anhtld commited on
Commit
2ceb4ca
·
verified ·
1 Parent(s): 8790fea

Auto-sync: 2026-06-30 23:21:36

Browse files
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_noopwgcontact_challenger0p01_summary.json ADDED
@@ -0,0 +1,505 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
3
+ "objective": "near_miss_policy_bc5",
4
+ "out_name": "policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_noopwgcontact_challenger0p01.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.34144927536231884,
9
+ "std_success": 0.00821880978478714,
10
+ "mean_progress": 0.561136360562357,
11
+ "mean_action_mse_to_best": 0.4220015877709333,
12
+ "gain_vs_h4": 0.04474927536231882,
13
+ "gain_vs_h16_rank_checkpoint": 0.04405797101449277,
14
+ "rows": [
15
+ {
16
+ "seed": 0,
17
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_noopwgcontact_challenger0p01.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": 0.2,
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": 4,
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
+ "residual_near_miss",
52
+ "residual_no_op",
53
+ "residual_wrong_gripper"
54
+ ],
55
+ "retrieval_residual_challenger_scales": [],
56
+ "retrieval_residual_challenger_margin": 0.01,
57
+ "retrieval_residual_challenger_type_margins": {},
58
+ "retrieval_residual_challenger_tasks": [],
59
+ "retrieval_residual_challenger_type_tasks": {
60
+ "residual_no_op": [
61
+ "PickCube-v1",
62
+ "PullCube-v1",
63
+ "StackCube-v1"
64
+ ],
65
+ "residual_wrong_gripper": [
66
+ "PickCube-v1",
67
+ "PullCube-v1",
68
+ "StackCube-v1"
69
+ ]
70
+ },
71
+ "lattice_exclude_types": [
72
+ "residual_random_negative",
73
+ "residual_wrong_direction",
74
+ "residual_near_miss+residual_no_op"
75
+ ],
76
+ "lattice_exclude_type_tasks": {},
77
+ "candidate_type_bonuses": {
78
+ "retrieval_residual_no_op": 0.03
79
+ },
80
+ "candidate_type_bonus_components": false,
81
+ "candidate_oracle_rollouts": 0,
82
+ "candidate_oracle_unique_tolerance": 1e-06,
83
+ "candidate_oracle_success_rate": null,
84
+ "candidate_oracle_progress": null,
85
+ "candidate_oracle_score": null,
86
+ "candidate_oracle_regret": null,
87
+ "candidate_oracle_score_gain_over_selected": null,
88
+ "candidate_oracle_unique_count": null,
89
+ "candidate_oracle_improvement_rate": null,
90
+ "candidate_oracle_selected_branch_success_rate": null,
91
+ "candidate_oracle_selected_branch_progress": null,
92
+ "candidate_oracle_type_counts": {},
93
+ "candidate_oracle_best_branch_rank": null,
94
+ "candidate_oracle_best_branch_rank_counts": {},
95
+ "candidate_oracle_branch_success_rates": [],
96
+ "candidate_oracle_branch_progress": [],
97
+ "candidate_oracle_branch_score_gains_over_selected": [],
98
+ "selected_residual_scale_counts": {
99
+ "0.35": 454,
100
+ "0.45": 106,
101
+ "0.4": 15
102
+ },
103
+ "policy_rollout_success_rate": 0.33217391304347826,
104
+ "policy_rollout_progress": 0.5468972360149729,
105
+ "oracle_success_rate": 0.8573913043478261,
106
+ "action_mse_to_best": 0.4010052319470307,
107
+ "best_policy_val": {
108
+ "bc_loss": 0.13721593966086706,
109
+ "field_effect_loss": 0.009290305380192068,
110
+ "field_potential_loss": 0.2666468388504452,
111
+ "field_preference_loss": 0.5130573478009965,
112
+ "lattice_edges": 3833.3333333333335,
113
+ "progress_mae": 0.1933159919248687,
114
+ "rank_acc": 0.8265031774838766,
115
+ "rank_loss": 0.5130523675017886,
116
+ "regret_mae": 0.3756548762321472,
117
+ "success_accuracy": 0.8773836526605818,
118
+ "total_loss": 1.5581054819954767
119
+ },
120
+ "per_task": {
121
+ "LiftPegUpright-v1": {
122
+ "action_mse_to_best": 0.32276735697709713,
123
+ "expert_success_rate": 0.8865979381443299,
124
+ "num_groups": 97,
125
+ "oracle_success_rate": 0.9278350515463918,
126
+ "policy_expert_regret": 1.0783334703175063,
127
+ "policy_oracle_regret": 1.1047489480259491,
128
+ "policy_rollout_progress": 0.5714634232914325,
129
+ "policy_rollout_success_rate": 0.2268041237113402,
130
+ "restore_max_error": 2.384185791015625e-07
131
+ },
132
+ "PickCube-v1": {
133
+ "action_mse_to_best": 0.3095931297597977,
134
+ "expert_success_rate": 0.9375,
135
+ "num_groups": 208,
136
+ "oracle_success_rate": 0.9471153846153846,
137
+ "policy_expert_regret": 1.0586748375611092,
138
+ "policy_oracle_regret": 1.0720977049876148,
139
+ "policy_rollout_progress": 0.5762841139579765,
140
+ "policy_rollout_success_rate": 0.28365384615384615,
141
+ "restore_max_error": 2.384185791015625e-07
142
+ },
143
+ "PullCube-v1": {
144
+ "action_mse_to_best": 0.6097412405463009,
145
+ "expert_success_rate": 0.19480519480519481,
146
+ "num_groups": 77,
147
+ "oracle_success_rate": 0.36363636363636365,
148
+ "policy_expert_regret": 0.24474660938432255,
149
+ "policy_oracle_regret": 0.39496795063761714,
150
+ "policy_rollout_progress": 0.33171507053468346,
151
+ "policy_rollout_success_rate": 0.23376623376623376,
152
+ "restore_max_error": 2.3096799850463867e-07
153
+ },
154
+ "PushCube-v1": {
155
+ "action_mse_to_best": 0.3675119048578942,
156
+ "expert_success_rate": 0.8617021276595744,
157
+ "num_groups": 94,
158
+ "oracle_success_rate": 0.9893617021276596,
159
+ "policy_expert_regret": 0.5508309629686335,
160
+ "policy_oracle_regret": 0.628381609441118,
161
+ "policy_rollout_progress": 0.7014056246014352,
162
+ "policy_rollout_success_rate": 0.6702127659574468,
163
+ "restore_max_error": 2.384185791015625e-07
164
+ },
165
+ "StackCube-v1": {
166
+ "action_mse_to_best": 0.5391718098385767,
167
+ "expert_success_rate": 0.6767676767676768,
168
+ "num_groups": 99,
169
+ "oracle_success_rate": 0.8585858585858586,
170
+ "policy_expert_regret": 0.9101122845001896,
171
+ "policy_oracle_regret": 1.0577925431607949,
172
+ "policy_rollout_progress": 0.4817441798219777,
173
+ "policy_rollout_success_rate": 0.29292929292929293,
174
+ "restore_max_error": 2.384185791015625e-07
175
+ }
176
+ }
177
+ },
178
+ {
179
+ "seed": 1,
180
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_noopwgcontact_challenger0p01.json",
181
+ "num_groups": 575,
182
+ "selection_mode": "retrieval_residual",
183
+ "num_candidates": 48,
184
+ "candidate_sigma": 0.0,
185
+ "proposal_types": [],
186
+ "proposal_lattice_types": [],
187
+ "selection_margin": 0.2,
188
+ "prepend_policy_candidate": false,
189
+ "field_optim_steps": 0,
190
+ "field_optim_step_size": 0.0,
191
+ "field_optim_trust_radius": 0.0,
192
+ "field_optim_l2_penalty": 0.0,
193
+ "retrieval_neighbors": 4,
194
+ "retrieval_metric": "raw",
195
+ "retrieval_type_min_success": 0.0,
196
+ "retrieval_type_success_bonus_scale": 0.0,
197
+ "retrieval_residual_consensus_penalty_scale": 0.0,
198
+ "retrieval_residual_min_source_progress": 0.0,
199
+ "retrieval_residual_min_source_advantage": -1000000000.0,
200
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
201
+ "retrieval_residual_source_score_bonus_scale": 0.0,
202
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
203
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
204
+ "retrieval_residual_scale": 1.0,
205
+ "retrieval_residual_scales": [
206
+ 0.35,
207
+ 0.4,
208
+ 0.45
209
+ ],
210
+ "retrieval_residual_anchor": "expert",
211
+ "retrieval_residual_direction": "candidate_minus_anchor",
212
+ "retrieval_residual_reduce": "compose_mean_by_type",
213
+ "retrieval_residual_challenger_types": [
214
+ "residual_near_miss",
215
+ "residual_no_op",
216
+ "residual_wrong_gripper"
217
+ ],
218
+ "retrieval_residual_challenger_scales": [],
219
+ "retrieval_residual_challenger_margin": 0.01,
220
+ "retrieval_residual_challenger_type_margins": {},
221
+ "retrieval_residual_challenger_tasks": [],
222
+ "retrieval_residual_challenger_type_tasks": {
223
+ "residual_no_op": [
224
+ "PickCube-v1",
225
+ "PullCube-v1",
226
+ "StackCube-v1"
227
+ ],
228
+ "residual_wrong_gripper": [
229
+ "PickCube-v1",
230
+ "PullCube-v1",
231
+ "StackCube-v1"
232
+ ]
233
+ },
234
+ "lattice_exclude_types": [
235
+ "residual_random_negative",
236
+ "residual_wrong_direction",
237
+ "residual_near_miss+residual_no_op"
238
+ ],
239
+ "lattice_exclude_type_tasks": {},
240
+ "candidate_type_bonuses": {
241
+ "retrieval_residual_no_op": 0.03
242
+ },
243
+ "candidate_type_bonus_components": false,
244
+ "candidate_oracle_rollouts": 0,
245
+ "candidate_oracle_unique_tolerance": 1e-06,
246
+ "candidate_oracle_success_rate": null,
247
+ "candidate_oracle_progress": null,
248
+ "candidate_oracle_score": null,
249
+ "candidate_oracle_regret": null,
250
+ "candidate_oracle_score_gain_over_selected": null,
251
+ "candidate_oracle_unique_count": null,
252
+ "candidate_oracle_improvement_rate": null,
253
+ "candidate_oracle_selected_branch_success_rate": null,
254
+ "candidate_oracle_selected_branch_progress": null,
255
+ "candidate_oracle_type_counts": {},
256
+ "candidate_oracle_best_branch_rank": null,
257
+ "candidate_oracle_best_branch_rank_counts": {},
258
+ "candidate_oracle_branch_success_rates": [],
259
+ "candidate_oracle_branch_progress": [],
260
+ "candidate_oracle_branch_score_gains_over_selected": [],
261
+ "selected_residual_scale_counts": {
262
+ "0.35": 461,
263
+ "0.45": 90,
264
+ "0.4": 24
265
+ },
266
+ "policy_rollout_success_rate": 0.3443478260869565,
267
+ "policy_rollout_progress": 0.5690277994780437,
268
+ "oracle_success_rate": 0.8695652173913043,
269
+ "action_mse_to_best": 0.4123674940139703,
270
+ "best_policy_val": {
271
+ "bc_loss": 0.1315989200439718,
272
+ "field_effect_loss": 0.010110370814800262,
273
+ "field_potential_loss": 0.31546536915832096,
274
+ "field_preference_loss": 0.4960637440284093,
275
+ "lattice_edges": 3833.3333333333335,
276
+ "progress_mae": 0.19023226698239645,
277
+ "rank_acc": 0.8255469501018524,
278
+ "rank_loss": 0.4960531195004781,
279
+ "regret_mae": 0.42156532737943864,
280
+ "success_accuracy": 0.8807473679383596,
281
+ "total_loss": 1.561984618504842
282
+ },
283
+ "per_task": {
284
+ "LiftPegUpright-v1": {
285
+ "action_mse_to_best": 0.3281630301287612,
286
+ "expert_success_rate": 0.8584070796460177,
287
+ "num_groups": 113,
288
+ "oracle_success_rate": 0.9380530973451328,
289
+ "policy_expert_regret": 0.987445328615408,
290
+ "policy_oracle_regret": 1.0666473587002374,
291
+ "policy_rollout_progress": 0.6354643670858535,
292
+ "policy_rollout_success_rate": 0.23008849557522124,
293
+ "restore_max_error": 2.384185791015625e-07
294
+ },
295
+ "PickCube-v1": {
296
+ "action_mse_to_best": 0.36263899948528927,
297
+ "expert_success_rate": 0.9402173913043478,
298
+ "num_groups": 184,
299
+ "oracle_success_rate": 0.9456521739130435,
300
+ "policy_expert_regret": 1.087151286698392,
301
+ "policy_oracle_regret": 1.0871752911733221,
302
+ "policy_rollout_progress": 0.5613672655764157,
303
+ "policy_rollout_success_rate": 0.2826086956521739,
304
+ "restore_max_error": 2.384185791015625e-07
305
+ },
306
+ "PullCube-v1": {
307
+ "action_mse_to_best": 0.5888737403836689,
308
+ "expert_success_rate": 0.25,
309
+ "num_groups": 76,
310
+ "oracle_success_rate": 0.40789473684210525,
311
+ "policy_expert_regret": 0.27039701540611294,
312
+ "policy_oracle_regret": 0.517146605214006,
313
+ "policy_rollout_progress": 0.2963535526865407,
314
+ "policy_rollout_success_rate": 0.18421052631578946,
315
+ "restore_max_error": 2.384185791015625e-07
316
+ },
317
+ "PushCube-v1": {
318
+ "action_mse_to_best": 0.3595503710948669,
319
+ "expert_success_rate": 0.8198198198198198,
320
+ "num_groups": 111,
321
+ "oracle_success_rate": 1.0,
322
+ "policy_expert_regret": 0.3128753544779511,
323
+ "policy_oracle_regret": 0.33004292161078064,
324
+ "policy_rollout_progress": 0.8411282495603906,
325
+ "policy_rollout_success_rate": 0.8288288288288288,
326
+ "restore_max_error": 2.3795291781425476e-07
327
+ },
328
+ "StackCube-v1": {
329
+ "action_mse_to_best": 0.53449247568682,
330
+ "expert_success_rate": 0.6923076923076923,
331
+ "num_groups": 91,
332
+ "oracle_success_rate": 0.8571428571428571,
333
+ "policy_expert_regret": 1.1330992091488052,
334
+ "policy_oracle_regret": 1.292761056305288,
335
+ "policy_rollout_progress": 0.3978442708542059,
336
+ "policy_rollout_success_rate": 0.15384615384615385,
337
+ "restore_max_error": 1.9371509552001953e-07
338
+ }
339
+ }
340
+ },
341
+ {
342
+ "seed": 2,
343
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_noopwgcontact_challenger0p01.json",
344
+ "num_groups": 575,
345
+ "selection_mode": "retrieval_residual",
346
+ "num_candidates": 48,
347
+ "candidate_sigma": 0.0,
348
+ "proposal_types": [],
349
+ "proposal_lattice_types": [],
350
+ "selection_margin": 0.2,
351
+ "prepend_policy_candidate": false,
352
+ "field_optim_steps": 0,
353
+ "field_optim_step_size": 0.0,
354
+ "field_optim_trust_radius": 0.0,
355
+ "field_optim_l2_penalty": 0.0,
356
+ "retrieval_neighbors": 4,
357
+ "retrieval_metric": "raw",
358
+ "retrieval_type_min_success": 0.0,
359
+ "retrieval_type_success_bonus_scale": 0.0,
360
+ "retrieval_residual_consensus_penalty_scale": 0.0,
361
+ "retrieval_residual_min_source_progress": 0.0,
362
+ "retrieval_residual_min_source_advantage": -1000000000.0,
363
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
364
+ "retrieval_residual_source_score_bonus_scale": 0.0,
365
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
366
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
367
+ "retrieval_residual_scale": 1.0,
368
+ "retrieval_residual_scales": [
369
+ 0.35,
370
+ 0.4,
371
+ 0.45
372
+ ],
373
+ "retrieval_residual_anchor": "expert",
374
+ "retrieval_residual_direction": "candidate_minus_anchor",
375
+ "retrieval_residual_reduce": "compose_mean_by_type",
376
+ "retrieval_residual_challenger_types": [
377
+ "residual_near_miss",
378
+ "residual_no_op",
379
+ "residual_wrong_gripper"
380
+ ],
381
+ "retrieval_residual_challenger_scales": [],
382
+ "retrieval_residual_challenger_margin": 0.01,
383
+ "retrieval_residual_challenger_type_margins": {},
384
+ "retrieval_residual_challenger_tasks": [],
385
+ "retrieval_residual_challenger_type_tasks": {
386
+ "residual_no_op": [
387
+ "PickCube-v1",
388
+ "PullCube-v1",
389
+ "StackCube-v1"
390
+ ],
391
+ "residual_wrong_gripper": [
392
+ "PickCube-v1",
393
+ "PullCube-v1",
394
+ "StackCube-v1"
395
+ ]
396
+ },
397
+ "lattice_exclude_types": [
398
+ "residual_random_negative",
399
+ "residual_wrong_direction",
400
+ "residual_near_miss+residual_no_op"
401
+ ],
402
+ "lattice_exclude_type_tasks": {},
403
+ "candidate_type_bonuses": {
404
+ "retrieval_residual_no_op": 0.03
405
+ },
406
+ "candidate_type_bonus_components": false,
407
+ "candidate_oracle_rollouts": 0,
408
+ "candidate_oracle_unique_tolerance": 1e-06,
409
+ "candidate_oracle_success_rate": null,
410
+ "candidate_oracle_progress": null,
411
+ "candidate_oracle_score": null,
412
+ "candidate_oracle_regret": null,
413
+ "candidate_oracle_score_gain_over_selected": null,
414
+ "candidate_oracle_unique_count": null,
415
+ "candidate_oracle_improvement_rate": null,
416
+ "candidate_oracle_selected_branch_success_rate": null,
417
+ "candidate_oracle_selected_branch_progress": null,
418
+ "candidate_oracle_type_counts": {},
419
+ "candidate_oracle_best_branch_rank": null,
420
+ "candidate_oracle_best_branch_rank_counts": {},
421
+ "candidate_oracle_branch_success_rates": [],
422
+ "candidate_oracle_branch_progress": [],
423
+ "candidate_oracle_branch_score_gains_over_selected": [],
424
+ "selected_residual_scale_counts": {
425
+ "0.35": 436,
426
+ "0.45": 108,
427
+ "0.4": 31
428
+ },
429
+ "policy_rollout_success_rate": 0.34782608695652173,
430
+ "policy_rollout_progress": 0.5674840461940545,
431
+ "oracle_success_rate": 0.8765217391304347,
432
+ "action_mse_to_best": 0.45263203735179874,
433
+ "best_policy_val": {
434
+ "bc_loss": 0.11367896075050037,
435
+ "field_effect_loss": 0.009670218582161598,
436
+ "field_potential_loss": 0.2641640139950646,
437
+ "field_preference_loss": 0.5130490180518892,
438
+ "lattice_edges": 3833.3333333333335,
439
+ "progress_mae": 0.2021729110015763,
440
+ "rank_acc": 0.8333857821093665,
441
+ "rank_loss": 0.5130119257503085,
442
+ "regret_mae": 0.3958987047274907,
443
+ "success_accuracy": 0.8680730561415354,
444
+ "total_loss": 1.4394984311527677
445
+ },
446
+ "per_task": {
447
+ "LiftPegUpright-v1": {
448
+ "action_mse_to_best": 0.35602021238203935,
449
+ "expert_success_rate": 0.8229166666666666,
450
+ "num_groups": 96,
451
+ "oracle_success_rate": 0.9270833333333334,
452
+ "policy_expert_regret": 0.8476849501021206,
453
+ "policy_oracle_regret": 0.9539546399998168,
454
+ "policy_rollout_progress": 0.6463205499264101,
455
+ "policy_rollout_success_rate": 0.3333333333333333,
456
+ "restore_max_error": 1.955777406692505e-07
457
+ },
458
+ "PickCube-v1": {
459
+ "action_mse_to_best": 0.3939900371725812,
460
+ "expert_success_rate": 0.9444444444444444,
461
+ "num_groups": 198,
462
+ "oracle_success_rate": 0.9595959595959596,
463
+ "policy_expert_regret": 1.0321517002232599,
464
+ "policy_oracle_regret": 1.0463813957523065,
465
+ "policy_rollout_progress": 0.603372491943189,
466
+ "policy_rollout_success_rate": 0.29797979797979796,
467
+ "restore_max_error": 2.384185791015625e-07
468
+ },
469
+ "PullCube-v1": {
470
+ "action_mse_to_best": 0.6625336924360858,
471
+ "expert_success_rate": 0.24444444444444444,
472
+ "num_groups": 90,
473
+ "oracle_success_rate": 0.4666666666666667,
474
+ "policy_expert_regret": 0.29977832058341136,
475
+ "policy_oracle_regret": 0.5640585053556909,
476
+ "policy_rollout_progress": 0.30767469643615186,
477
+ "policy_rollout_success_rate": 0.2,
478
+ "restore_max_error": 2.384185791015625e-07
479
+ },
480
+ "PushCube-v1": {
481
+ "action_mse_to_best": 0.397049247997232,
482
+ "expert_success_rate": 0.8514851485148515,
483
+ "num_groups": 101,
484
+ "oracle_success_rate": 1.0,
485
+ "policy_expert_regret": 0.3614020819711213,
486
+ "policy_oracle_regret": 0.41630433337523204,
487
+ "policy_rollout_progress": 0.8015174488029858,
488
+ "policy_rollout_success_rate": 0.7821782178217822,
489
+ "restore_max_error": 2.384185791015625e-07
490
+ },
491
+ "StackCube-v1": {
492
+ "action_mse_to_best": 0.5371716373496586,
493
+ "expert_success_rate": 0.7666666666666667,
494
+ "num_groups": 90,
495
+ "oracle_success_rate": 0.9111111111111111,
496
+ "policy_expert_regret": 1.1891138343347443,
497
+ "policy_oracle_regret": 1.351234385702345,
498
+ "policy_rollout_progress": 0.40160905950599246,
499
+ "policy_rollout_success_rate": 0.13333333333333333,
500
+ "restore_max_error": 2.2351741790771484e-07
501
+ }
502
+ }
503
+ }
504
+ ]
505
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_noopwgcontact_challenger0p01_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: `near_miss_policy_bc5`
5
+ Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_noopwgcontact_challenger0p01.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: 34.14% +/- 0.82%
11
+ Gain vs h=16 rank checkpoint: +4.41%
12
+ Mean progress: 56.11%
13
+ Mean action MSE to best: 0.422
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 | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_no_op,residual_wrong_gripper with residual_no_op=PickCube-v1,PullCube-v1,StackCube-v1;residual_wrong_gripper=PickCube-v1,PullCube-v1,StackCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 33.22% | 54.69% | 85.74% | n/a | n/a | 0.401 |
18
+ | 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_no_op,residual_wrong_gripper with residual_no_op=PickCube-v1,PullCube-v1,StackCube-v1;residual_wrong_gripper=PickCube-v1,PullCube-v1,StackCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 34.43% | 56.90% | 86.96% | n/a | n/a | 0.412 |
19
+ | 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_no_op,residual_wrong_gripper with residual_no_op=PickCube-v1,PullCube-v1,StackCube-v1;residual_wrong_gripper=PickCube-v1,PullCube-v1,StackCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 34.78% | 56.75% | 87.65% | n/a | n/a | 0.453 |
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpullstack_challenger0p01_summary.json ADDED
@@ -0,0 +1,487 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
3
+ "objective": "near_miss_policy_bc5",
4
+ "out_name": "policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpullstack_challenger0p01.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.36,
9
+ "std_success": 0.01485913694837829,
10
+ "mean_progress": 0.5743561674773712,
11
+ "mean_action_mse_to_best": 0.4207441219177259,
12
+ "gain_vs_h4": 0.06329999999999997,
13
+ "gain_vs_h16_rank_checkpoint": 0.06260869565217392,
14
+ "rows": [
15
+ {
16
+ "seed": 0,
17
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpullstack_challenger0p01.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": 0.2,
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": 4,
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
+ "residual_near_miss",
52
+ "residual_wrong_gripper"
53
+ ],
54
+ "retrieval_residual_challenger_scales": [],
55
+ "retrieval_residual_challenger_margin": 0.01,
56
+ "retrieval_residual_challenger_type_margins": {},
57
+ "retrieval_residual_challenger_tasks": [],
58
+ "retrieval_residual_challenger_type_tasks": {
59
+ "residual_wrong_gripper": [
60
+ "PickCube-v1",
61
+ "PullCube-v1",
62
+ "StackCube-v1"
63
+ ]
64
+ },
65
+ "lattice_exclude_types": [
66
+ "residual_random_negative",
67
+ "residual_wrong_direction",
68
+ "residual_near_miss+residual_no_op"
69
+ ],
70
+ "lattice_exclude_type_tasks": {},
71
+ "candidate_type_bonuses": {
72
+ "retrieval_residual_no_op": 0.03
73
+ },
74
+ "candidate_type_bonus_components": false,
75
+ "candidate_oracle_rollouts": 0,
76
+ "candidate_oracle_unique_tolerance": 1e-06,
77
+ "candidate_oracle_success_rate": null,
78
+ "candidate_oracle_progress": null,
79
+ "candidate_oracle_score": null,
80
+ "candidate_oracle_regret": null,
81
+ "candidate_oracle_score_gain_over_selected": null,
82
+ "candidate_oracle_unique_count": null,
83
+ "candidate_oracle_improvement_rate": null,
84
+ "candidate_oracle_selected_branch_success_rate": null,
85
+ "candidate_oracle_selected_branch_progress": null,
86
+ "candidate_oracle_type_counts": {},
87
+ "candidate_oracle_best_branch_rank": null,
88
+ "candidate_oracle_best_branch_rank_counts": {},
89
+ "candidate_oracle_branch_success_rates": [],
90
+ "candidate_oracle_branch_progress": [],
91
+ "candidate_oracle_branch_score_gains_over_selected": [],
92
+ "selected_residual_scale_counts": {
93
+ "0.35": 501,
94
+ "0.45": 66,
95
+ "0.4": 8
96
+ },
97
+ "policy_rollout_success_rate": 0.3582608695652174,
98
+ "policy_rollout_progress": 0.5613569916915689,
99
+ "oracle_success_rate": 0.8573913043478261,
100
+ "action_mse_to_best": 0.39735263349083455,
101
+ "best_policy_val": {
102
+ "bc_loss": 0.13721593966086706,
103
+ "field_effect_loss": 0.009290305380192068,
104
+ "field_potential_loss": 0.2666468388504452,
105
+ "field_preference_loss": 0.5130573478009965,
106
+ "lattice_edges": 3833.3333333333335,
107
+ "progress_mae": 0.1933159919248687,
108
+ "rank_acc": 0.8265031774838766,
109
+ "rank_loss": 0.5130523675017886,
110
+ "regret_mae": 0.3756548762321472,
111
+ "success_accuracy": 0.8773836526605818,
112
+ "total_loss": 1.5581054819954767
113
+ },
114
+ "per_task": {
115
+ "LiftPegUpright-v1": {
116
+ "action_mse_to_best": 0.32276735697709713,
117
+ "expert_success_rate": 0.8865979381443299,
118
+ "num_groups": 97,
119
+ "oracle_success_rate": 0.9278350515463918,
120
+ "policy_expert_regret": 1.0783334703175063,
121
+ "policy_oracle_regret": 1.1047489480259491,
122
+ "policy_rollout_progress": 0.5714634232914325,
123
+ "policy_rollout_success_rate": 0.2268041237113402,
124
+ "restore_max_error": 2.384185791015625e-07
125
+ },
126
+ "PickCube-v1": {
127
+ "action_mse_to_best": 0.3064204964142006,
128
+ "expert_success_rate": 0.9375,
129
+ "num_groups": 208,
130
+ "oracle_success_rate": 0.9471153846153846,
131
+ "policy_expert_regret": 0.9335763099818275,
132
+ "policy_oracle_regret": 0.9469991774083331,
133
+ "policy_rollout_progress": 0.6244595646141813,
134
+ "policy_rollout_success_rate": 0.3605769230769231,
135
+ "restore_max_error": 2.384185791015625e-07
136
+ },
137
+ "PullCube-v1": {
138
+ "action_mse_to_best": 0.6037605747387007,
139
+ "expert_success_rate": 0.19480519480519481,
140
+ "num_groups": 77,
141
+ "oracle_success_rate": 0.36363636363636365,
142
+ "policy_expert_regret": 0.24754985438728278,
143
+ "policy_oracle_regret": 0.4211023997069932,
144
+ "policy_rollout_progress": 0.31856763445232034,
145
+ "policy_rollout_success_rate": 0.22077922077922077,
146
+ "restore_max_error": 2.3096799850463867e-07
147
+ },
148
+ "PushCube-v1": {
149
+ "action_mse_to_best": 0.3675119048578942,
150
+ "expert_success_rate": 0.8617021276595744,
151
+ "num_groups": 94,
152
+ "oracle_success_rate": 0.9893617021276596,
153
+ "policy_expert_regret": 0.5508309629686335,
154
+ "policy_oracle_regret": 0.628381609441118,
155
+ "policy_rollout_progress": 0.7014056246014352,
156
+ "policy_rollout_success_rate": 0.6702127659574468,
157
+ "restore_max_error": 2.384185791015625e-07
158
+ },
159
+ "StackCube-v1": {
160
+ "action_mse_to_best": 0.529274586512886,
161
+ "expert_success_rate": 0.6767676767676768,
162
+ "num_groups": 99,
163
+ "oracle_success_rate": 0.8585858585858586,
164
+ "policy_expert_regret": 0.9171201798951987,
165
+ "policy_oracle_regret": 1.064800438555804,
166
+ "policy_rollout_progress": 0.4747362844269685,
167
+ "policy_rollout_success_rate": 0.29292929292929293,
168
+ "restore_max_error": 2.384185791015625e-07
169
+ }
170
+ }
171
+ },
172
+ {
173
+ "seed": 1,
174
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpullstack_challenger0p01.json",
175
+ "num_groups": 575,
176
+ "selection_mode": "retrieval_residual",
177
+ "num_candidates": 48,
178
+ "candidate_sigma": 0.0,
179
+ "proposal_types": [],
180
+ "proposal_lattice_types": [],
181
+ "selection_margin": 0.2,
182
+ "prepend_policy_candidate": false,
183
+ "field_optim_steps": 0,
184
+ "field_optim_step_size": 0.0,
185
+ "field_optim_trust_radius": 0.0,
186
+ "field_optim_l2_penalty": 0.0,
187
+ "retrieval_neighbors": 4,
188
+ "retrieval_metric": "raw",
189
+ "retrieval_type_min_success": 0.0,
190
+ "retrieval_type_success_bonus_scale": 0.0,
191
+ "retrieval_residual_consensus_penalty_scale": 0.0,
192
+ "retrieval_residual_min_source_progress": 0.0,
193
+ "retrieval_residual_min_source_advantage": -1000000000.0,
194
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
195
+ "retrieval_residual_source_score_bonus_scale": 0.0,
196
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
197
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
198
+ "retrieval_residual_scale": 1.0,
199
+ "retrieval_residual_scales": [
200
+ 0.35,
201
+ 0.4,
202
+ 0.45
203
+ ],
204
+ "retrieval_residual_anchor": "expert",
205
+ "retrieval_residual_direction": "candidate_minus_anchor",
206
+ "retrieval_residual_reduce": "compose_mean_by_type",
207
+ "retrieval_residual_challenger_types": [
208
+ "residual_near_miss",
209
+ "residual_wrong_gripper"
210
+ ],
211
+ "retrieval_residual_challenger_scales": [],
212
+ "retrieval_residual_challenger_margin": 0.01,
213
+ "retrieval_residual_challenger_type_margins": {},
214
+ "retrieval_residual_challenger_tasks": [],
215
+ "retrieval_residual_challenger_type_tasks": {
216
+ "residual_wrong_gripper": [
217
+ "PickCube-v1",
218
+ "PullCube-v1",
219
+ "StackCube-v1"
220
+ ]
221
+ },
222
+ "lattice_exclude_types": [
223
+ "residual_random_negative",
224
+ "residual_wrong_direction",
225
+ "residual_near_miss+residual_no_op"
226
+ ],
227
+ "lattice_exclude_type_tasks": {},
228
+ "candidate_type_bonuses": {
229
+ "retrieval_residual_no_op": 0.03
230
+ },
231
+ "candidate_type_bonus_components": false,
232
+ "candidate_oracle_rollouts": 0,
233
+ "candidate_oracle_unique_tolerance": 1e-06,
234
+ "candidate_oracle_success_rate": null,
235
+ "candidate_oracle_progress": null,
236
+ "candidate_oracle_score": null,
237
+ "candidate_oracle_regret": null,
238
+ "candidate_oracle_score_gain_over_selected": null,
239
+ "candidate_oracle_unique_count": null,
240
+ "candidate_oracle_improvement_rate": null,
241
+ "candidate_oracle_selected_branch_success_rate": null,
242
+ "candidate_oracle_selected_branch_progress": null,
243
+ "candidate_oracle_type_counts": {},
244
+ "candidate_oracle_best_branch_rank": null,
245
+ "candidate_oracle_best_branch_rank_counts": {},
246
+ "candidate_oracle_branch_success_rates": [],
247
+ "candidate_oracle_branch_progress": [],
248
+ "candidate_oracle_branch_score_gains_over_selected": [],
249
+ "selected_residual_scale_counts": {
250
+ "0.35": 511,
251
+ "0.45": 54,
252
+ "0.4": 10
253
+ },
254
+ "policy_rollout_success_rate": 0.34608695652173915,
255
+ "policy_rollout_progress": 0.5718237993347904,
256
+ "oracle_success_rate": 0.8695652173913043,
257
+ "action_mse_to_best": 0.4139976060374275,
258
+ "best_policy_val": {
259
+ "bc_loss": 0.1315989200439718,
260
+ "field_effect_loss": 0.010110370814800262,
261
+ "field_potential_loss": 0.31546536915832096,
262
+ "field_preference_loss": 0.4960637440284093,
263
+ "lattice_edges": 3833.3333333333335,
264
+ "progress_mae": 0.19023226698239645,
265
+ "rank_acc": 0.8255469501018524,
266
+ "rank_loss": 0.4960531195004781,
267
+ "regret_mae": 0.42156532737943864,
268
+ "success_accuracy": 0.8807473679383596,
269
+ "total_loss": 1.561984618504842
270
+ },
271
+ "per_task": {
272
+ "LiftPegUpright-v1": {
273
+ "action_mse_to_best": 0.3281630301287612,
274
+ "expert_success_rate": 0.8584070796460177,
275
+ "num_groups": 113,
276
+ "oracle_success_rate": 0.9380530973451328,
277
+ "policy_expert_regret": 0.987445328615408,
278
+ "policy_oracle_regret": 1.0666473587002374,
279
+ "policy_rollout_progress": 0.6354643670858535,
280
+ "policy_rollout_success_rate": 0.23008849557522124,
281
+ "restore_max_error": 2.384185791015625e-07
282
+ },
283
+ "PickCube-v1": {
284
+ "action_mse_to_best": 0.36477837357265147,
285
+ "expert_success_rate": 0.9402173913043478,
286
+ "num_groups": 184,
287
+ "oracle_success_rate": 0.9456521739130435,
288
+ "policy_expert_regret": 1.08854141729155,
289
+ "policy_oracle_regret": 1.08856542176648,
290
+ "policy_rollout_progress": 0.5675299971506161,
291
+ "policy_rollout_success_rate": 0.2826086956521739,
292
+ "restore_max_error": 2.384185791015625e-07
293
+ },
294
+ "PullCube-v1": {
295
+ "action_mse_to_best": 0.5960272875075278,
296
+ "expert_success_rate": 0.25,
297
+ "num_groups": 76,
298
+ "oracle_success_rate": 0.40789473684210525,
299
+ "policy_expert_regret": 0.27355212223176895,
300
+ "policy_oracle_regret": 0.4974974064940685,
301
+ "policy_rollout_progress": 0.30258720147570495,
302
+ "policy_rollout_success_rate": 0.19736842105263158,
303
+ "restore_max_error": 2.384185791015625e-07
304
+ },
305
+ "PushCube-v1": {
306
+ "action_mse_to_best": 0.3595503710948669,
307
+ "expert_success_rate": 0.8198198198198198,
308
+ "num_groups": 111,
309
+ "oracle_success_rate": 1.0,
310
+ "policy_expert_regret": 0.3128753544779511,
311
+ "policy_oracle_regret": 0.33004292161078064,
312
+ "policy_rollout_progress": 0.8411282495603906,
313
+ "policy_rollout_success_rate": 0.8288288288288288,
314
+ "restore_max_error": 2.3795291781425476e-07
315
+ },
316
+ "StackCube-v1": {
317
+ "action_mse_to_best": 0.53449247568682,
318
+ "expert_success_rate": 0.6923076923076923,
319
+ "num_groups": 91,
320
+ "oracle_success_rate": 0.8571428571428571,
321
+ "policy_expert_regret": 1.1330992091488052,
322
+ "policy_oracle_regret": 1.292761056305288,
323
+ "policy_rollout_progress": 0.3978442708542059,
324
+ "policy_rollout_success_rate": 0.15384615384615385,
325
+ "restore_max_error": 1.9371509552001953e-07
326
+ }
327
+ }
328
+ },
329
+ {
330
+ "seed": 2,
331
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpullstack_challenger0p01.json",
332
+ "num_groups": 575,
333
+ "selection_mode": "retrieval_residual",
334
+ "num_candidates": 48,
335
+ "candidate_sigma": 0.0,
336
+ "proposal_types": [],
337
+ "proposal_lattice_types": [],
338
+ "selection_margin": 0.2,
339
+ "prepend_policy_candidate": false,
340
+ "field_optim_steps": 0,
341
+ "field_optim_step_size": 0.0,
342
+ "field_optim_trust_radius": 0.0,
343
+ "field_optim_l2_penalty": 0.0,
344
+ "retrieval_neighbors": 4,
345
+ "retrieval_metric": "raw",
346
+ "retrieval_type_min_success": 0.0,
347
+ "retrieval_type_success_bonus_scale": 0.0,
348
+ "retrieval_residual_consensus_penalty_scale": 0.0,
349
+ "retrieval_residual_min_source_progress": 0.0,
350
+ "retrieval_residual_min_source_advantage": -1000000000.0,
351
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
352
+ "retrieval_residual_source_score_bonus_scale": 0.0,
353
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
354
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
355
+ "retrieval_residual_scale": 1.0,
356
+ "retrieval_residual_scales": [
357
+ 0.35,
358
+ 0.4,
359
+ 0.45
360
+ ],
361
+ "retrieval_residual_anchor": "expert",
362
+ "retrieval_residual_direction": "candidate_minus_anchor",
363
+ "retrieval_residual_reduce": "compose_mean_by_type",
364
+ "retrieval_residual_challenger_types": [
365
+ "residual_near_miss",
366
+ "residual_wrong_gripper"
367
+ ],
368
+ "retrieval_residual_challenger_scales": [],
369
+ "retrieval_residual_challenger_margin": 0.01,
370
+ "retrieval_residual_challenger_type_margins": {},
371
+ "retrieval_residual_challenger_tasks": [],
372
+ "retrieval_residual_challenger_type_tasks": {
373
+ "residual_wrong_gripper": [
374
+ "PickCube-v1",
375
+ "PullCube-v1",
376
+ "StackCube-v1"
377
+ ]
378
+ },
379
+ "lattice_exclude_types": [
380
+ "residual_random_negative",
381
+ "residual_wrong_direction",
382
+ "residual_near_miss+residual_no_op"
383
+ ],
384
+ "lattice_exclude_type_tasks": {},
385
+ "candidate_type_bonuses": {
386
+ "retrieval_residual_no_op": 0.03
387
+ },
388
+ "candidate_type_bonus_components": false,
389
+ "candidate_oracle_rollouts": 0,
390
+ "candidate_oracle_unique_tolerance": 1e-06,
391
+ "candidate_oracle_success_rate": null,
392
+ "candidate_oracle_progress": null,
393
+ "candidate_oracle_score": null,
394
+ "candidate_oracle_regret": null,
395
+ "candidate_oracle_score_gain_over_selected": null,
396
+ "candidate_oracle_unique_count": null,
397
+ "candidate_oracle_improvement_rate": null,
398
+ "candidate_oracle_selected_branch_success_rate": null,
399
+ "candidate_oracle_selected_branch_progress": null,
400
+ "candidate_oracle_type_counts": {},
401
+ "candidate_oracle_best_branch_rank": null,
402
+ "candidate_oracle_best_branch_rank_counts": {},
403
+ "candidate_oracle_branch_success_rates": [],
404
+ "candidate_oracle_branch_progress": [],
405
+ "candidate_oracle_branch_score_gains_over_selected": [],
406
+ "selected_residual_scale_counts": {
407
+ "0.35": 480,
408
+ "0.45": 72,
409
+ "0.4": 23
410
+ },
411
+ "policy_rollout_success_rate": 0.37565217391304345,
412
+ "policy_rollout_progress": 0.5898877114057541,
413
+ "oracle_success_rate": 0.8765217391304347,
414
+ "action_mse_to_best": 0.45088212622491564,
415
+ "best_policy_val": {
416
+ "bc_loss": 0.11367896075050037,
417
+ "field_effect_loss": 0.009670218582161598,
418
+ "field_potential_loss": 0.2641640139950646,
419
+ "field_preference_loss": 0.5130490180518892,
420
+ "lattice_edges": 3833.3333333333335,
421
+ "progress_mae": 0.2021729110015763,
422
+ "rank_acc": 0.8333857821093665,
423
+ "rank_loss": 0.5130119257503085,
424
+ "regret_mae": 0.3958987047274907,
425
+ "success_accuracy": 0.8680730561415354,
426
+ "total_loss": 1.4394984311527677
427
+ },
428
+ "per_task": {
429
+ "LiftPegUpright-v1": {
430
+ "action_mse_to_best": 0.35602021238203935,
431
+ "expert_success_rate": 0.8229166666666666,
432
+ "num_groups": 96,
433
+ "oracle_success_rate": 0.9270833333333334,
434
+ "policy_expert_regret": 0.8476849501021206,
435
+ "policy_oracle_regret": 0.9539546399998168,
436
+ "policy_rollout_progress": 0.6463205499264101,
437
+ "policy_rollout_success_rate": 0.3333333333333333,
438
+ "restore_max_error": 1.955777406692505e-07
439
+ },
440
+ "PickCube-v1": {
441
+ "action_mse_to_best": 0.39817221942498826,
442
+ "expert_success_rate": 0.9444444444444444,
443
+ "num_groups": 198,
444
+ "oracle_success_rate": 0.9595959595959596,
445
+ "policy_expert_regret": 0.9662310388127361,
446
+ "policy_oracle_regret": 0.9733144893414445,
447
+ "policy_rollout_progress": 0.6401566648197294,
448
+ "policy_rollout_success_rate": 0.3484848484848485,
449
+ "restore_max_error": 2.384185791015625e-07
450
+ },
451
+ "PullCube-v1": {
452
+ "action_mse_to_best": 0.6421529037257035,
453
+ "expert_success_rate": 0.24444444444444444,
454
+ "num_groups": 90,
455
+ "oracle_success_rate": 0.4666666666666667,
456
+ "policy_expert_regret": 0.2687155395539271,
457
+ "policy_oracle_regret": 0.45486517548561095,
458
+ "policy_rollout_progress": 0.36988404384917684,
459
+ "policy_rollout_success_rate": 0.26666666666666666,
460
+ "restore_max_error": 2.384185791015625e-07
461
+ },
462
+ "PushCube-v1": {
463
+ "action_mse_to_best": 0.397049247997232,
464
+ "expert_success_rate": 0.8514851485148515,
465
+ "num_groups": 101,
466
+ "oracle_success_rate": 1.0,
467
+ "policy_expert_regret": 0.3614020819711213,
468
+ "policy_oracle_regret": 0.41630433337523204,
469
+ "policy_rollout_progress": 0.8015174488029858,
470
+ "policy_rollout_success_rate": 0.7821782178217822,
471
+ "restore_max_error": 2.384185791015625e-07
472
+ },
473
+ "StackCube-v1": {
474
+ "action_mse_to_best": 0.5371716373496586,
475
+ "expert_success_rate": 0.7666666666666667,
476
+ "num_groups": 90,
477
+ "oracle_success_rate": 0.9111111111111111,
478
+ "policy_expert_regret": 1.1891138343347443,
479
+ "policy_oracle_regret": 1.351234385702345,
480
+ "policy_rollout_progress": 0.40160905950599246,
481
+ "policy_rollout_success_rate": 0.13333333333333333,
482
+ "restore_max_error": 2.2351741790771484e-07
483
+ }
484
+ }
485
+ }
486
+ ]
487
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpullstack_challenger0p01_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: `near_miss_policy_bc5`
5
+ Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpullstack_challenger0p01.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: 36.00% +/- 1.49%
11
+ Gain vs h=16 rank checkpoint: +6.26%
12
+ Mean progress: 57.44%
13
+ Mean action MSE to best: 0.421
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 | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper with residual_wrong_gripper=PickCube-v1,PullCube-v1,StackCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 35.83% | 56.14% | 85.74% | n/a | n/a | 0.397 |
18
+ | 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper with residual_wrong_gripper=PickCube-v1,PullCube-v1,StackCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 34.61% | 57.18% | 86.96% | n/a | n/a | 0.414 |
19
+ | 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper with residual_wrong_gripper=PickCube-v1,PullCube-v1,StackCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 37.57% | 58.99% | 87.65% | n/a | n/a | 0.451 |