "num_groups": 99,
193
+ "oracle_success_rate": 0.8585858585858586,
194
+ "policy_expert_regret": 0.8869800036302721,
195
+ "policy_oracle_regret": 1.0460948816152533,
196
+ "policy_rollout_progress": 0.5194412894321211,
197
+ "policy_rollout_success_rate": 0.2828282828282828,
198
+ "restore_max_error": 2.384185791015625e-07
199
+ }
200
+ }
201
+ },
202
+ {
203
+ "seed": 1,
204
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_1/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025.json",
205
+ "num_groups": 575,
206
+ "selection_mode": "retrieval_residual",
207
+ "num_candidates": 48,
208
+ "candidate_sigma": 0.0,
209
+ "proposal_types": [],
210
+ "proposal_lattice_types": [],
211
+ "selection_margin": 0.0,
212
+ "prepend_policy_candidate": false,
213
+ "field_optim_steps": 0,
214
+ "field_optim_step_size": 0.0,
215
+ "field_optim_trust_radius": 0.0,
216
+ "field_optim_l2_penalty": 0.0,
217
+ "retrieval_neighbors": 6,
218
+ "retrieval_metric": "raw",
219
+ "retrieval_type_min_success": 0.0,
220
+ "retrieval_type_success_bonus_scale": 0.0,
221
+ "retrieval_residual_consensus_penalty_scale": 0.0,
222
+ "retrieval_residual_min_source_progress": 0.0,
223
+ "retrieval_residual_min_source_advantage": -1000000000.0,
224
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
225
+ "retrieval_residual_source_score_bonus_scale": 0.0,
226
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
227
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
228
+ "retrieval_residual_scale": 1.0,
229
+ "retrieval_residual_scales": [
230
+ 0.35,
231
+ 0.4,
232
+ 0.45
233
+ ],
234
+ "retrieval_residual_anchor": "expert",
235
+ "retrieval_residual_direction": "candidate_minus_anchor",
236
+ "retrieval_residual_reduce": "compose_mean_by_type",
237
+ "retrieval_residual_challenger_types": [],
238
+ "retrieval_residual_challenger_scales": [],
239
+ "retrieval_residual_challenger_margin": 0.0,
240
+ "retrieval_residual_challenger_type_margins": {},
241
+ "retrieval_residual_challenger_tasks": [],
242
+ "retrieval_residual_challenger_type_tasks": {},
243
+ "lattice_exclude_types": [
244
+ "residual_random_negative",
245
+ "residual_wrong_direction",
246
+ "residual_near_miss+residual_no_op",
247
+ "residual_no_op+residual_wrong_gripper"
248
+ ],
249
+ "lattice_exclude_type_tasks": {},
250
+ "candidate_type_bonuses": {},
251
+ "candidate_type_bonuses_by_task": {},
252
+ "residual_scale_bonuses_by_task": {
253
+ "*": {
254
+ "0.35": -9.674979510607974e-05,
255
+ "0.4": 0.0003106990953088251,
256
+ "0.45": -0.00016780893426614763
257
+ },
258
+ "LiftPegUpright-v1": {
259
+ "0.35": -8.968927557415074e-05,
260
+ "0.4": -0.0003124815588572361,
261
+ "0.45": 0.0005748239586551157
262
+ },
263
+ "PickCube-v1": {
264
+ "0.35": -0.00016158133519031948,
265
+ "0.4": 0.00011917837628256934,
266
+ "0.45": 0.00015914294978351952
267
+ },
268
+ "PullCube-v1": {
269
+ "0.35": -0.0005339817564724847,
270
+ "0.4": 0.0008820353746935294,
271
+ "0.45": -1.8324470630295964e-05
272
+ },
273
+ "PushCube-v1": {
274
+ "0.35": -0.0005731775627019931,
275
+ "0.4": 0.0009944043024089578,
276
+ "0.45": -6.036984813302393e-05
277
+ },
278
+ "StackCube-v1": {
279
+ "0.35": 0.00026410051799642675,
280
+ "0.4": -0.00037601939216213854,
281
+ "0.45": -9.074650458396062e-06
282
+ }
283
+ },
284
+ "candidate_type_bonus_components": false,
285
+ "candidate_oracle_rollouts": 0,
286
+ "candidate_oracle_unique_tolerance": 1e-06,
287
+ "candidate_oracle_success_rate": null,
288
+ "candidate_oracle_progress": null,
289
+ "candidate_oracle_score": null,
290
+ "candidate_oracle_regret": null,
291
+ "candidate_oracle_score_gain_over_selected": null,
292
+ "candidate_oracle_unique_count": null,
293
+ "candidate_oracle_improvement_rate": null,
294
+ "candidate_oracle_selected_branch_success_rate": null,
295
+ "candidate_oracle_selected_branch_progress": null,
296
+ "candidate_oracle_type_counts": {},
297
+ "candidate_oracle_best_branch_rank": null,
298
+ "candidate_oracle_best_branch_rank_counts": {},
299
+ "candidate_oracle_branch_success_rates": [],
300
+ "candidate_oracle_branch_progress": [],
301
+ "candidate_oracle_branch_score_gains_over_selected": [],
302
+ "selected_residual_scale_counts": {
303
+ "0.45": 315,
304
+ "0.4": 121,
305
+ "0.35": 139
306
+ },
307
+ "policy_rollout_success_rate": 0.3791304347826087,
308
+ "policy_rollout_progress": 0.5928320967908138,
309
+ "oracle_success_rate": 0.8695652173913043,
310
+ "action_mse_to_best": 0.5136578851476635,
311
+ "best_policy_val": {
312
+ "bc_loss": 0.31955528176493114,
313
+ "field_effect_loss": 0.010269693254182736,
314
+ "field_potential_loss": 0.30970407525698346,
315
+ "field_preference_loss": 0.5266202224625481,
316
+ "lattice_edges": 3833.3333333333335,
317
+ "progress_mae": 0.20275857630703184,
318
+ "proposal_loss": 0.0,
319
+ "rank_acc": 0.8124439285861121,
320
+ "rank_loss": 0.5266130566596985,
321
+ "regret_mae": 0.40278742214043933,
322
+ "success_accuracy": 0.8677195310592651,
323
+ "total_loss": 0.2706700298521254,
324
+ "transport_field_candidates": 0.0,
325
+ "transport_field_edges": 0.0,
326
+ "transport_field_loss": 0.0,
327
+ "transport_field_potential_loss": 0.0,
328
+ "transport_field_preference_loss": 0.0,
329
+ "transport_field_rank_acc": 0.0
330
+ },
331
+ "per_task": {
332
+ "LiftPegUpright-v1": {
333
+ "action_mse_to_best": 0.4174908100528461,
334
+ "expert_success_rate": 0.8584070796460177,
335
+ "num_groups": 113,
336
+ "oracle_success_rate": 0.9380530973451328,
337
+ "policy_expert_regret": 0.9912996379147588,
338
+ "policy_oracle_regret": 1.0707513929995816,
339
+ "policy_rollout_progress": 0.6289191406912509,
340
+ "policy_rollout_success_rate": 0.22123893805309736,
341
+ "restore_max_error": 2.384185791015625e-07
342
+ },
343
+ "PickCube-v1": {
344
+ "action_mse_to_best": 0.45537832047304383,
345
+ "expert_success_rate": 0.9402173913043478,
346
+ "num_groups": 184,
347
+ "oracle_success_rate": 0.9456521739130435,
348
+ "policy_expert_regret": 1.0321165590658379,
349
+ "policy_oracle_regret": 1.032140563540768,
350
+ "policy_rollout_progress": 0.5804765945067629,
351
+ "policy_rollout_success_rate": 0.32608695652173914,
352
+ "restore_max_error": 2.384185791015625e-07
353
+ },
354
+ "PullCube-v1": {
355
+ "action_mse_to_best": 0.6721879841857835,
356
+ "expert_success_rate": 0.25,
357
+ "num_groups": 76,
358
+ "oracle_success_rate": 0.40789473684210525,
359
+ "policy_expert_regret": 0.21536361078094496,
360
+ "policy_oracle_regret": 0.4112855422947752,
361
+ "policy_rollout_progress": 0.3406818090006709,
362
+ "policy_rollout_success_rate": 0.2236842105263158,
363
+ "restore_max_error": 2.384185791015625e-07
364
+ },
365
+ "PushCube-v1": {
366
+ "action_mse_to_best": 0.4296965829289711,
367
+ "expert_success_rate": 0.8198198198198198,
368
+ "num_groups": 111,
369
+ "oracle_success_rate": 1.0,
370
+ "policy_expert_regret": 0.26138654270687617,
371
+ "policy_oracle_regret": 0.27855410983970574,
372
+ "policy_rollout_progress": 0.8655900343044384,
373
+ "policy_rollout_success_rate": 0.8558558558558559,
374
+ "restore_max_error": 2.3795291781425476e-07
375
+ },
376
+ "StackCube-v1": {
377
+ "action_mse_to_best": 0.7209297137764784,
378
+ "expert_success_rate": 0.6923076923076923,
379
+ "num_groups": 91,
380
+ "oracle_success_rate": 0.8571428571428571,
381
+ "policy_expert_regret": 1.002974340548882,
382
+ "policy_oracle_regret": 1.1627966669889598,
383
+ "policy_rollout_progress": 0.45088558324745726,
384
+ "policy_rollout_success_rate": 0.23076923076923078,
385
+ "restore_max_error": 1.9371509552001953e-07
386
+ }
387
+ }
388
+ },
389
+ {
390
+ "seed": 2,
391
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_2/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025.json",
392
+ "num_groups": 575,
393
+ "selection_mode": "retrieval_residual",
394
+ "num_candidates": 48,
395
+ "candidate_sigma": 0.0,
396
+ "proposal_types": [],
397
+ "proposal_lattice_types": [],
398
+ "selection_margin": 0.0,
399
+ "prepend_policy_candidate": false,
400
+ "field_optim_steps": 0,
401
+ "field_optim_step_size": 0.0,
402
+ "field_optim_trust_radius": 0.0,
403
+ "field_optim_l2_penalty": 0.0,
404
+ "retrieval_neighbors": 6,
405
+ "retrieval_metric": "raw",
406
+ "retrieval_type_min_success": 0.0,
407
+ "retrieval_type_success_bonus_scale": 0.0,
408
+ "retrieval_residual_consensus_penalty_scale": 0.0,
409
+ "retrieval_residual_min_source_progress": 0.0,
410
+ "retrieval_residual_min_source_advantage": -1000000000.0,
411
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
412
+ "retrieval_residual_source_score_bonus_scale": 0.0,
413
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
414
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
415
+ "retrieval_residual_scale": 1.0,
416
+ "retrieval_residual_scales": [
417
+ 0.35,
418
+ 0.4,
419
+ 0.45
420
+ ],
421
+ "retrieval_residual_anchor": "expert",
422
+ "retrieval_residual_direction": "candidate_minus_anchor",
423
+ "retrieval_residual_reduce": "compose_mean_by_type",
424
+ "retrieval_residual_challenger_types": [],
425
+ "retrieval_residual_challenger_scales": [],
426
+ "retrieval_residual_challenger_margin": 0.0,
427
+ "retrieval_residual_challenger_type_margins": {},
428
+ "retrieval_residual_challenger_tasks": [],
429
+ "retrieval_residual_challenger_type_tasks": {},
430
+ "lattice_exclude_types": [
431
+ "residual_random_negative",
432
+ "residual_wrong_direction",
433
+ "residual_near_miss+residual_no_op",
434
+ "residual_no_op+residual_wrong_gripper"
435
+ ],
436
+ "lattice_exclude_type_tasks": {},
437
+ "candidate_type_bonuses": {},
438
+ "candidate_type_bonuses_by_task": {},
439
+ "residual_scale_bonuses_by_task": {
440
+ "*": {
441
+ "0.35": -9.674979510607974e-05,
442
+ "0.4": 0.0003106990953088251,
443
+ "0.45": -0.00016780893426614763
444
+ },
445
+ "LiftPegUpright-v1": {
446
+ "0.35": -8.968927557415074e-05,
447
+ "0.4": -0.0003124815588572361,
448
+ "0.45": 0.0005748239586551157
449
+ },
450
+ "PickCube-v1": {
451
+ "0.35": -0.00016158133519031948,
452
+ "0.4": 0.00011917837628256934,
453
+ "0.45": 0.00015914294978351952
454
+ },
455
+ "PullCube-v1": {
456
+ "0.35": -0.0005339817564724847,
457
+ "0.4": 0.0008820353746935294,
458
+ "0.45": -1.8324470630295964e-05
459
+ },
460
+ "PushCube-v1": {
461
+ "0.35": -0.0005731775627019931,
462
+ "0.4": 0.0009944043024089578,
463
+ "0.45": -6.036984813302393e-05
464
+ },
465
+ "StackCube-v1": {
466
+ "0.35": 0.00026410051799642675,
467
+ "0.4": -0.00037601939216213854,
468
+ "0.45": -9.074650458396062e-06
469
+ }
470
+ },
471
+ "candidate_type_bonus_components": false,
472
+ "candidate_oracle_rollouts": 0,
473
+ "candidate_oracle_unique_tolerance": 1e-06,
474
+ "candidate_oracle_success_rate": null,
475
+ "candidate_oracle_progress": null,
476
+ "candidate_oracle_score": null,
477
+ "candidate_oracle_regret": null,
478
+ "candidate_oracle_score_gain_over_selected": null,
479
+ "candidate_oracle_unique_count": null,
480
+ "candidate_oracle_improvement_rate": null,
481
+ "candidate_oracle_selected_branch_success_rate": null,
482
+ "candidate_oracle_selected_branch_progress": null,
483
+ "candidate_oracle_type_counts": {},
484
+ "candidate_oracle_best_branch_rank": null,
485
+ "candidate_oracle_best_branch_rank_counts": {},
486
+ "candidate_oracle_branch_success_rates": [],
487
+ "candidate_oracle_branch_progress": [],
488
+ "candidate_oracle_branch_score_gains_over_selected": [],
489
+ "selected_residual_scale_counts": {
490
+ "0.45": 287,
491
+ "0.4": 123,
492
+ "0.35": 165
493
+ },
494
+ "policy_rollout_success_rate": 0.40869565217391307,
495
+ "policy_rollout_progress": 0.6181983790514262,
496
+ "oracle_success_rate": 0.8765217391304347,
497
+ "action_mse_to_best": 0.5270935040372221,
498
+ "best_policy_val": {
499
+ "bc_loss": 0.2974168997671869,
500
+ "field_effect_loss": 0.009809228720971279,
501
+ "field_potential_loss": 0.29470594475666684,
502
+ "field_preference_loss": 0.5080636441707611,
503
+ "lattice_edges": 3833.3333333333335,
504
+ "progress_mae": 0.1944621221886741,
505
+ "proposal_loss": 0.0,
506
+ "rank_acc": 0.8252745270729065,
507
+ "rank_loss": 0.5080037944846683,
508
+ "regret_mae": 0.3989163488149643,
509
+ "success_accuracy": 0.870691223276986,
510
+ "total_loss": 0.2595750474267536,
511
+ "transport_field_candidates": 0.0,
512
+ "transport_field_edges": 0.0,
513
+ "transport_field_loss": 0.0,
514
+ "transport_field_potential_loss": 0.0,
515
+ "transport_field_preference_loss": 0.0,
516
+ "transport_field_rank_acc": 0.0
517
+ },
518
+ "per_task": {
519
+ "LiftPegUpright-v1": {
520
+ "action_mse_to_best": 0.44695540916291066,
521
+ "expert_success_rate": 0.8229166666666666,
522
+ "num_groups": 96,
523
+ "oracle_success_rate": 0.9270833333333334,
524
+ "policy_expert_regret": 0.8176391956706842,
525
+ "policy_oracle_regret": 0.9106746912002563,
526
+ "policy_rollout_progress": 0.6583504987259706,
527
+ "policy_rollout_success_rate": 0.3645833333333333,
528
+ "restore_max_error": 1.955777406692505e-07
529
+ },
530
+ "PickCube-v1": {
531
+ "action_mse_to_best": 0.4654190246053416,
532
+ "expert_success_rate": 0.9444444444444444,
533
+ "num_groups": 198,
534
+ "oracle_success_rate": 0.9595959595959596,
535
+ "policy_expert_regret": 0.9609503087051439,
536
+ "policy_oracle_regret": 0.9680337592338523,
537
+ "policy_rollout_progress": 0.638304318941076,
538
+ "policy_rollout_success_rate": 0.3484848484848485,
539
+ "restore_max_error": 2.384185791015625e-07
540
+ },
541
+ "PullCube-v1": {
542
+ "action_mse_to_best": 0.6707304260796971,
543
+ "expert_success_rate": 0.24444444444444444,
544
+ "num_groups": 90,
545
+ "oracle_success_rate": 0.4666666666666667,
546
+ "policy_expert_regret": 0.30656092938863566,
547
+ "policy_oracle_regret": 0.45043551944610144,
548
+ "policy_rollout_progress": 0.3851172113791108,
549
+ "policy_rollout_success_rate": 0.2777777777777778,
550
+ "restore_max_error": 2.384185791015625e-07
551
+ },
552
+ "PushCube-v1": {
553
+ "action_mse_to_best": 0.5131540888606911,
554
+ "expert_success_rate": 0.8514851485148515,
555
+ "num_groups": 101,
556
+ "oracle_success_rate": 1.0,
557
+ "policy_expert_regret": 0.2486148280377435,
558
+ "policy_oracle_regret": 0.28494056705201026,
559
+ "policy_rollout_progress": 0.8635742844331382,
560
+ "policy_rollout_success_rate": 0.8514851485148515,
561
+ "restore_max_error": 2.384185791015625e-07
562
+ },
563
+ "StackCube-v1": {
564
+ "action_mse_to_best": 0.620264192753368,
565
+ "expert_success_rate": 0.7666666666666667,
566
+ "num_groups": 90,
567
+ "oracle_success_rate": 0.9111111111111111,
568
+ "policy_expert_regret": 1.0516069930460719,
569
+ "policy_oracle_regret": 1.1881121820873684,
570
+ "policy_rollout_progress": 0.48885125749640995,
571
+ "policy_rollout_success_rate": 0.2222222222222222,
572
+ "restore_max_error": 2.2351741790771484e-07
573
+ }
574
+ }
575
+ }
576
+ ]
577
+ }
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025_summary.md ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # h=16 Best-Policy Checkpoint Rollout
2
+
3
+ Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
4
+ Objective: `transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1`
5
+ Result file: `policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025.json`
6
+ Completed seeds: 3
7
+ Baseline h=4 policy success: 29.67%
8
+ Baseline h=16 rank-checkpoint success: 29.74%
9
+
10
+ Mean success: 38.84% +/- 1.76%
11
+ Gain vs h=16 rank checkpoint: +9.10%
12
+ Mean progress: 59.94%
13
+ Mean action MSE to best: 0.513
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 37.74% | 58.71% | 85.74% | n/a | n/a | 0.498 |
18
+ | 1 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 37.91% | 59.28% | 86.96% | n/a | n/a | 0.514 |
19
+ | 2 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 40.87% | 61.82% | 87.65% | n/a | n/a | 0.527 |
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005_summary.json ADDED
@@ -0,0 +1,577 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
3
+ "objective": "transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1",
4
+ "out_name": "policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.38782608695652177,
9
+ "std_success": 0.01659024698116427,
10
+ "mean_progress": 0.5989091341704994,
11
+ "mean_action_mse_to_best": 0.5130268671047752,
12
+ "gain_vs_h4": 0.09112608695652175,
13
+ "gain_vs_h16_rank_checkpoint": 0.0904347826086957,
14
+ "rows": [
15
+ {
16
+ "seed": 0,
17
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_0/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 48,
21
+ "candidate_sigma": 0.0,
22
+ "proposal_types": [],
23
+ "proposal_lattice_types": [],
24
+ "selection_margin": 0.0,
25
+ "prepend_policy_candidate": false,
26
+ "field_optim_steps": 0,
27
+ "field_optim_step_size": 0.0,
28
+ "field_optim_trust_radius": 0.0,
29
+ "field_optim_l2_penalty": 0.0,
30
+ "retrieval_neighbors": 6,
31
+ "retrieval_metric": "raw",
32
+ "retrieval_type_min_success": 0.0,
33
+ "retrieval_type_success_bonus_scale": 0.0,
34
+ "retrieval_residual_consensus_penalty_scale": 0.0,
35
+ "retrieval_residual_min_source_progress": 0.0,
36
+ "retrieval_residual_min_source_advantage": -1000000000.0,
37
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
38
+ "retrieval_residual_source_score_bonus_scale": 0.0,
39
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
40
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
41
+ "retrieval_residual_scale": 1.0,
42
+ "retrieval_residual_scales": [
43
+ 0.35,
44
+ 0.4,
45
+ 0.45
46
+ ],
47
+ "retrieval_residual_anchor": "expert",
48
+ "retrieval_residual_direction": "candidate_minus_anchor",
49
+ "retrieval_residual_reduce": "compose_mean_by_type",
50
+ "retrieval_residual_challenger_types": [],
51
+ "retrieval_residual_challenger_scales": [],
52
+ "retrieval_residual_challenger_margin": 0.0,
53
+ "retrieval_residual_challenger_type_margins": {},
54
+ "retrieval_residual_challenger_tasks": [],
55
+ "retrieval_residual_challenger_type_tasks": {},
56
+ "lattice_exclude_types": [
57
+ "residual_random_negative",
58
+ "residual_wrong_direction",
59
+ "residual_near_miss+residual_no_op",
60
+ "residual_no_op+residual_wrong_gripper"
61
+ ],
62
+ "lattice_exclude_type_tasks": {},
63
+ "candidate_type_bonuses": {},
64
+ "candidate_type_bonuses_by_task": {},
65
+ "residual_scale_bonuses_by_task": {
66
+ "*": {
67
+ "0.35": -0.00019349959021215947,
68
+ "0.4": 0.0006213981906176502,
69
+ "0.45": -0.00033561786853229526
70
+ },
71
+ "LiftPegUpright-v1": {
72
+ "0.35": -0.00017937855114830148,
73
+ "0.4": -0.0006249631177144721,
74
+ "0.45": 0.0011496479173102314
75
+ },
76
+ "PickCube-v1": {
77
+ "0.35": -0.00032316267038063896,
78
+ "0.4": 0.00023835675256513868,
79
+ "0.45": 0.00031828589956703904
80
+ },
81
+ "PullCube-v1": {
82
+ "0.35": -0.0010679635129449694,
83
+ "0.4": 0.0017640707493870589,
84
+ "0.45": -3.664894126059193e-05
85
+ },
86
+ "PushCube-v1": {
87
+ "0.35": -0.0011463551254039862,
88
+ "0.4": 0.0019888086048179157,
89
+ "0.45": -0.00012073969626604787
90
+ },
91
+ "StackCube-v1": {
92
+ "0.35": 0.0005282010359928535,
93
+ "0.4": -0.0007520387843242771,
94
+ "0.45": -1.8149300916792124e-05
95
+ }
96
+ },
97
+ "candidate_type_bonus_components": false,
98
+ "candidate_oracle_rollouts": 0,
99
+ "candidate_oracle_unique_tolerance": 1e-06,
100
+ "candidate_oracle_success_rate": null,
101
+ "candidate_oracle_progress": null,
102
+ "candidate_oracle_score": null,
103
+ "candidate_oracle_regret": null,
104
+ "candidate_oracle_score_gain_over_selected": null,
105
+ "candidate_oracle_unique_count": null,
106
+ "candidate_oracle_improvement_rate": null,
107
+ "candidate_oracle_selected_branch_success_rate": null,
108
+ "candidate_oracle_selected_branch_progress": null,
109
+ "candidate_oracle_type_counts": {},
110
+ "candidate_oracle_best_branch_rank": null,
111
+ "candidate_oracle_best_branch_rank_counts": {},
112
+ "candidate_oracle_branch_success_rates": [],
113
+ "candidate_oracle_branch_progress": [],
114
+ "candidate_oracle_branch_score_gains_over_selected": [],
115
+ "selected_residual_scale_counts": {
116
+ "0.45": 276,
117
+ "0.35": 142,
118
+ "0.4": 157
119
+ },
120
+ "policy_rollout_success_rate": 0.3773913043478261,
121
+ "policy_rollout_progress": 0.5870847139799077,
122
+ "oracle_success_rate": 0.8573913043478261,
123
+ "action_mse_to_best": 0.4981027377445413,
124
+ "best_policy_val": {
125
+ "bc_loss": 0.3445456624031067,
126
+ "field_effect_loss": 0.009637858977334367,
127
+ "field_potential_loss": 0.28480341533819836,
128
+ "field_preference_loss": 0.5244891477955712,
129
+ "lattice_edges": 3833.3333333333335,
130
+ "progress_mae": 0.19576937291357252,
131
+ "proposal_loss": 0.0,
132
+ "rank_acc": 0.8192069795396593,
133
+ "rank_loss": 0.524484740363227,
134
+ "regret_mae": 0.42191632754272884,
135
+ "success_accuracy": 0.8693191342883639,
136
+ "total_loss": 0.261704473859734,
137
+ "transport_field_candidates": 0.0,
138
+ "transport_field_edges": 0.0,
139
+ "transport_field_loss": 0.0,
140
+ "transport_field_potential_loss": 0.0,
141
+ "transport_field_preference_loss": 0.0,
142
+ "transport_field_rank_acc": 0.0
143
+ },
144
+ "per_task": {
145
+ "LiftPegUpright-v1": {
146
+ "action_mse_to_best": 0.33983143654258285,
147
+ "expert_success_rate": 0.8865979381443299,
148
+ "num_groups": 97,
149
+ "oracle_success_rate": 0.9278350515463918,
150
+ "policy_expert_regret": 1.07819144181984,
151
+ "policy_oracle_regret": 1.1046069195282828,
152
+ "policy_rollout_progress": 0.5716054517890989,
153
+ "policy_rollout_success_rate": 0.2268041237113402,
154
+ "restore_max_error": 2.384185791015625e-07
155
+ },
156
+ "PickCube-v1": {
157
+ "action_mse_to_best": 0.46389446242784077,
158
+ "expert_success_rate": 0.9375,
159
+ "num_groups": 208,
160
+ "oracle_success_rate": 0.9471153846153846,
161
+ "policy_expert_regret": 0.9422860153180619,
162
+ "policy_oracle_regret": 0.9490635434183508,
163
+ "policy_rollout_progress": 0.6223951986041636,
164
+ "policy_rollout_success_rate": 0.3605769230769231,
165
+ "restore_max_error": 2.384185791015625e-07
166
+ },
167
+ "PullCube-v1": {
168
+ "action_mse_to_best": 0.6115246802181392,
169
+ "expert_success_rate": 0.19480519480519481,
170
+ "num_groups": 77,
171
+ "oracle_success_rate": 0.36363636363636365,
172
+ "policy_expert_regret": 0.21602793449124733,
173
+ "policy_oracle_regret": 0.40532328940734463,
174
+ "policy_rollout_progress": 0.36328704135081213,
175
+ "policy_rollout_success_rate": 0.2597402597402597,
176
+ "restore_max_error": 2.3096799850463867e-07
177
+ },
178
+ "PushCube-v1": {
179
+ "action_mse_to_best": 0.5037185225556505,
180
+ "expert_success_rate": 0.8617021276595744,
181
+ "num_groups": 94,
182
+ "oracle_success_rate": 0.9893617021276596,
183
+ "policy_expert_regret": 0.37636212211974124,
184
+ "policy_oracle_regret": 0.4349256801478406,
185
+ "policy_rollout_progress": 0.7794892613558059,
186
+ "policy_rollout_success_rate": 0.7659574468085106,
187
+ "restore_max_error": 2.384185791015625e-07
188
+ },
189
+ "StackCube-v1": {
190
+ "action_mse_to_best": 0.6314993452167872,
191
+ "expert_success_rate": 0.6767676767676768,
192
+ "num_groups": 99,
193
+ "oracle_success_rate": 0.8585858585858586,
194
+ "policy_expert_regret": 0.8869800036302721,
195
+ "policy_oracle_regret": 1.0460948816152533,
196
+ "policy_rollout_progress": 0.5194412894321211,
197
+ "policy_rollout_success_rate": 0.2828282828282828,
198
+ "restore_max_error": 2.384185791015625e-07
199
+ }
200
+ }
201
+ },
202
+ {
203
+ "seed": 1,
204
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_1/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005.json",
205
+ "num_groups": 575,
206
+ "selection_mode": "retrieval_residual",
207
+ "num_candidates": 48,
208
+ "candidate_sigma": 0.0,
209
+ "proposal_types": [],
210
+ "proposal_lattice_types": [],
211
+ "selection_margin": 0.0,
212
+ "prepend_policy_candidate": false,
213
+ "field_optim_steps": 0,
214
+ "field_optim_step_size": 0.0,
215
+ "field_optim_trust_radius": 0.0,
216
+ "field_optim_l2_penalty": 0.0,
217
+ "retrieval_neighbors": 6,
218
+ "retrieval_metric": "raw",
219
+ "retrieval_type_min_success": 0.0,
220
+ "retrieval_type_success_bonus_scale": 0.0,
221
+ "retrieval_residual_consensus_penalty_scale": 0.0,
222
+ "retrieval_residual_min_source_progress": 0.0,
223
+ "retrieval_residual_min_source_advantage": -1000000000.0,
224
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
225
+ "retrieval_residual_source_score_bonus_scale": 0.0,
226
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
227
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
228
+ "retrieval_residual_scale": 1.0,
229
+ "retrieval_residual_scales": [
230
+ 0.35,
231
+ 0.4,
232
+ 0.45
233
+ ],
234
+ "retrieval_residual_anchor": "expert",
235
+ "retrieval_residual_direction": "candidate_minus_anchor",
236
+ "retrieval_residual_reduce": "compose_mean_by_type",
237
+ "retrieval_residual_challenger_types": [],
238
+ "retrieval_residual_challenger_scales": [],
239
+ "retrieval_residual_challenger_margin": 0.0,
240
+ "retrieval_residual_challenger_type_margins": {},
241
+ "retrieval_residual_challenger_tasks": [],
242
+ "retrieval_residual_challenger_type_tasks": {},
243
+ "lattice_exclude_types": [
244
+ "residual_random_negative",
245
+ "residual_wrong_direction",
246
+ "residual_near_miss+residual_no_op",
247
+ "residual_no_op+residual_wrong_gripper"
248
+ ],
249
+ "lattice_exclude_type_tasks": {},
250
+ "candidate_type_bonuses": {},
251
+ "candidate_type_bonuses_by_task": {},
252
+ "residual_scale_bonuses_by_task": {
253
+ "*": {
254
+ "0.35": -0.00019349959021215947,
255
+ "0.4": 0.0006213981906176502,
256
+ "0.45": -0.00033561786853229526
257
+ },
258
+ "LiftPegUpright-v1": {
259
+ "0.35": -0.00017937855114830148,
260
+ "0.4": -0.0006249631177144721,
261
+ "0.45": 0.0011496479173102314
262
+ },
263
+ "PickCube-v1": {
264
+ "0.35": -0.00032316267038063896,
265
+ "0.4": 0.00023835675256513868,
266
+ "0.45": 0.00031828589956703904
267
+ },
268
+ "PullCube-v1": {
269
+ "0.35": -0.0010679635129449694,
270
+ "0.4": 0.0017640707493870589,
271
+ "0.45": -3.664894126059193e-05
272
+ },
273
+ "PushCube-v1": {
274
+ "0.35": -0.0011463551254039862,
275
+ "0.4": 0.0019888086048179157,
276
+ "0.45": -0.00012073969626604787
277
+ },
278
+ "StackCube-v1": {
279
+ "0.35": 0.0005282010359928535,
280
+ "0.4": -0.0007520387843242771,
281
+ "0.45": -1.8149300916792124e-05
282
+ }
283
+ },
284
+ "candidate_type_bonus_components": false,
285
+ "candidate_oracle_rollouts": 0,
286
+ "candidate_oracle_unique_tolerance": 1e-06,
287
+ "candidate_oracle_success_rate": null,
288
+ "candidate_oracle_progress": null,
289
+ "candidate_oracle_score": null,
290
+ "candidate_oracle_regret": null,
291
+ "candidate_oracle_score_gain_over_selected": null,
292
+ "candidate_oracle_unique_count": null,
293
+ "candidate_oracle_improvement_rate": null,
294
+ "candidate_oracle_selected_branch_success_rate": null,
295
+ "candidate_oracle_selected_branch_progress": null,
296
+ "candidate_oracle_type_counts": {},
297
+ "candidate_oracle_best_branch_rank": null,
298
+ "candidate_oracle_best_branch_rank_counts": {},
299
+ "candidate_oracle_branch_success_rates": [],
300
+ "candidate_oracle_branch_progress": [],
301
+ "candidate_oracle_branch_score_gains_over_selected": [],
302
+ "selected_residual_scale_counts": {
303
+ "0.45": 314,
304
+ "0.35": 132,
305
+ "0.4": 129
306
+ },
307
+ "policy_rollout_success_rate": 0.3791304347826087,
308
+ "policy_rollout_progress": 0.5928355075111208,
309
+ "oracle_success_rate": 0.8695652173913043,
310
+ "action_mse_to_best": 0.5134497352065923,
311
+ "best_policy_val": {
312
+ "bc_loss": 0.31955528176493114,
313
+ "field_effect_loss": 0.010269693254182736,
314
+ "field_potential_loss": 0.30970407525698346,
315
+ "field_preference_loss": 0.5266202224625481,
316
+ "lattice_edges": 3833.3333333333335,
317
+ "progress_mae": 0.20275857630703184,
318
+ "proposal_loss": 0.0,
319
+ "rank_acc": 0.8124439285861121,
320
+ "rank_loss": 0.5266130566596985,
321
+ "regret_mae": 0.40278742214043933,
322
+ "success_accuracy": 0.8677195310592651,
323
+ "total_loss": 0.2706700298521254,
324
+ "transport_field_candidates": 0.0,
325
+ "transport_field_edges": 0.0,
326
+ "transport_field_loss": 0.0,
327
+ "transport_field_potential_loss": 0.0,
328
+ "transport_field_preference_loss": 0.0,
329
+ "transport_field_rank_acc": 0.0
330
+ },
331
+ "per_task": {
332
+ "LiftPegUpright-v1": {
333
+ "action_mse_to_best": 0.41769555068896275,
334
+ "expert_success_rate": 0.8584070796460177,
335
+ "num_groups": 113,
336
+ "oracle_success_rate": 0.9380530973451328,
337
+ "policy_expert_regret": 0.9912851655905226,
338
+ "policy_oracle_regret": 1.0707369206753452,
339
+ "policy_rollout_progress": 0.6289336130154871,
340
+ "policy_rollout_success_rate": 0.22123893805309736,
341
+ "restore_max_error": 2.384185791015625e-07
342
+ },
343
+ "PickCube-v1": {
344
+ "action_mse_to_best": 0.45538117469328904,
345
+ "expert_success_rate": 0.9402173913043478,
346
+ "num_groups": 184,
347
+ "oracle_success_rate": 0.9456521739130435,
348
+ "policy_expert_regret": 1.0321345428237692,
349
+ "policy_oracle_regret": 1.0321585472986994,
350
+ "policy_rollout_progress": 0.5804586107488316,
351
+ "policy_rollout_success_rate": 0.32608695652173914,
352
+ "restore_max_error": 2.384185791015625e-07
353
+ },
354
+ "PullCube-v1": {
355
+ "action_mse_to_best": 0.6727483035310319,
356
+ "expert_success_rate": 0.25,
357
+ "num_groups": 76,
358
+ "oracle_success_rate": 0.40789473684210525,
359
+ "policy_expert_regret": 0.21536361078094496,
360
+ "policy_oracle_regret": 0.4112207699861181,
361
+ "policy_rollout_progress": 0.34074658130932794,
362
+ "policy_rollout_success_rate": 0.2236842105263158,
363
+ "restore_max_error": 2.384185791015625e-07
364
+ },
365
+ "PushCube-v1": {
366
+ "action_mse_to_best": 0.42832537565950873,
367
+ "expert_success_rate": 0.8198198198198198,
368
+ "num_groups": 111,
369
+ "oracle_success_rate": 1.0,
370
+ "policy_expert_regret": 0.26135670601784644,
371
+ "policy_oracle_regret": 0.278524273150676,
372
+ "policy_rollout_progress": 0.8656198709934682,
373
+ "policy_rollout_success_rate": 0.8558558558558559,
374
+ "restore_max_error": 2.3795291781425476e-07
375
+ },
376
+ "StackCube-v1": {
377
+ "action_mse_to_best": 0.7205590835803157,
378
+ "expert_success_rate": 0.6923076923076923,
379
+ "num_groups": 91,
380
+ "oracle_success_rate": 0.8571428571428571,
381
+ "policy_expert_regret": 1.003024887416389,
382
+ "policy_oracle_regret": 1.1628472138564665,
383
+ "policy_rollout_progress": 0.45083503637995037,
384
+ "policy_rollout_success_rate": 0.23076923076923078,
385
+ "restore_max_error": 1.9371509552001953e-07
386
+ }
387
+ }
388
+ },
389
+ {
390
+ "seed": 2,
391
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_2/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005.json",
392
+ "num_groups": 575,
393
+ "selection_mode": "retrieval_residual",
394
+ "num_candidates": 48,
395
+ "candidate_sigma": 0.0,
396
+ "proposal_types": [],
397
+ "proposal_lattice_types": [],
398
+ "selection_margin": 0.0,
399
+ "prepend_policy_candidate": false,
400
+ "field_optim_steps": 0,
401
+ "field_optim_step_size": 0.0,
402
+ "field_optim_trust_radius": 0.0,
403
+ "field_optim_l2_penalty": 0.0,
404
+ "retrieval_neighbors": 6,
405
+ "retrieval_metric": "raw",
406
+ "retrieval_type_min_success": 0.0,
407
+ "retrieval_type_success_bonus_scale": 0.0,
408
+ "retrieval_residual_consensus_penalty_scale": 0.0,
409
+ "retrieval_residual_min_source_progress": 0.0,
410
+ "retrieval_residual_min_source_advantage": -1000000000.0,
411
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
412
+ "retrieval_residual_source_score_bonus_scale": 0.0,
413
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
414
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
415
+ "retrieval_residual_scale": 1.0,
416
+ "retrieval_residual_scales": [
417
+ 0.35,
418
+ 0.4,
419
+ 0.45
420
+ ],
421
+ "retrieval_residual_anchor": "expert",
422
+ "retrieval_residual_direction": "candidate_minus_anchor",
423
+ "retrieval_residual_reduce": "compose_mean_by_type",
424
+ "retrieval_residual_challenger_types": [],
425
+ "retrieval_residual_challenger_scales": [],
426
+ "retrieval_residual_challenger_margin": 0.0,
427
+ "retrieval_residual_challenger_type_margins": {},
428
+ "retrieval_residual_challenger_tasks": [],
429
+ "retrieval_residual_challenger_type_tasks": {},
430
+ "lattice_exclude_types": [
431
+ "residual_random_negative",
432
+ "residual_wrong_direction",
433
+ "residual_near_miss+residual_no_op",
434
+ "residual_no_op+residual_wrong_gripper"
435
+ ],
436
+ "lattice_exclude_type_tasks": {},
437
+ "candidate_type_bonuses": {},
438
+ "candidate_type_bonuses_by_task": {},
439
+ "residual_scale_bonuses_by_task": {
440
+ "*": {
441
+ "0.35": -0.00019349959021215947,
442
+ "0.4": 0.0006213981906176502,
443
+ "0.45": -0.00033561786853229526
444
+ },
445
+ "LiftPegUpright-v1": {
446
+ "0.35": -0.00017937855114830148,
447
+ "0.4": -0.0006249631177144721,
448
+ "0.45": 0.0011496479173102314
449
+ },
450
+ "PickCube-v1": {
451
+ "0.35": -0.00032316267038063896,
452
+ "0.4": 0.00023835675256513868,
453
+ "0.45": 0.00031828589956703904
454
+ },
455
+ "PullCube-v1": {
456
+ "0.35": -0.0010679635129449694,
457
+ "0.4": 0.0017640707493870589,
458
+ "0.45": -3.664894126059193e-05
459
+ },
460
+ "PushCube-v1": {
461
+ "0.35": -0.0011463551254039862,
462
+ "0.4": 0.0019888086048179157,
463
+ "0.45": -0.00012073969626604787
464
+ },
465
+ "StackCube-v1": {
466
+ "0.35": 0.0005282010359928535,
467
+ "0.4": -0.0007520387843242771,
468
+ "0.45": -1.8149300916792124e-05
469
+ }
470
+ },
471
+ "candidate_type_bonus_components": false,
472
+ "candidate_oracle_rollouts": 0,
473
+ "candidate_oracle_unique_tolerance": 1e-06,
474
+ "candidate_oracle_success_rate": null,
475
+ "candidate_oracle_progress": null,
476
+ "candidate_oracle_score": null,
477
+ "candidate_oracle_regret": null,
478
+ "candidate_oracle_score_gain_over_selected": null,
479
+ "candidate_oracle_unique_count": null,
480
+ "candidate_oracle_improvement_rate": null,
481
+ "candidate_oracle_selected_branch_success_rate": null,
482
+ "candidate_oracle_selected_branch_progress": null,
483
+ "candidate_oracle_type_counts": {},
484
+ "candidate_oracle_best_branch_rank": null,
485
+ "candidate_oracle_best_branch_rank_counts": {},
486
+ "candidate_oracle_branch_success_rates": [],
487
+ "candidate_oracle_branch_progress": [],
488
+ "candidate_oracle_branch_score_gains_over_selected": [],
489
+ "selected_residual_scale_counts": {
490
+ "0.45": 283,
491
+ "0.4": 133,
492
+ "0.35": 159
493
+ },
494
+ "policy_rollout_success_rate": 0.40695652173913044,
495
+ "policy_rollout_progress": 0.6168071810204697,
496
+ "oracle_success_rate": 0.8765217391304347,
497
+ "action_mse_to_best": 0.5275281283631921,
498
+ "best_policy_val": {
499
+ "bc_loss": 0.2974168997671869,
500
+ "field_effect_loss": 0.009809228720971279,
501
+ "field_potential_loss": 0.29470594475666684,
502
+ "field_preference_loss": 0.5080636441707611,
503
+ "lattice_edges": 3833.3333333333335,
504
+ "progress_mae": 0.1944621221886741,
505
+ "proposal_loss": 0.0,
506
+ "rank_acc": 0.8252745270729065,
507
+ "rank_loss": 0.5080037944846683,
508
+ "regret_mae": 0.3989163488149643,
509
+ "success_accuracy": 0.870691223276986,
510
+ "total_loss": 0.2595750474267536,
511
+ "transport_field_candidates": 0.0,
512
+ "transport_field_edges": 0.0,
513
+ "transport_field_loss": 0.0,
514
+ "transport_field_potential_loss": 0.0,
515
+ "transport_field_preference_loss": 0.0,
516
+ "transport_field_rank_acc": 0.0
517
+ },
518
+ "per_task": {
519
+ "LiftPegUpright-v1": {
520
+ "action_mse_to_best": 0.44695540916291066,
521
+ "expert_success_rate": 0.8229166666666666,
522
+ "num_groups": 96,
523
+ "oracle_success_rate": 0.9270833333333334,
524
+ "policy_expert_regret": 0.8176391956706842,
525
+ "policy_oracle_regret": 0.9106746912002563,
526
+ "policy_rollout_progress": 0.6583504987259706,
527
+ "policy_rollout_success_rate": 0.3645833333333333,
528
+ "restore_max_error": 1.955777406692505e-07
529
+ },
530
+ "PickCube-v1": {
531
+ "action_mse_to_best": 0.4659398831238951,
532
+ "expert_success_rate": 0.9444444444444444,
533
+ "num_groups": 198,
534
+ "oracle_success_rate": 0.9595959595959596,
535
+ "policy_expert_regret": 0.9609504700591317,
536
+ "policy_oracle_regret": 0.9680339205878401,
537
+ "policy_rollout_progress": 0.6383041575870881,
538
+ "policy_rollout_success_rate": 0.3484848484848485,
539
+ "restore_max_error": 2.384185791015625e-07
540
+ },
541
+ "PullCube-v1": {
542
+ "action_mse_to_best": 0.6718037481109301,
543
+ "expert_success_rate": 0.24444444444444444,
544
+ "num_groups": 90,
545
+ "oracle_success_rate": 0.4666666666666667,
546
+ "policy_expert_regret": 0.30672622924256654,
547
+ "policy_oracle_regret": 0.4515341889630589,
548
+ "policy_rollout_progress": 0.3762278470624652,
549
+ "policy_rollout_success_rate": 0.26666666666666666,
550
+ "restore_max_error": 2.384185791015625e-07
551
+ },
552
+ "PushCube-v1": {
553
+ "action_mse_to_best": 0.5141648006911325,
554
+ "expert_success_rate": 0.8514851485148515,
555
+ "num_groups": 101,
556
+ "oracle_success_rate": 1.0,
557
+ "policy_expert_regret": 0.2486148280377435,
558
+ "policy_oracle_regret": 0.28494056705201026,
559
+ "policy_rollout_progress": 0.8635742844331382,
560
+ "policy_rollout_success_rate": 0.8514851485148515,
561
+ "restore_max_error": 2.384185791015625e-07
562
+ },
563
+ "StackCube-v1": {
564
+ "action_mse_to_best": 0.6196875052319633,
565
+ "expert_success_rate": 0.7666666666666667,
566
+ "num_groups": 90,
567
+ "oracle_success_rate": 0.9111111111111111,
568
+ "policy_expert_regret": 1.0516054833928743,
569
+ "policy_oracle_regret": 1.188110672434171,
570
+ "policy_rollout_progress": 0.48885276714960735,
571
+ "policy_rollout_success_rate": 0.2222222222222222,
572
+ "restore_max_error": 2.2351741790771484e-07
573
+ }
574
+ }
575
+ }
576
+ ]
577
+ }
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005_summary.md ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # h=16 Best-Policy Checkpoint Rollout
2
+
3
+ Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
4
+ Objective: `transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1`
5
+ Result file: `policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005.json`
6
+ Completed seeds: 3
7
+ Baseline h=4 policy success: 29.67%
8
+ Baseline h=16 rank-checkpoint success: 29.74%
9
+
10
+ Mean success: 38.78% +/- 1.66%
11
+ Gain vs h=16 rank checkpoint: +9.04%
12
+ Mean progress: 59.89%
13
+ Mean action MSE to best: 0.513
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 37.74% | 58.71% | 85.74% | n/a | n/a | 0.498 |
18
+ | 1 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 37.91% | 59.28% | 86.96% | n/a | n/a | 0.513 |
19
+ | 2 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 40.70% | 61.68% | 87.65% | n/a | n/a | 0.528 |
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010_summary.json ADDED
@@ -0,0 +1,577 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
3
+ "objective": "transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1",
4
+ "out_name": "policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.38666666666666666,
9
+ "std_success": 0.016159257814221853,
10
+ "mean_progress": 0.5982465150498826,
11
+ "mean_action_mse_to_best": 0.513236562266028,
12
+ "gain_vs_h4": 0.08996666666666664,
13
+ "gain_vs_h16_rank_checkpoint": 0.08927536231884059,
14
+ "rows": [
15
+ {
16
+ "seed": 0,
17
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_0/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 48,
21
+ "candidate_sigma": 0.0,
22
+ "proposal_types": [],
23
+ "proposal_lattice_types": [],
24
+ "selection_margin": 0.0,
25
+ "prepend_policy_candidate": false,
26
+ "field_optim_steps": 0,
27
+ "field_optim_step_size": 0.0,
28
+ "field_optim_trust_radius": 0.0,
29
+ "field_optim_l2_penalty": 0.0,
30
+ "retrieval_neighbors": 6,
31
+ "retrieval_metric": "raw",
32
+ "retrieval_type_min_success": 0.0,
33
+ "retrieval_type_success_bonus_scale": 0.0,
34
+ "retrieval_residual_consensus_penalty_scale": 0.0,
35
+ "retrieval_residual_min_source_progress": 0.0,
36
+ "retrieval_residual_min_source_advantage": -1000000000.0,
37
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
38
+ "retrieval_residual_source_score_bonus_scale": 0.0,
39
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
40
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
41
+ "retrieval_residual_scale": 1.0,
42
+ "retrieval_residual_scales": [
43
+ 0.35,
44
+ 0.4,
45
+ 0.45
46
+ ],
47
+ "retrieval_residual_anchor": "expert",
48
+ "retrieval_residual_direction": "candidate_minus_anchor",
49
+ "retrieval_residual_reduce": "compose_mean_by_type",
50
+ "retrieval_residual_challenger_types": [],
51
+ "retrieval_residual_challenger_scales": [],
52
+ "retrieval_residual_challenger_margin": 0.0,
53
+ "retrieval_residual_challenger_type_margins": {},
54
+ "retrieval_residual_challenger_tasks": [],
55
+ "retrieval_residual_challenger_type_tasks": {},
56
+ "lattice_exclude_types": [
57
+ "residual_random_negative",
58
+ "residual_wrong_direction",
59
+ "residual_near_miss+residual_no_op",
60
+ "residual_no_op+residual_wrong_gripper"
61
+ ],
62
+ "lattice_exclude_type_tasks": {},
63
+ "candidate_type_bonuses": {},
64
+ "candidate_type_bonuses_by_task": {},
65
+ "residual_scale_bonuses_by_task": {
66
+ "*": {
67
+ "0.35": -0.00038699918042431894,
68
+ "0.4": 0.0012427963812353005,
69
+ "0.45": -0.0006712357370645905
70
+ },
71
+ "LiftPegUpright-v1": {
72
+ "0.35": -0.00035875710229660296,
73
+ "0.4": -0.0012499262354289443,
74
+ "0.45": 0.002299295834620463
75
+ },
76
+ "PickCube-v1": {
77
+ "0.35": -0.0006463253407612779,
78
+ "0.4": 0.00047671350513027737,
79
+ "0.45": 0.0006365717991340781
80
+ },
81
+ "PullCube-v1": {
82
+ "0.35": -0.0021359270258899388,
83
+ "0.4": 0.0035281414987741178,
84
+ "0.45": -7.329788252118386e-05
85
+ },
86
+ "PushCube-v1": {
87
+ "0.35": -0.0022927102508079724,
88
+ "0.4": 0.003977617209635831,
89
+ "0.45": -0.00024147939253209573
90
+ },
91
+ "StackCube-v1": {
92
+ "0.35": 0.001056402071985707,
93
+ "0.4": -0.0015040775686485542,
94
+ "0.45": -3.629860183358425e-05
95
+ }
96
+ },
97
+ "candidate_type_bonus_components": false,
98
+ "candidate_oracle_rollouts": 0,
99
+ "candidate_oracle_unique_tolerance": 1e-06,
100
+ "candidate_oracle_success_rate": null,
101
+ "candidate_oracle_progress": null,
102
+ "candidate_oracle_score": null,
103
+ "candidate_oracle_regret": null,
104
+ "candidate_oracle_score_gain_over_selected": null,
105
+ "candidate_oracle_unique_count": null,
106
+ "candidate_oracle_improvement_rate": null,
107
+ "candidate_oracle_selected_branch_success_rate": null,
108
+ "candidate_oracle_selected_branch_progress": null,
109
+ "candidate_oracle_type_counts": {},
110
+ "candidate_oracle_best_branch_rank": null,
111
+ "candidate_oracle_best_branch_rank_counts": {},
112
+ "candidate_oracle_branch_success_rates": [],
113
+ "candidate_oracle_branch_progress": [],
114
+ "candidate_oracle_branch_score_gains_over_selected": [],
115
+ "selected_residual_scale_counts": {
116
+ "0.45": 278,
117
+ "0.35": 128,
118
+ "0.4": 169
119
+ },
120
+ "policy_rollout_success_rate": 0.37565217391304345,
121
+ "policy_rollout_progress": 0.5863374990386807,
122
+ "oracle_success_rate": 0.8573913043478261,
123
+ "action_mse_to_best": 0.49939822750409013,
124
+ "best_policy_val": {
125
+ "bc_loss": 0.3445456624031067,
126
+ "field_effect_loss": 0.009637858977334367,
127
+ "field_potential_loss": 0.28480341533819836,
128
+ "field_preference_loss": 0.5244891477955712,
129
+ "lattice_edges": 3833.3333333333335,
130
+ "progress_mae": 0.19576937291357252,
131
+ "proposal_loss": 0.0,
132
+ "rank_acc": 0.8192069795396593,
133
+ "rank_loss": 0.524484740363227,
134
+ "regret_mae": 0.42191632754272884,
135
+ "success_accuracy": 0.8693191342883639,
136
+ "total_loss": 0.261704473859734,
137
+ "transport_field_candidates": 0.0,
138
+ "transport_field_edges": 0.0,
139
+ "transport_field_loss": 0.0,
140
+ "transport_field_potential_loss": 0.0,
141
+ "transport_field_preference_loss": 0.0,
142
+ "transport_field_rank_acc": 0.0
143
+ },
144
+ "per_task": {
145
+ "LiftPegUpright-v1": {
146
+ "action_mse_to_best": 0.33983143654258285,
147
+ "expert_success_rate": 0.8865979381443299,
148
+ "num_groups": 97,
149
+ "oracle_success_rate": 0.9278350515463918,
150
+ "policy_expert_regret": 1.07819144181984,
151
+ "policy_oracle_regret": 1.1046069195282828,
152
+ "policy_rollout_progress": 0.5716054517890989,
153
+ "policy_rollout_success_rate": 0.2268041237113402,
154
+ "restore_max_error": 2.384185791015625e-07
155
+ },
156
+ "PickCube-v1": {
157
+ "action_mse_to_best": 0.46664860253580487,
158
+ "expert_success_rate": 0.9375,
159
+ "num_groups": 208,
160
+ "oracle_success_rate": 0.9471153846153846,
161
+ "policy_expert_regret": 0.9491616923881409,
162
+ "policy_oracle_regret": 0.9559392204884296,
163
+ "policy_rollout_progress": 0.620327213841777,
164
+ "policy_rollout_success_rate": 0.3557692307692308,
165
+ "restore_max_error": 2.384185791015625e-07
166
+ },
167
+ "PullCube-v1": {
168
+ "action_mse_to_best": 0.6109810099973307,
169
+ "expert_success_rate": 0.19480519480519481,
170
+ "num_groups": 77,
171
+ "oracle_success_rate": 0.36363636363636365,
172
+ "policy_expert_regret": 0.2160475609412828,
173
+ "policy_oracle_regret": 0.40534284928595865,
174
+ "policy_rollout_progress": 0.3632675172736892,
175
+ "policy_rollout_success_rate": 0.2597402597402597,
176
+ "restore_max_error": 2.3096799850463867e-07
177
+ },
178
+ "PushCube-v1": {
179
+ "action_mse_to_best": 0.5050888739209226,
180
+ "expert_success_rate": 0.8617021276595744,
181
+ "num_groups": 94,
182
+ "oracle_success_rate": 0.9893617021276596,
183
+ "policy_expert_regret": 0.37636212211974124,
184
+ "policy_oracle_regret": 0.4349256801478406,
185
+ "policy_rollout_progress": 0.7794892613558059,
186
+ "policy_rollout_success_rate": 0.7659574468085106,
187
+ "restore_max_error": 2.384185791015625e-07
188
+ },
189
+ "StackCube-v1": {
190
+ "action_mse_to_best": 0.6323588911556837,
191
+ "expert_success_rate": 0.6767676767676768,
192
+ "num_groups": 99,
193
+ "oracle_success_rate": 0.8585858585858586,
194
+ "policy_expert_regret": 0.8869598461220963,
195
+ "policy_oracle_regret": 1.0460747241070776,
196
+ "policy_rollout_progress": 0.5194614469402968,
197
+ "policy_rollout_success_rate": 0.2828282828282828,
198
+ "restore_max_error": 2.384185791015625e-07
199
+ }
200
+ }
201
+ },
202
+ {
203
+ "seed": 1,
204
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_1/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010.json",
205
+ "num_groups": 575,
206
+ "selection_mode": "retrieval_residual",
207
+ "num_candidates": 48,
208
+ "candidate_sigma": 0.0,
209
+ "proposal_types": [],
210
+ "proposal_lattice_types": [],
211
+ "selection_margin": 0.0,
212
+ "prepend_policy_candidate": false,
213
+ "field_optim_steps": 0,
214
+ "field_optim_step_size": 0.0,
215
+ "field_optim_trust_radius": 0.0,
216
+ "field_optim_l2_penalty": 0.0,
217
+ "retrieval_neighbors": 6,
218
+ "retrieval_metric": "raw",
219
+ "retrieval_type_min_success": 0.0,
220
+ "retrieval_type_success_bonus_scale": 0.0,
221
+ "retrieval_residual_consensus_penalty_scale": 0.0,
222
+ "retrieval_residual_min_source_progress": 0.0,
223
+ "retrieval_residual_min_source_advantage": -1000000000.0,
224
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
225
+ "retrieval_residual_source_score_bonus_scale": 0.0,
226
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
227
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
228
+ "retrieval_residual_scale": 1.0,
229
+ "retrieval_residual_scales": [
230
+ 0.35,
231
+ 0.4,
232
+ 0.45
233
+ ],
234
+ "retrieval_residual_anchor": "expert",
235
+ "retrieval_residual_direction": "candidate_minus_anchor",
236
+ "retrieval_residual_reduce": "compose_mean_by_type",
237
+ "retrieval_residual_challenger_types": [],
238
+ "retrieval_residual_challenger_scales": [],
239
+ "retrieval_residual_challenger_margin": 0.0,
240
+ "retrieval_residual_challenger_type_margins": {},
241
+ "retrieval_residual_challenger_tasks": [],
242
+ "retrieval_residual_challenger_type_tasks": {},
243
+ "lattice_exclude_types": [
244
+ "residual_random_negative",
245
+ "residual_wrong_direction",
246
+ "residual_near_miss+residual_no_op",
247
+ "residual_no_op+residual_wrong_gripper"
248
+ ],
249
+ "lattice_exclude_type_tasks": {},
250
+ "candidate_type_bonuses": {},
251
+ "candidate_type_bonuses_by_task": {},
252
+ "residual_scale_bonuses_by_task": {
253
+ "*": {
254
+ "0.35": -0.00038699918042431894,
255
+ "0.4": 0.0012427963812353005,
256
+ "0.45": -0.0006712357370645905
257
+ },
258
+ "LiftPegUpright-v1": {
259
+ "0.35": -0.00035875710229660296,
260
+ "0.4": -0.0012499262354289443,
261
+ "0.45": 0.002299295834620463
262
+ },
263
+ "PickCube-v1": {
264
+ "0.35": -0.0006463253407612779,
265
+ "0.4": 0.00047671350513027737,
266
+ "0.45": 0.0006365717991340781
267
+ },
268
+ "PullCube-v1": {
269
+ "0.35": -0.0021359270258899388,
270
+ "0.4": 0.0035281414987741178,
271
+ "0.45": -7.329788252118386e-05
272
+ },
273
+ "PushCube-v1": {
274
+ "0.35": -0.0022927102508079724,
275
+ "0.4": 0.003977617209635831,
276
+ "0.45": -0.00024147939253209573
277
+ },
278
+ "StackCube-v1": {
279
+ "0.35": 0.001056402071985707,
280
+ "0.4": -0.0015040775686485542,
281
+ "0.45": -3.629860183358425e-05
282
+ }
283
+ },
284
+ "candidate_type_bonus_components": false,
285
+ "candidate_oracle_rollouts": 0,
286
+ "candidate_oracle_unique_tolerance": 1e-06,
287
+ "candidate_oracle_success_rate": null,
288
+ "candidate_oracle_progress": null,
289
+ "candidate_oracle_score": null,
290
+ "candidate_oracle_regret": null,
291
+ "candidate_oracle_score_gain_over_selected": null,
292
+ "candidate_oracle_unique_count": null,
293
+ "candidate_oracle_improvement_rate": null,
294
+ "candidate_oracle_selected_branch_success_rate": null,
295
+ "candidate_oracle_selected_branch_progress": null,
296
+ "candidate_oracle_type_counts": {},
297
+ "candidate_oracle_best_branch_rank": null,
298
+ "candidate_oracle_best_branch_rank_counts": {},
299
+ "candidate_oracle_branch_success_rates": [],
300
+ "candidate_oracle_branch_progress": [],
301
+ "candidate_oracle_branch_score_gains_over_selected": [],
302
+ "selected_residual_scale_counts": {
303
+ "0.45": 307,
304
+ "0.35": 128,
305
+ "0.4": 140
306
+ },
307
+ "policy_rollout_success_rate": 0.3791304347826087,
308
+ "policy_rollout_progress": 0.5929921348668311,
309
+ "oracle_success_rate": 0.8695652173913043,
310
+ "action_mse_to_best": 0.5128354561774303,
311
+ "best_policy_val": {
312
+ "bc_loss": 0.31955528176493114,
313
+ "field_effect_loss": 0.010269693254182736,
314
+ "field_potential_loss": 0.30970407525698346,
315
+ "field_preference_loss": 0.5266202224625481,
316
+ "lattice_edges": 3833.3333333333335,
317
+ "progress_mae": 0.20275857630703184,
318
+ "proposal_loss": 0.0,
319
+ "rank_acc": 0.8124439285861121,
320
+ "rank_loss": 0.5266130566596985,
321
+ "regret_mae": 0.40278742214043933,
322
+ "success_accuracy": 0.8677195310592651,
323
+ "total_loss": 0.2706700298521254,
324
+ "transport_field_candidates": 0.0,
325
+ "transport_field_edges": 0.0,
326
+ "transport_field_loss": 0.0,
327
+ "transport_field_potential_loss": 0.0,
328
+ "transport_field_preference_loss": 0.0,
329
+ "transport_field_rank_acc": 0.0
330
+ },
331
+ "per_task": {
332
+ "LiftPegUpright-v1": {
333
+ "action_mse_to_best": 0.41795733929982626,
334
+ "expert_success_rate": 0.8584070796460177,
335
+ "num_groups": 113,
336
+ "oracle_success_rate": 0.9380530973451328,
337
+ "policy_expert_regret": 0.9912920312016411,
338
+ "policy_oracle_regret": 1.0707437862864637,
339
+ "policy_rollout_progress": 0.6289267474043686,
340
+ "policy_rollout_success_rate": 0.22123893805309736,
341
+ "restore_max_error": 2.384185791015625e-07
342
+ },
343
+ "PickCube-v1": {
344
+ "action_mse_to_best": 0.4553450090891641,
345
+ "expert_success_rate": 0.9402173913043478,
346
+ "num_groups": 184,
347
+ "oracle_success_rate": 0.9456521739130435,
348
+ "policy_expert_regret": 1.032134129154577,
349
+ "policy_oracle_regret": 1.032158133629507,
350
+ "policy_rollout_progress": 0.580459024418024,
351
+ "policy_rollout_success_rate": 0.32608695652173914,
352
+ "restore_max_error": 2.384185791015625e-07
353
+ },
354
+ "PullCube-v1": {
355
+ "action_mse_to_best": 0.6720873946814161,
356
+ "expert_success_rate": 0.25,
357
+ "num_groups": 76,
358
+ "oracle_success_rate": 0.40789473684210525,
359
+ "policy_expert_regret": 0.214007350714191,
360
+ "policy_oracle_regret": 0.4097537608130982,
361
+ "policy_rollout_progress": 0.34219757898857717,
362
+ "policy_rollout_success_rate": 0.2236842105263158,
363
+ "restore_max_error": 2.384185791015625e-07
364
+ },
365
+ "PushCube-v1": {
366
+ "action_mse_to_best": 0.42792120408098977,
367
+ "expert_success_rate": 0.8198198198198198,
368
+ "num_groups": 111,
369
+ "oracle_success_rate": 1.0,
370
+ "policy_expert_regret": 0.2613556649502333,
371
+ "policy_oracle_regret": 0.27852323208306284,
372
+ "policy_rollout_progress": 0.8656209120610813,
373
+ "policy_rollout_success_rate": 0.8558558558558559,
374
+ "restore_max_error": 2.3795291781425476e-07
375
+ },
376
+ "StackCube-v1": {
377
+ "action_mse_to_best": 0.717470666373169,
378
+ "expert_success_rate": 0.6923076923076923,
379
+ "num_groups": 91,
380
+ "oracle_success_rate": 0.8571428571428571,
381
+ "policy_expert_regret": 1.0032406121819883,
382
+ "policy_oracle_regret": 1.163062938622066,
383
+ "policy_rollout_progress": 0.450619311614351,
384
+ "policy_rollout_success_rate": 0.23076923076923078,
385
+ "restore_max_error": 1.9371509552001953e-07
386
+ }
387
+ }
388
+ },
389
+ {
390
+ "seed": 2,
391
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_2/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010.json",
392
+ "num_groups": 575,
393
+ "selection_mode": "retrieval_residual",
394
+ "num_candidates": 48,
395
+ "candidate_sigma": 0.0,
396
+ "proposal_types": [],
397
+ "proposal_lattice_types": [],
398
+ "selection_margin": 0.0,
399
+ "prepend_policy_candidate": false,
400
+ "field_optim_steps": 0,
401
+ "field_optim_step_size": 0.0,
402
+ "field_optim_trust_radius": 0.0,
403
+ "field_optim_l2_penalty": 0.0,
404
+ "retrieval_neighbors": 6,
405
+ "retrieval_metric": "raw",
406
+ "retrieval_type_min_success": 0.0,
407
+ "retrieval_type_success_bonus_scale": 0.0,
408
+ "retrieval_residual_consensus_penalty_scale": 0.0,
409
+ "retrieval_residual_min_source_progress": 0.0,
410
+ "retrieval_residual_min_source_advantage": -1000000000.0,
411
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
412
+ "retrieval_residual_source_score_bonus_scale": 0.0,
413
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
414
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
415
+ "retrieval_residual_scale": 1.0,
416
+ "retrieval_residual_scales": [
417
+ 0.35,
418
+ 0.4,
419
+ 0.45
420
+ ],
421
+ "retrieval_residual_anchor": "expert",
422
+ "retrieval_residual_direction": "candidate_minus_anchor",
423
+ "retrieval_residual_reduce": "compose_mean_by_type",
424
+ "retrieval_residual_challenger_types": [],
425
+ "retrieval_residual_challenger_scales": [],
426
+ "retrieval_residual_challenger_margin": 0.0,
427
+ "retrieval_residual_challenger_type_margins": {},
428
+ "retrieval_residual_challenger_tasks": [],
429
+ "retrieval_residual_challenger_type_tasks": {},
430
+ "lattice_exclude_types": [
431
+ "residual_random_negative",
432
+ "residual_wrong_direction",
433
+ "residual_near_miss+residual_no_op",
434
+ "residual_no_op+residual_wrong_gripper"
435
+ ],
436
+ "lattice_exclude_type_tasks": {},
437
+ "candidate_type_bonuses": {},
438
+ "candidate_type_bonuses_by_task": {},
439
+ "residual_scale_bonuses_by_task": {
440
+ "*": {
441
+ "0.35": -0.00038699918042431894,
442
+ "0.4": 0.0012427963812353005,
443
+ "0.45": -0.0006712357370645905
444
+ },
445
+ "LiftPegUpright-v1": {
446
+ "0.35": -0.00035875710229660296,
447
+ "0.4": -0.0012499262354289443,
448
+ "0.45": 0.002299295834620463
449
+ },
450
+ "PickCube-v1": {
451
+ "0.35": -0.0006463253407612779,
452
+ "0.4": 0.00047671350513027737,
453
+ "0.45": 0.0006365717991340781
454
+ },
455
+ "PullCube-v1": {
456
+ "0.35": -0.0021359270258899388,
457
+ "0.4": 0.0035281414987741178,
458
+ "0.45": -7.329788252118386e-05
459
+ },
460
+ "PushCube-v1": {
461
+ "0.35": -0.0022927102508079724,
462
+ "0.4": 0.003977617209635831,
463
+ "0.45": -0.00024147939253209573
464
+ },
465
+ "StackCube-v1": {
466
+ "0.35": 0.001056402071985707,
467
+ "0.4": -0.0015040775686485542,
468
+ "0.45": -3.629860183358425e-05
469
+ }
470
+ },
471
+ "candidate_type_bonus_components": false,
472
+ "candidate_oracle_rollouts": 0,
473
+ "candidate_oracle_unique_tolerance": 1e-06,
474
+ "candidate_oracle_success_rate": null,
475
+ "candidate_oracle_progress": null,
476
+ "candidate_oracle_score": null,
477
+ "candidate_oracle_regret": null,
478
+ "candidate_oracle_score_gain_over_selected": null,
479
+ "candidate_oracle_unique_count": null,
480
+ "candidate_oracle_improvement_rate": null,
481
+ "candidate_oracle_selected_branch_success_rate": null,
482
+ "candidate_oracle_selected_branch_progress": null,
483
+ "candidate_oracle_type_counts": {},
484
+ "candidate_oracle_best_branch_rank": null,
485
+ "candidate_oracle_best_branch_rank_counts": {},
486
+ "candidate_oracle_branch_success_rates": [],
487
+ "candidate_oracle_branch_progress": [],
488
+ "candidate_oracle_branch_score_gains_over_selected": [],
489
+ "selected_residual_scale_counts": {
490
+ "0.45": 278,
491
+ "0.4": 148,
492
+ "0.35": 149
493
+ },
494
+ "policy_rollout_success_rate": 0.4052173913043478,
495
+ "policy_rollout_progress": 0.6154099112441358,
496
+ "oracle_success_rate": 0.8765217391304347,
497
+ "action_mse_to_best": 0.5274760031165636,
498
+ "best_policy_val": {
499
+ "bc_loss": 0.2974168997671869,
500
+ "field_effect_loss": 0.009809228720971279,
501
+ "field_potential_loss": 0.29470594475666684,
502
+ "field_preference_loss": 0.5080636441707611,
503
+ "lattice_edges": 3833.3333333333335,
504
+ "progress_mae": 0.1944621221886741,
505
+ "proposal_loss": 0.0,
506
+ "rank_acc": 0.8252745270729065,
507
+ "rank_loss": 0.5080037944846683,
508
+ "regret_mae": 0.3989163488149643,
509
+ "success_accuracy": 0.870691223276986,
510
+ "total_loss": 0.2595750474267536,
511
+ "transport_field_candidates": 0.0,
512
+ "transport_field_edges": 0.0,
513
+ "transport_field_loss": 0.0,
514
+ "transport_field_potential_loss": 0.0,
515
+ "transport_field_preference_loss": 0.0,
516
+ "transport_field_rank_acc": 0.0
517
+ },
518
+ "per_task": {
519
+ "LiftPegUpright-v1": {
520
+ "action_mse_to_best": 0.44695540916291066,
521
+ "expert_success_rate": 0.8229166666666666,
522
+ "num_groups": 96,
523
+ "oracle_success_rate": 0.9270833333333334,
524
+ "policy_expert_regret": 0.8176391956706842,
525
+ "policy_oracle_regret": 0.9106746912002563,
526
+ "policy_rollout_progress": 0.6583504987259706,
527
+ "policy_rollout_success_rate": 0.3645833333333333,
528
+ "restore_max_error": 1.955777406692505e-07
529
+ },
530
+ "PickCube-v1": {
531
+ "action_mse_to_best": 0.4653542496527385,
532
+ "expert_success_rate": 0.9444444444444444,
533
+ "num_groups": 198,
534
+ "oracle_success_rate": 0.9595959595959596,
535
+ "policy_expert_regret": 0.9609521525356957,
536
+ "policy_oracle_regret": 0.968035603064404,
537
+ "policy_rollout_progress": 0.6383024751105242,
538
+ "policy_rollout_success_rate": 0.3484848484848485,
539
+ "restore_max_error": 2.384185791015625e-07
540
+ },
541
+ "PullCube-v1": {
542
+ "action_mse_to_best": 0.6751722337471114,
543
+ "expert_success_rate": 0.24444444444444444,
544
+ "num_groups": 90,
545
+ "oracle_success_rate": 0.4666666666666667,
546
+ "policy_expert_regret": 0.3269892188720405,
547
+ "policy_oracle_regret": 0.471620392271628,
548
+ "policy_rollout_progress": 0.3672527788724336,
549
+ "policy_rollout_success_rate": 0.25555555555555554,
550
+ "restore_max_error": 2.384185791015625e-07
551
+ },
552
+ "PushCube-v1": {
553
+ "action_mse_to_best": 0.5135416065407271,
554
+ "expert_success_rate": 0.8514851485148515,
555
+ "num_groups": 101,
556
+ "oracle_success_rate": 1.0,
557
+ "policy_expert_regret": 0.24849979933535699,
558
+ "policy_oracle_regret": 0.2848255383496237,
559
+ "policy_rollout_progress": 0.8636893131355248,
560
+ "policy_rollout_success_rate": 0.8514851485148515,
561
+ "restore_max_error": 2.384185791015625e-07
562
+ },
563
+ "StackCube-v1": {
564
+ "action_mse_to_best": 0.617973753147655,
565
+ "expert_success_rate": 0.7666666666666667,
566
+ "num_groups": 90,
567
+ "oracle_success_rate": 0.9111111111111111,
568
+ "policy_expert_regret": 1.051682802869214,
569
+ "policy_oracle_regret": 1.1881879919105105,
570
+ "policy_rollout_progress": 0.4887754476732678,
571
+ "policy_rollout_success_rate": 0.2222222222222222,
572
+ "restore_max_error": 2.2351741790771484e-07
573
+ }
574
+ }
575
+ }
576
+ ]
577
+ }
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010_summary.md ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # h=16 Best-Policy Checkpoint Rollout
2
+
3
+ Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
4
+ Objective: `transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1`
5
+ Result file: `policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010.json`
6
+ Completed seeds: 3
7
+ Baseline h=4 policy success: 29.67%
8
+ Baseline h=16 rank-checkpoint success: 29.74%
9
+
10
+ Mean success: 38.67% +/- 1.62%
11
+ Gain vs h=16 rank checkpoint: +8.93%
12
+ Mean progress: 59.82%
13
+ Mean action MSE to best: 0.513
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 37.57% | 58.63% | 85.74% | n/a | n/a | 0.499 |
18
+ | 1 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 37.91% | 59.30% | 86.96% | n/a | n/a | 0.513 |
19
+ | 2 | retrieval_residual | 48 | no | 6 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 40.52% | 61.54% | 87.65% | n/a | n/a | 0.527 |
workspace/results/paper_analysis.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "best_clean_key": "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005",
3
- "generated_utc": "2026-07-02T15:50:54+00:00",
4
  "mechanism_gap": {
5
  "best_clean_vs_direct_same_ckpt": 0.1060869565217391,
6
  "best_clean_vs_h16": 0.09159420289855075,
@@ -13041,25 +13041,292 @@
13041
  "std_success": 0.017764119937442965
13042
  },
13043
  "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025": {
 
 
13044
  "headline_metric": "mean_success",
13045
  "headline_metric_label": "deployment",
 
 
13046
  "label": "K6-matched transported residual field re-grounding, all-train tangent-length prior 0.025",
13047
- "missing": true,
13048
- "source": "results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025_summary.json"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13049
  },
13050
  "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005": {
 
 
13051
  "headline_metric": "mean_success",
13052
  "headline_metric_label": "deployment",
 
 
13053
  "label": "K6-matched transported residual field re-grounding, all-train tangent-length prior 0.05",
13054
- "missing": true,
13055
- "source": "results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005_summary.json"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13056
  },
13057
  "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010": {
 
 
13058
  "headline_metric": "mean_success",
13059
  "headline_metric_label": "deployment",
 
 
13060
  "label": "K6-matched transported residual field re-grounding, all-train tangent-length prior 0.10",
13061
- "missing": true,
13062
- "source": "results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010_summary.json"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13063
  },
13064
  "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_typesuccess001": {
13065
  "ci95_success": 0.04240637681159422,
 
1
  {
2
  "best_clean_key": "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005",
3
+ "generated_utc": "2026-07-02T16:52:46+00:00",
4
  "mechanism_gap": {
5
  "best_clean_vs_direct_same_ckpt": 0.1060869565217391,
6
  "best_clean_vs_h16": 0.09159420289855075,
 
13041
  "std_success": 0.017764119937442965
13042
  },
13043
  "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025": {
13044
+ "ci95_success": 0.043707043917985716,
13045
+ "headline_ci95_success": 0.043707043917985716,
13046
  "headline_metric": "mean_success",
13047
  "headline_metric_label": "deployment",
13048
+ "headline_std_success": 0.01759303293390554,
13049
+ "headline_success": 0.38840579710144935,
13050
  "label": "K6-matched transported residual field re-grounding, all-train tangent-length prior 0.025",
13051
+ "mean_action_mse_to_best": 0.5128867500585814,
13052
+ "mean_progress": 0.5993607256448139,
13053
+ "mean_success": 0.38840579710144935,
13054
+ "num_completed": 3,
13055
+ "per_task_success": {
13056
+ "LiftPegUpright-v1": {
13057
+ "mean_num_groups": 102.0,
13058
+ "mean_success": 0.27087546503259025,
13059
+ "std_success": 0.08120108533113783
13060
+ },
13061
+ "PickCube-v1": {
13062
+ "mean_num_groups": 196.66666666666666,
13063
+ "mean_success": 0.3450495760278369,
13064
+ "std_success": 0.01749972202307378
13065
+ },
13066
+ "PullCube-v1": {
13067
+ "mean_num_groups": 81.0,
13068
+ "mean_success": 0.2537340826814511,
13069
+ "std_success": 0.02754240597049266
13070
+ },
13071
+ "PushCube-v1": {
13072
+ "mean_num_groups": 102.0,
13073
+ "mean_success": 0.8244328170597393,
13074
+ "std_success": 0.05068828725828965
13075
+ },
13076
+ "StackCube-v1": {
13077
+ "mean_num_groups": 93.33333333333333,
13078
+ "mean_success": 0.24527324527324526,
13079
+ "std_success": 0.03280317778959639
13080
+ }
13081
+ },
13082
+ "seed_action_mse_to_best": {
13083
+ "0": 0.49790886099085857,
13084
+ "1": 0.5136578851476635,
13085
+ "2": 0.5270935040372221
13086
+ },
13087
+ "seed_progress": {
13088
+ "0": 0.5870517010922017,
13089
+ "1": 0.5928320967908138,
13090
+ "2": 0.6181983790514262
13091
+ },
13092
+ "seed_success": {
13093
+ "0": 0.3773913043478261,
13094
+ "1": 0.3791304347826087,
13095
+ "2": 0.40869565217391307
13096
+ },
13097
+ "selected_candidate_type_counts": {
13098
+ "retrieval_residual_policy_residual": 613,
13099
+ "retrieval_residual_residual_near_miss": 202,
13100
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 212,
13101
+ "retrieval_residual_residual_no_op": 234,
13102
+ "retrieval_residual_residual_wrong_gripper": 464
13103
+ },
13104
+ "selected_residual_scale_counts": {
13105
+ "0.35": 452,
13106
+ "0.4": 395,
13107
+ "0.45": 878
13108
+ },
13109
+ "selected_type_outcomes": {
13110
+ "retrieval_residual_policy_residual": {
13111
+ "count": 613.0,
13112
+ "mean_progress": 0.5483487874776927,
13113
+ "success_rate": 0.2969004893964111
13114
+ },
13115
+ "retrieval_residual_residual_near_miss": {
13116
+ "count": 202.0,
13117
+ "mean_progress": 0.7936865563624271,
13118
+ "success_rate": 0.6534653465346535
13119
+ },
13120
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": {
13121
+ "count": 212.0,
13122
+ "mean_progress": 0.7113448614957478,
13123
+ "success_rate": 0.5094339622641509
13124
+ },
13125
+ "retrieval_residual_residual_no_op": {
13126
+ "count": 234.0,
13127
+ "mean_progress": 0.5392063331002226,
13128
+ "success_rate": 0.3547008547008547
13129
+ },
13130
+ "retrieval_residual_residual_wrong_gripper": {
13131
+ "count": 464.0,
13132
+ "mean_progress": 0.5613262242364601,
13133
+ "success_rate": 0.35560344827586204
13134
+ }
13135
+ },
13136
+ "source": "results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025_summary.json",
13137
+ "std_success": 0.01759303293390554
13138
  },
13139
  "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005": {
13140
+ "ci95_success": 0.04121578446082103,
13141
+ "headline_ci95_success": 0.04121578446082103,
13142
  "headline_metric": "mean_success",
13143
  "headline_metric_label": "deployment",
13144
+ "headline_std_success": 0.01659024698116427,
13145
+ "headline_success": 0.38782608695652177,
13146
  "label": "K6-matched transported residual field re-grounding, all-train tangent-length prior 0.05",
13147
+ "mean_action_mse_to_best": 0.5130268671047752,
13148
+ "mean_progress": 0.5989091341704994,
13149
+ "mean_success": 0.38782608695652177,
13150
+ "num_completed": 3,
13151
+ "per_task_success": {
13152
+ "LiftPegUpright-v1": {
13153
+ "mean_num_groups": 102.0,
13154
+ "mean_success": 0.27087546503259025,
13155
+ "std_success": 0.08120108533113783
13156
+ },
13157
+ "PickCube-v1": {
13158
+ "mean_num_groups": 196.66666666666666,
13159
+ "mean_success": 0.3450495760278369,
13160
+ "std_success": 0.01749972202307378
13161
+ },
13162
+ "PullCube-v1": {
13163
+ "mean_num_groups": 81.0,
13164
+ "mean_success": 0.2500303789777474,
13165
+ "std_success": 0.023077786338402385
13166
+ },
13167
+ "PushCube-v1": {
13168
+ "mean_num_groups": 102.0,
13169
+ "mean_success": 0.8244328170597393,
13170
+ "std_success": 0.05068828725828965
13171
+ },
13172
+ "StackCube-v1": {
13173
+ "mean_num_groups": 93.33333333333333,
13174
+ "mean_success": 0.24527324527324526,
13175
+ "std_success": 0.03280317778959639
13176
+ }
13177
+ },
13178
+ "seed_action_mse_to_best": {
13179
+ "0": 0.4981027377445413,
13180
+ "1": 0.5134497352065923,
13181
+ "2": 0.5275281283631921
13182
+ },
13183
+ "seed_progress": {
13184
+ "0": 0.5870847139799077,
13185
+ "1": 0.5928355075111208,
13186
+ "2": 0.6168071810204697
13187
+ },
13188
+ "seed_success": {
13189
+ "0": 0.3773913043478261,
13190
+ "1": 0.3791304347826087,
13191
+ "2": 0.40695652173913044
13192
+ },
13193
+ "selected_candidate_type_counts": {
13194
+ "retrieval_residual_policy_residual": 619,
13195
+ "retrieval_residual_residual_near_miss": 201,
13196
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 212,
13197
+ "retrieval_residual_residual_no_op": 234,
13198
+ "retrieval_residual_residual_wrong_gripper": 459
13199
+ },
13200
+ "selected_residual_scale_counts": {
13201
+ "0.35": 433,
13202
+ "0.4": 419,
13203
+ "0.45": 873
13204
+ },
13205
+ "selected_type_outcomes": {
13206
+ "retrieval_residual_policy_residual": {
13207
+ "count": 619.0,
13208
+ "mean_progress": 0.5483998409329472,
13209
+ "success_rate": 0.2972536348949919
13210
+ },
13211
+ "retrieval_residual_residual_near_miss": {
13212
+ "count": 201.0,
13213
+ "mean_progress": 0.7949256523516938,
13214
+ "success_rate": 0.6567164179104478
13215
+ },
13216
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": {
13217
+ "count": 212.0,
13218
+ "mean_progress": 0.7113448614957478,
13219
+ "success_rate": 0.5094339622641509
13220
+ },
13221
+ "retrieval_residual_residual_no_op": {
13222
+ "count": 234.0,
13223
+ "mean_progress": 0.5393127208957688,
13224
+ "success_rate": 0.3547008547008547
13225
+ },
13226
+ "retrieval_residual_residual_wrong_gripper": {
13227
+ "count": 459.0,
13228
+ "mean_progress": 0.5596392406475342,
13229
+ "success_rate": 0.35294117647058826
13230
+ }
13231
+ },
13232
+ "source": "results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005_summary.json",
13233
+ "std_success": 0.01659024698116427
13234
  },
13235
  "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010": {
13236
+ "ci95_success": 0.04014506160601271,
13237
+ "headline_ci95_success": 0.04014506160601271,
13238
  "headline_metric": "mean_success",
13239
  "headline_metric_label": "deployment",
13240
+ "headline_std_success": 0.016159257814221853,
13241
+ "headline_success": 0.38666666666666666,
13242
  "label": "K6-matched transported residual field re-grounding, all-train tangent-length prior 0.10",
13243
+ "mean_action_mse_to_best": 0.513236562266028,
13244
+ "mean_progress": 0.5982465150498825,
13245
+ "mean_success": 0.38666666666666666,
13246
+ "num_completed": 3,
13247
+ "per_task_success": {
13248
+ "LiftPegUpright-v1": {
13249
+ "mean_num_groups": 102.0,
13250
+ "mean_success": 0.27087546503259025,
13251
+ "std_success": 0.08120108533113783
13252
+ },
13253
+ "PickCube-v1": {
13254
+ "mean_num_groups": 196.66666666666666,
13255
+ "mean_success": 0.34344701192527277,
13256
+ "std_success": 0.01546913698389715
13257
+ },
13258
+ "PullCube-v1": {
13259
+ "mean_num_groups": 81.0,
13260
+ "mean_success": 0.2463266752740437,
13261
+ "std_success": 0.01972026482244712
13262
+ },
13263
+ "PushCube-v1": {
13264
+ "mean_num_groups": 102.0,
13265
+ "mean_success": 0.8244328170597393,
13266
+ "std_success": 0.05068828725828965
13267
+ },
13268
+ "StackCube-v1": {
13269
+ "mean_num_groups": 93.33333333333333,
13270
+ "mean_success": 0.24527324527324526,
13271
+ "std_success": 0.03280317778959639
13272
+ }
13273
+ },
13274
+ "seed_action_mse_to_best": {
13275
+ "0": 0.49939822750409013,
13276
+ "1": 0.5128354561774303,
13277
+ "2": 0.5274760031165636
13278
+ },
13279
+ "seed_progress": {
13280
+ "0": 0.5863374990386807,
13281
+ "1": 0.5929921348668311,
13282
+ "2": 0.6154099112441358
13283
+ },
13284
+ "seed_success": {
13285
+ "0": 0.37565217391304345,
13286
+ "1": 0.3791304347826087,
13287
+ "2": 0.4052173913043478
13288
+ },
13289
+ "selected_candidate_type_counts": {
13290
+ "retrieval_residual_policy_residual": 624,
13291
+ "retrieval_residual_residual_near_miss": 202,
13292
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 213,
13293
+ "retrieval_residual_residual_no_op": 232,
13294
+ "retrieval_residual_residual_wrong_gripper": 454
13295
+ },
13296
+ "selected_residual_scale_counts": {
13297
+ "0.35": 405,
13298
+ "0.4": 457,
13299
+ "0.45": 863
13300
+ },
13301
+ "selected_type_outcomes": {
13302
+ "retrieval_residual_policy_residual": {
13303
+ "count": 624.0,
13304
+ "mean_progress": 0.5462547187762072,
13305
+ "success_rate": 0.2948717948717949
13306
+ },
13307
+ "retrieval_residual_residual_near_miss": {
13308
+ "count": 202.0,
13309
+ "mean_progress": 0.7938871512945631,
13310
+ "success_rate": 0.6534653465346535
13311
+ },
13312
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": {
13313
+ "count": 213.0,
13314
+ "mean_progress": 0.7080770921434315,
13315
+ "success_rate": 0.5070422535211268
13316
+ },
13317
+ "retrieval_residual_residual_no_op": {
13318
+ "count": 232.0,
13319
+ "mean_progress": 0.5436024628097897,
13320
+ "success_rate": 0.3577586206896552
13321
+ },
13322
+ "retrieval_residual_residual_wrong_gripper": {
13323
+ "count": 454.0,
13324
+ "mean_progress": 0.55905484005456,
13325
+ "success_rate": 0.3524229074889868
13326
+ }
13327
+ },
13328
+ "source": "results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010_summary.json",
13329
+ "std_success": 0.016159257814221853
13330
  },
13331
  "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_typesuccess001": {
13332
  "ci95_success": 0.04240637681159422,
workspace/results/paper_analysis.md CHANGED
@@ -1,6 +1,6 @@
1
  # Paper Analysis
2
 
3
- Generated: `2026-07-02T15:50:54+00:00`
4
 
5
  ## Main Seed Statistics
6
 
@@ -72,9 +72,9 @@ Generated: `2026-07-02T15:50:54+00:00`
72
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore001_metric_taskrel | K6-matched transported residual field re-grounding, source-score 0.01, task-relative retrieval chart | 3 | deployment | 37.22% +/- 1.66 | 37.22% | +/- 4.12 | 58.49% | 0.512 | +7.48 pp |
73
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore001_metric_taskrelz | K6-matched transported residual field re-grounding, source-score 0.01, task-relative z-score retrieval chart | 3 | deployment | 36.99% +/- 2.02 | 36.99% | +/- 5.01 | 58.49% | 0.511 | +7.25 pp |
74
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005 | K6-matched transported residual field re-grounding, task-conditioned source-score prior | 3 | deployment | 38.90% +/- 1.86 | 38.90% | +/- 4.62 | 59.96% | 0.513 | +9.16 pp |
75
- | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.025 | 0 | deployment | missing | missing | missing | missing | missing | missing |
76
- | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.05 | 0 | deployment | missing | missing | missing | missing | missing | missing |
77
- | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.10 | 0 | deployment | missing | missing | missing | missing | missing | missing |
78
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r005_t005 | K6-matched transported residual field re-grounding, train-oracle calibrated selector | 3 | deployment | 38.43% +/- 1.68 | 38.43% | +/- 4.17 | 59.70% | 0.512 | +8.70 pp |
79
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r0025_t0025 | K6-matched transported residual field re-grounding, small train-oracle selector calibration | 3 | deployment | 38.67% +/- 1.78 | 38.67% | +/- 4.41 | 59.79% | 0.513 | +8.93 pp |
80
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_rank005 | K6-matched transported residual field re-grounding, train-oracle rank calibration | 3 | deployment | 38.84% +/- 1.76 | 38.84% | +/- 4.37 | 59.93% | 0.513 | +9.10 pp |
 
1
  # Paper Analysis
2
 
3
+ Generated: `2026-07-02T16:52:46+00:00`
4
 
5
  ## Main Seed Statistics
6
 
 
72
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore001_metric_taskrel | K6-matched transported residual field re-grounding, source-score 0.01, task-relative retrieval chart | 3 | deployment | 37.22% +/- 1.66 | 37.22% | +/- 4.12 | 58.49% | 0.512 | +7.48 pp |
73
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore001_metric_taskrelz | K6-matched transported residual field re-grounding, source-score 0.01, task-relative z-score retrieval chart | 3 | deployment | 36.99% +/- 2.02 | 36.99% | +/- 5.01 | 58.49% | 0.511 | +7.25 pp |
74
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005 | K6-matched transported residual field re-grounding, task-conditioned source-score prior | 3 | deployment | 38.90% +/- 1.86 | 38.90% | +/- 4.62 | 59.96% | 0.513 | +9.16 pp |
75
+ | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.025 | 3 | deployment | 38.84% +/- 1.76 | 38.84% | +/- 4.37 | 59.94% | 0.513 | +9.10 pp |
76
+ | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.05 | 3 | deployment | 38.78% +/- 1.66 | 38.78% | +/- 4.12 | 59.89% | 0.513 | +9.04 pp |
77
+ | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.10 | 3 | deployment | 38.67% +/- 1.62 | 38.67% | +/- 4.01 | 59.82% | 0.513 | +8.93 pp |
78
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r005_t005 | K6-matched transported residual field re-grounding, train-oracle calibrated selector | 3 | deployment | 38.43% +/- 1.68 | 38.43% | +/- 4.17 | 59.70% | 0.512 | +8.70 pp |
79
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r0025_t0025 | K6-matched transported residual field re-grounding, small train-oracle selector calibration | 3 | deployment | 38.67% +/- 1.78 | 38.67% | +/- 4.41 | 59.79% | 0.513 | +8.93 pp |
80
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_rank005 | K6-matched transported residual field re-grounding, train-oracle rank calibration | 3 | deployment | 38.84% +/- 1.76 | 38.84% | +/- 4.37 | 59.93% | 0.513 | +9.10 pp |
workspace/results/paper_table_status.json CHANGED
@@ -1780,15 +1780,15 @@
1780
  "expert_proposal": "no",
1781
  "story_role": "tangent-length calibration learned from all train-split candidate-oracle traces",
1782
  "fallback_success": null,
1783
- "pending_job": "15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55",
1784
- "path_exists": false,
1785
- "status": "pending",
1786
- "success": null,
1787
- "std_success": null,
1788
  "completed_seeds": null,
1789
- "num_completed": null,
1790
  "best_config": null,
1791
- "gain_vs_h16_policy": null
1792
  },
1793
  {
1794
  "key": "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005",
@@ -1799,15 +1799,15 @@
1799
  "expert_proposal": "no",
1800
  "story_role": "tangent-length calibration learned from all train-split candidate-oracle traces",
1801
  "fallback_success": null,
1802
- "pending_job": "15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55",
1803
- "path_exists": false,
1804
- "status": "pending",
1805
- "success": null,
1806
- "std_success": null,
1807
  "completed_seeds": null,
1808
- "num_completed": null,
1809
  "best_config": null,
1810
- "gain_vs_h16_policy": null
1811
  },
1812
  {
1813
  "key": "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010",
@@ -1818,15 +1818,15 @@
1818
  "expert_proposal": "no",
1819
  "story_role": "tangent-length calibration learned from all train-split candidate-oracle traces",
1820
  "fallback_success": null,
1821
- "pending_job": "15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55",
1822
- "path_exists": false,
1823
- "status": "pending",
1824
- "success": null,
1825
- "std_success": null,
1826
  "completed_seeds": null,
1827
- "num_completed": null,
1828
  "best_config": null,
1829
- "gain_vs_h16_policy": null
1830
  },
1831
  {
1832
  "key": "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r005_t005",
 
1780
  "expert_proposal": "no",
1781
  "story_role": "tangent-length calibration learned from all train-split candidate-oracle traces",
1782
  "fallback_success": null,
1783
+ "pending_job": "train oracle 15056041; validation 15059951-53; summaries 15059963-65",
1784
+ "path_exists": true,
1785
+ "status": "complete",
1786
+ "success": 0.3884057971014493,
1787
+ "std_success": 0.01759303293390554,
1788
  "completed_seeds": null,
1789
+ "num_completed": 3,
1790
  "best_config": null,
1791
+ "gain_vs_h16_policy": 0.09101449275362322
1792
  },
1793
  {
1794
  "key": "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005",
 
1799
  "expert_proposal": "no",
1800
  "story_role": "tangent-length calibration learned from all train-split candidate-oracle traces",
1801
  "fallback_success": null,
1802
+ "pending_job": "train oracle 15056041; validation 15059951-53; summaries 15059963-65",
1803
+ "path_exists": true,
1804
+ "status": "complete",
1805
+ "success": 0.38782608695652177,
1806
+ "std_success": 0.01659024698116427,
1807
  "completed_seeds": null,
1808
+ "num_completed": 3,
1809
  "best_config": null,
1810
+ "gain_vs_h16_policy": 0.0904347826086957
1811
  },
1812
  {
1813
  "key": "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010",
 
1818
  "expert_proposal": "no",
1819
  "story_role": "tangent-length calibration learned from all train-split candidate-oracle traces",
1820
  "fallback_success": null,
1821
+ "pending_job": "train oracle 15056041; validation 15059951-53; summaries 15059963-65",
1822
+ "path_exists": true,
1823
+ "status": "complete",
1824
+ "success": 0.38666666666666666,
1825
+ "std_success": 0.016159257814221853,
1826
  "completed_seeds": null,
1827
+ "num_completed": 3,
1828
  "best_config": null,
1829
+ "gain_vs_h16_policy": 0.08927536231884059
1830
  },
1831
  {
1832
  "key": "transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r005_t005",
workspace/results/paper_table_status.md CHANGED
@@ -96,9 +96,9 @@ Baseline h=16 policy: 29.74%
96
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore001_metric_taskrel | K6-matched transported residual field re-grounding, source-score 0.01, task-relative retrieval chart | complete | 37.22% | +7.48 pp | yes | no | no | object-relative local chart ablation for transported residual support selection |
97
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore001_metric_taskrelz | K6-matched transported residual field re-grounding, source-score 0.01, task-relative z-score retrieval chart | complete | 36.99% | +7.25 pp | yes | no | no | normalized object-relative local chart ablation for transported residual support selection |
98
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005 | K6-matched transported residual field re-grounding, task-conditioned source-score prior | complete | 38.90% | +9.16 pp | yes | no | no | task-conditioned train-source utility prior chosen from per-task source-score sensitivity |
99
- | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.025 | pending 15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55 | pending | pending | yes | no | no | tangent-length calibration learned from all train-split candidate-oracle traces |
100
- | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.05 | pending 15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55 | pending | pending | yes | no | no | tangent-length calibration learned from all train-split candidate-oracle traces |
101
- | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.10 | pending 15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55 | pending | pending | yes | no | no | tangent-length calibration learned from all train-split candidate-oracle traces |
102
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r005_t005 | K6-matched transported residual field re-grounding, train-oracle calibrated selector | complete | 38.43% | +8.70 pp | yes | no | no | selector calibration learned from train-split candidate-oracle rollouts and evaluated on held-out validation |
103
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r0025_t0025 | K6-matched transported residual field re-grounding, small train-oracle selector calibration | complete | 38.67% | +8.93 pp | yes | no | no | smaller train-oracle rank/type selector calibration learned without validation outcomes |
104
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_rank005 | K6-matched transported residual field re-grounding, train-oracle rank calibration | complete | 38.84% | +9.10 pp | yes | no | no | rank-only selector calibration learned from train-split candidate-oracle rollouts |
 
96
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore001_metric_taskrel | K6-matched transported residual field re-grounding, source-score 0.01, task-relative retrieval chart | complete | 37.22% | +7.48 pp | yes | no | no | object-relative local chart ablation for transported residual support selection |
97
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore001_metric_taskrelz | K6-matched transported residual field re-grounding, source-score 0.01, task-relative z-score retrieval chart | complete | 36.99% | +7.25 pp | yes | no | no | normalized object-relative local chart ablation for transported residual support selection |
98
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005 | K6-matched transported residual field re-grounding, task-conditioned source-score prior | complete | 38.90% | +9.16 pp | yes | no | no | task-conditioned train-source utility prior chosen from per-task source-score sensitivity |
99
+ | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale0025 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.025 | complete | 38.84% | +9.10 pp | yes | no | no | tangent-length calibration learned from all train-split candidate-oracle traces |
100
+ | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.05 | complete | 38.78% | +9.04 pp | yes | no | no | tangent-length calibration learned from all train-split candidate-oracle traces |
101
+ | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010 | K6-matched transported residual field re-grounding, all-train tangent-length prior 0.10 | complete | 38.67% | +8.93 pp | yes | no | no | tangent-length calibration learned from all train-split candidate-oracle traces |
102
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r005_t005 | K6-matched transported residual field re-grounding, train-oracle calibrated selector | complete | 38.43% | +8.70 pp | yes | no | no | selector calibration learned from train-split candidate-oracle rollouts and evaluated on held-out validation |
103
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r0025_t0025 | K6-matched transported residual field re-grounding, small train-oracle selector calibration | complete | 38.67% | +8.93 pp | yes | no | no | smaller train-oracle rank/type selector calibration learned without validation outcomes |
104
  | transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_rank005 | K6-matched transported residual field re-grounding, train-oracle rank calibration | complete | 38.84% | +9.10 pp | yes | no | no | rank-only selector calibration learned from train-split candidate-oracle rollouts |
workspace/scripts/build_paper_table_status.py CHANGED
@@ -935,7 +935,7 @@ SPECS = [
935
  same_state_proposals="no",
936
  expert_proposal="no",
937
  story_role="tangent-length calibration learned from all train-split candidate-oracle traces",
938
- pending_job="15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55",
939
  ),
940
  ResultSpec(
941
  key="transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005",
@@ -945,7 +945,7 @@ SPECS = [
945
  same_state_proposals="no",
946
  expert_proposal="no",
947
  story_role="tangent-length calibration learned from all train-split candidate-oracle traces",
948
- pending_job="15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55",
949
  ),
950
  ResultSpec(
951
  key="transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010",
@@ -955,7 +955,7 @@ SPECS = [
955
  same_state_proposals="no",
956
  expert_proposal="no",
957
  story_role="tangent-length calibration learned from all train-split candidate-oracle traces",
958
- pending_job="15056041 -> 15056043-45 -> 15056050-52 -> 15056053-55",
959
  ),
960
  ResultSpec(
961
  key="transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r005_t005",
 
935
  same_state_proposals="no",
936
  expert_proposal="no",
937
  story_role="tangent-length calibration learned from all train-split candidate-oracle traces",
938
+ pending_job="train oracle 15056041; validation 15059951-53; summaries 15059963-65",
939
  ),
940
  ResultSpec(
941
  key="transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale005",
 
945
  same_state_proposals="no",
946
  expert_proposal="no",
947
  story_role="tangent-length calibration learned from all train-split candidate-oracle traces",
948
+ pending_job="train oracle 15056041; validation 15059951-53; summaries 15059963-65",
949
  ),
950
  ResultSpec(
951
  key="transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_scaleoracle_trainall_k4_scale010",
 
955
  same_state_proposals="no",
956
  expert_proposal="no",
957
  story_role="tangent-length calibration learned from all train-split candidate-oracle traces",
958
+ pending_job="train oracle 15056041; validation 15059951-53; summaries 15059963-65",
959
  ),
960
  ResultSpec(
961
  key="transport_field_reground_fieldonly_k6matched_b12_clean_k6_dropnoopwg_retargeted_srcscore_task_pick001_stack005_oraclecal_train800_k4_r005_t005",