anhtld commited on
Commit
c81118b
·
verified ·
1 Parent(s): 37f2d7e

Auto-sync: 2026-06-30 21:48:20

Browse files
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_wgmargin0p03_challenger0p01_summary.json ADDED
@@ -0,0 +1,490 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_wgpickpull_wgmargin0p03_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": 0.01593939372158552,
10
+ "mean_progress": 0.5738671411409219,
11
+ "mean_action_mse_to_best": 0.41668552688795374,
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_wgpickpull_wgmargin0p03_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
+ "residual_wrong_gripper": 0.03
58
+ },
59
+ "retrieval_residual_challenger_tasks": [],
60
+ "retrieval_residual_challenger_type_tasks": {
61
+ "residual_wrong_gripper": [
62
+ "PickCube-v1",
63
+ "PullCube-v1"
64
+ ]
65
+ },
66
+ "lattice_exclude_types": [
67
+ "residual_random_negative",
68
+ "residual_wrong_direction",
69
+ "residual_near_miss+residual_no_op"
70
+ ],
71
+ "lattice_exclude_type_tasks": {},
72
+ "candidate_type_bonuses": {
73
+ "retrieval_residual_no_op": 0.03
74
+ },
75
+ "candidate_type_bonus_components": false,
76
+ "candidate_oracle_rollouts": 0,
77
+ "candidate_oracle_unique_tolerance": 1e-06,
78
+ "candidate_oracle_success_rate": null,
79
+ "candidate_oracle_progress": null,
80
+ "candidate_oracle_score": null,
81
+ "candidate_oracle_regret": null,
82
+ "candidate_oracle_score_gain_over_selected": null,
83
+ "candidate_oracle_unique_count": null,
84
+ "candidate_oracle_improvement_rate": null,
85
+ "candidate_oracle_selected_branch_success_rate": null,
86
+ "candidate_oracle_selected_branch_progress": null,
87
+ "candidate_oracle_type_counts": {},
88
+ "candidate_oracle_best_branch_rank": null,
89
+ "candidate_oracle_best_branch_rank_counts": {},
90
+ "candidate_oracle_branch_success_rates": [],
91
+ "candidate_oracle_branch_progress": [],
92
+ "candidate_oracle_branch_score_gains_over_selected": [],
93
+ "selected_residual_scale_counts": {
94
+ "0.35": 514,
95
+ "0.45": 55,
96
+ "0.4": 6
97
+ },
98
+ "policy_rollout_success_rate": 0.3565217391304348,
99
+ "policy_rollout_progress": 0.5592340455844027,
100
+ "oracle_success_rate": 0.8573913043478261,
101
+ "action_mse_to_best": 0.39585299389679796,
102
+ "best_policy_val": {
103
+ "bc_loss": 0.13721593966086706,
104
+ "field_effect_loss": 0.009290305380192068,
105
+ "field_potential_loss": 0.2666468388504452,
106
+ "field_preference_loss": 0.5130573478009965,
107
+ "lattice_edges": 3833.3333333333335,
108
+ "progress_mae": 0.1933159919248687,
109
+ "rank_acc": 0.8265031774838766,
110
+ "rank_loss": 0.5130523675017886,
111
+ "regret_mae": 0.3756548762321472,
112
+ "success_accuracy": 0.8773836526605818,
113
+ "total_loss": 1.5581054819954767
114
+ },
115
+ "per_task": {
116
+ "LiftPegUpright-v1": {
117
+ "action_mse_to_best": 0.32276735697709713,
118
+ "expert_success_rate": 0.8865979381443299,
119
+ "num_groups": 97,
120
+ "oracle_success_rate": 0.9278350515463918,
121
+ "policy_expert_regret": 1.0783334703175063,
122
+ "policy_oracle_regret": 1.1047489480259491,
123
+ "policy_rollout_progress": 0.5714634232914325,
124
+ "policy_rollout_success_rate": 0.2268041237113402,
125
+ "restore_max_error": 2.384185791015625e-07
126
+ },
127
+ "PickCube-v1": {
128
+ "action_mse_to_best": 0.3055856230382163,
129
+ "expert_success_rate": 0.9375,
130
+ "num_groups": 208,
131
+ "oracle_success_rate": 0.9471153846153846,
132
+ "policy_expert_regret": 0.9405335786465842,
133
+ "policy_oracle_regret": 0.9539564460730896,
134
+ "policy_rollout_progress": 0.622309988257117,
135
+ "policy_rollout_success_rate": 0.3557692307692308,
136
+ "restore_max_error": 2.384185791015625e-07
137
+ },
138
+ "PullCube-v1": {
139
+ "action_mse_to_best": 0.6027194387339926,
140
+ "expert_success_rate": 0.19480519480519481,
141
+ "num_groups": 77,
142
+ "oracle_success_rate": 0.36363636363636365,
143
+ "policy_expert_regret": 0.2471909495977709,
144
+ "policy_oracle_regret": 0.4206590882091263,
145
+ "policy_rollout_progress": 0.31901094595018725,
146
+ "policy_rollout_success_rate": 0.22077922077922077,
147
+ "restore_max_error": 2.3096799850463867e-07
148
+ },
149
+ "PushCube-v1": {
150
+ "action_mse_to_best": 0.3675119048578942,
151
+ "expert_success_rate": 0.8617021276595744,
152
+ "num_groups": 94,
153
+ "oracle_success_rate": 0.9893617021276596,
154
+ "policy_expert_regret": 0.5508309629686335,
155
+ "policy_oracle_regret": 0.628381609441118,
156
+ "policy_rollout_progress": 0.7014056246014352,
157
+ "policy_rollout_success_rate": 0.6702127659574468,
158
+ "restore_max_error": 2.384185791015625e-07
159
+ },
160
+ "StackCube-v1": {
161
+ "action_mse_to_best": 0.5231284084118376,
162
+ "expert_success_rate": 0.6767676767676768,
163
+ "num_groups": 99,
164
+ "oracle_success_rate": 0.8585858585858586,
165
+ "policy_expert_regret": 0.9252789386294105,
166
+ "policy_oracle_regret": 1.0729591972900159,
167
+ "policy_rollout_progress": 0.46657752569275673,
168
+ "policy_rollout_success_rate": 0.29292929292929293,
169
+ "restore_max_error": 2.384185791015625e-07
170
+ }
171
+ }
172
+ },
173
+ {
174
+ "seed": 1,
175
+ "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_wgpickpull_wgmargin0p03_challenger0p01.json",
176
+ "num_groups": 575,
177
+ "selection_mode": "retrieval_residual",
178
+ "num_candidates": 48,
179
+ "candidate_sigma": 0.0,
180
+ "proposal_types": [],
181
+ "proposal_lattice_types": [],
182
+ "selection_margin": 0.2,
183
+ "prepend_policy_candidate": false,
184
+ "field_optim_steps": 0,
185
+ "field_optim_step_size": 0.0,
186
+ "field_optim_trust_radius": 0.0,
187
+ "field_optim_l2_penalty": 0.0,
188
+ "retrieval_neighbors": 4,
189
+ "retrieval_metric": "raw",
190
+ "retrieval_type_min_success": 0.0,
191
+ "retrieval_type_success_bonus_scale": 0.0,
192
+ "retrieval_residual_consensus_penalty_scale": 0.0,
193
+ "retrieval_residual_min_source_progress": 0.0,
194
+ "retrieval_residual_min_source_advantage": -1000000000.0,
195
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
196
+ "retrieval_residual_source_score_bonus_scale": 0.0,
197
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
198
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
199
+ "retrieval_residual_scale": 1.0,
200
+ "retrieval_residual_scales": [
201
+ 0.35,
202
+ 0.4,
203
+ 0.45
204
+ ],
205
+ "retrieval_residual_anchor": "expert",
206
+ "retrieval_residual_direction": "candidate_minus_anchor",
207
+ "retrieval_residual_reduce": "compose_mean_by_type",
208
+ "retrieval_residual_challenger_types": [
209
+ "residual_near_miss",
210
+ "residual_wrong_gripper"
211
+ ],
212
+ "retrieval_residual_challenger_scales": [],
213
+ "retrieval_residual_challenger_margin": 0.01,
214
+ "retrieval_residual_challenger_type_margins": {
215
+ "residual_wrong_gripper": 0.03
216
+ },
217
+ "retrieval_residual_challenger_tasks": [],
218
+ "retrieval_residual_challenger_type_tasks": {
219
+ "residual_wrong_gripper": [
220
+ "PickCube-v1",
221
+ "PullCube-v1"
222
+ ]
223
+ },
224
+ "lattice_exclude_types": [
225
+ "residual_random_negative",
226
+ "residual_wrong_direction",
227
+ "residual_near_miss+residual_no_op"
228
+ ],
229
+ "lattice_exclude_type_tasks": {},
230
+ "candidate_type_bonuses": {
231
+ "retrieval_residual_no_op": 0.03
232
+ },
233
+ "candidate_type_bonus_components": false,
234
+ "candidate_oracle_rollouts": 0,
235
+ "candidate_oracle_unique_tolerance": 1e-06,
236
+ "candidate_oracle_success_rate": null,
237
+ "candidate_oracle_progress": null,
238
+ "candidate_oracle_score": null,
239
+ "candidate_oracle_regret": null,
240
+ "candidate_oracle_score_gain_over_selected": null,
241
+ "candidate_oracle_unique_count": null,
242
+ "candidate_oracle_improvement_rate": null,
243
+ "candidate_oracle_selected_branch_success_rate": null,
244
+ "candidate_oracle_selected_branch_progress": null,
245
+ "candidate_oracle_type_counts": {},
246
+ "candidate_oracle_best_branch_rank": null,
247
+ "candidate_oracle_best_branch_rank_counts": {},
248
+ "candidate_oracle_branch_success_rates": [],
249
+ "candidate_oracle_branch_progress": [],
250
+ "candidate_oracle_branch_score_gains_over_selected": [],
251
+ "selected_residual_scale_counts": {
252
+ "0.35": 522,
253
+ "0.45": 47,
254
+ "0.4": 6
255
+ },
256
+ "policy_rollout_success_rate": 0.34608695652173915,
257
+ "policy_rollout_progress": 0.5719369965759309,
258
+ "oracle_success_rate": 0.8695652173913043,
259
+ "action_mse_to_best": 0.40845387818858675,
260
+ "best_policy_val": {
261
+ "bc_loss": 0.1315989200439718,
262
+ "field_effect_loss": 0.010110370814800262,
263
+ "field_potential_loss": 0.31546536915832096,
264
+ "field_preference_loss": 0.4960637440284093,
265
+ "lattice_edges": 3833.3333333333335,
266
+ "progress_mae": 0.19023226698239645,
267
+ "rank_acc": 0.8255469501018524,
268
+ "rank_loss": 0.4960531195004781,
269
+ "regret_mae": 0.42156532737943864,
270
+ "success_accuracy": 0.8807473679383596,
271
+ "total_loss": 1.561984618504842
272
+ },
273
+ "per_task": {
274
+ "LiftPegUpright-v1": {
275
+ "action_mse_to_best": 0.3281630301287612,
276
+ "expert_success_rate": 0.8584070796460177,
277
+ "num_groups": 113,
278
+ "oracle_success_rate": 0.9380530973451328,
279
+ "policy_expert_regret": 0.987445328615408,
280
+ "policy_oracle_regret": 1.0666473587002374,
281
+ "policy_rollout_progress": 0.6354643670858535,
282
+ "policy_rollout_success_rate": 0.23008849557522124,
283
+ "restore_max_error": 2.384185791015625e-07
284
+ },
285
+ "PickCube-v1": {
286
+ "action_mse_to_best": 0.35166226872040524,
287
+ "expert_success_rate": 0.9402173913043478,
288
+ "num_groups": 184,
289
+ "oracle_success_rate": 0.9456521739130435,
290
+ "policy_expert_regret": 1.0885420930268161,
291
+ "policy_oracle_regret": 1.0885660975017464,
292
+ "policy_rollout_progress": 0.5675293214153498,
293
+ "policy_rollout_success_rate": 0.2826086956521739,
294
+ "restore_max_error": 2.384185791015625e-07
295
+ },
296
+ "PullCube-v1": {
297
+ "action_mse_to_best": 0.594962232501099,
298
+ "expert_success_rate": 0.25,
299
+ "num_groups": 76,
300
+ "oracle_success_rate": 0.40789473684210525,
301
+ "policy_expert_regret": 0.27355250936785813,
302
+ "policy_oracle_regret": 0.49749779363015767,
303
+ "policy_rollout_progress": 0.30258676218555164,
304
+ "policy_rollout_success_rate": 0.19736842105263158,
305
+ "restore_max_error": 2.384185791015625e-07
306
+ },
307
+ "PushCube-v1": {
308
+ "action_mse_to_best": 0.3595503710948669,
309
+ "expert_success_rate": 0.8198198198198198,
310
+ "num_groups": 111,
311
+ "oracle_success_rate": 1.0,
312
+ "policy_expert_regret": 0.3128753544779511,
313
+ "policy_oracle_regret": 0.33004292161078064,
314
+ "policy_rollout_progress": 0.8411282495603906,
315
+ "policy_rollout_success_rate": 0.8288288288288288,
316
+ "restore_max_error": 2.3795291781425476e-07
317
+ },
318
+ "StackCube-v1": {
319
+ "action_mse_to_best": 0.526873398326583,
320
+ "expert_success_rate": 0.6923076923076923,
321
+ "num_groups": 91,
322
+ "oracle_success_rate": 0.8571428571428571,
323
+ "policy_expert_regret": 1.132382218654339,
324
+ "policy_oracle_regret": 1.2920440658108219,
325
+ "policy_rollout_progress": 0.398561261348672,
326
+ "policy_rollout_success_rate": 0.15384615384615385,
327
+ "restore_max_error": 1.9371509552001953e-07
328
+ }
329
+ }
330
+ },
331
+ {
332
+ "seed": 2,
333
+ "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_wgpickpull_wgmargin0p03_challenger0p01.json",
334
+ "num_groups": 575,
335
+ "selection_mode": "retrieval_residual",
336
+ "num_candidates": 48,
337
+ "candidate_sigma": 0.0,
338
+ "proposal_types": [],
339
+ "proposal_lattice_types": [],
340
+ "selection_margin": 0.2,
341
+ "prepend_policy_candidate": false,
342
+ "field_optim_steps": 0,
343
+ "field_optim_step_size": 0.0,
344
+ "field_optim_trust_radius": 0.0,
345
+ "field_optim_l2_penalty": 0.0,
346
+ "retrieval_neighbors": 4,
347
+ "retrieval_metric": "raw",
348
+ "retrieval_type_min_success": 0.0,
349
+ "retrieval_type_success_bonus_scale": 0.0,
350
+ "retrieval_residual_consensus_penalty_scale": 0.0,
351
+ "retrieval_residual_min_source_progress": 0.0,
352
+ "retrieval_residual_min_source_advantage": -1000000000.0,
353
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
354
+ "retrieval_residual_source_score_bonus_scale": 0.0,
355
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
356
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
357
+ "retrieval_residual_scale": 1.0,
358
+ "retrieval_residual_scales": [
359
+ 0.35,
360
+ 0.4,
361
+ 0.45
362
+ ],
363
+ "retrieval_residual_anchor": "expert",
364
+ "retrieval_residual_direction": "candidate_minus_anchor",
365
+ "retrieval_residual_reduce": "compose_mean_by_type",
366
+ "retrieval_residual_challenger_types": [
367
+ "residual_near_miss",
368
+ "residual_wrong_gripper"
369
+ ],
370
+ "retrieval_residual_challenger_scales": [],
371
+ "retrieval_residual_challenger_margin": 0.01,
372
+ "retrieval_residual_challenger_type_margins": {
373
+ "residual_wrong_gripper": 0.03
374
+ },
375
+ "retrieval_residual_challenger_tasks": [],
376
+ "retrieval_residual_challenger_type_tasks": {
377
+ "residual_wrong_gripper": [
378
+ "PickCube-v1",
379
+ "PullCube-v1"
380
+ ]
381
+ },
382
+ "lattice_exclude_types": [
383
+ "residual_random_negative",
384
+ "residual_wrong_direction",
385
+ "residual_near_miss+residual_no_op"
386
+ ],
387
+ "lattice_exclude_type_tasks": {},
388
+ "candidate_type_bonuses": {
389
+ "retrieval_residual_no_op": 0.03
390
+ },
391
+ "candidate_type_bonus_components": false,
392
+ "candidate_oracle_rollouts": 0,
393
+ "candidate_oracle_unique_tolerance": 1e-06,
394
+ "candidate_oracle_success_rate": null,
395
+ "candidate_oracle_progress": null,
396
+ "candidate_oracle_score": null,
397
+ "candidate_oracle_regret": null,
398
+ "candidate_oracle_score_gain_over_selected": null,
399
+ "candidate_oracle_unique_count": null,
400
+ "candidate_oracle_improvement_rate": null,
401
+ "candidate_oracle_selected_branch_success_rate": null,
402
+ "candidate_oracle_selected_branch_progress": null,
403
+ "candidate_oracle_type_counts": {},
404
+ "candidate_oracle_best_branch_rank": null,
405
+ "candidate_oracle_best_branch_rank_counts": {},
406
+ "candidate_oracle_branch_success_rates": [],
407
+ "candidate_oracle_branch_progress": [],
408
+ "candidate_oracle_branch_score_gains_over_selected": [],
409
+ "selected_residual_scale_counts": {
410
+ "0.35": 494,
411
+ "0.45": 64,
412
+ "0.4": 17
413
+ },
414
+ "policy_rollout_success_rate": 0.3773913043478261,
415
+ "policy_rollout_progress": 0.590430381262432,
416
+ "oracle_success_rate": 0.8765217391304347,
417
+ "action_mse_to_best": 0.44574970857847646,
418
+ "best_policy_val": {
419
+ "bc_loss": 0.11367896075050037,
420
+ "field_effect_loss": 0.009670218582161598,
421
+ "field_potential_loss": 0.2641640139950646,
422
+ "field_preference_loss": 0.5130490180518892,
423
+ "lattice_edges": 3833.3333333333335,
424
+ "progress_mae": 0.2021729110015763,
425
+ "rank_acc": 0.8333857821093665,
426
+ "rank_loss": 0.5130119257503085,
427
+ "regret_mae": 0.3958987047274907,
428
+ "success_accuracy": 0.8680730561415354,
429
+ "total_loss": 1.4394984311527677
430
+ },
431
+ "per_task": {
432
+ "LiftPegUpright-v1": {
433
+ "action_mse_to_best": 0.35602021238203935,
434
+ "expert_success_rate": 0.8229166666666666,
435
+ "num_groups": 96,
436
+ "oracle_success_rate": 0.9270833333333334,
437
+ "policy_expert_regret": 0.8476849501021206,
438
+ "policy_oracle_regret": 0.9539546399998168,
439
+ "policy_rollout_progress": 0.6463205499264101,
440
+ "policy_rollout_success_rate": 0.3333333333333333,
441
+ "restore_max_error": 1.955777406692505e-07
442
+ },
443
+ "PickCube-v1": {
444
+ "action_mse_to_best": 0.3902869703756137,
445
+ "expert_success_rate": 0.9444444444444444,
446
+ "num_groups": 198,
447
+ "oracle_success_rate": 0.9595959595959596,
448
+ "policy_expert_regret": 0.9662351405455006,
449
+ "policy_oracle_regret": 0.973318591074209,
450
+ "policy_rollout_progress": 0.6401525630869649,
451
+ "policy_rollout_success_rate": 0.3484848484848485,
452
+ "restore_max_error": 2.384185791015625e-07
453
+ },
454
+ "PullCube-v1": {
455
+ "action_mse_to_best": 0.6412372935977247,
456
+ "expert_success_rate": 0.24444444444444444,
457
+ "num_groups": 90,
458
+ "oracle_success_rate": 0.4666666666666667,
459
+ "policy_expert_regret": 0.2687283302760787,
460
+ "policy_oracle_regret": 0.45487796620776255,
461
+ "policy_rollout_progress": 0.37023526067949003,
462
+ "policy_rollout_success_rate": 0.26666666666666666,
463
+ "restore_max_error": 2.384185791015625e-07
464
+ },
465
+ "PushCube-v1": {
466
+ "action_mse_to_best": 0.397049247997232,
467
+ "expert_success_rate": 0.8514851485148515,
468
+ "num_groups": 101,
469
+ "oracle_success_rate": 1.0,
470
+ "policy_expert_regret": 0.3614020819711213,
471
+ "policy_oracle_regret": 0.41630433337523204,
472
+ "policy_rollout_progress": 0.8015174488029858,
473
+ "policy_rollout_success_rate": 0.7821782178217822,
474
+ "restore_max_error": 2.384185791015625e-07
475
+ },
476
+ "StackCube-v1": {
477
+ "action_mse_to_best": 0.5226443493117888,
478
+ "expert_success_rate": 0.7666666666666667,
479
+ "num_groups": 90,
480
+ "oracle_success_rate": 0.9111111111111111,
481
+ "policy_expert_regret": 1.1748778588242001,
482
+ "policy_oracle_regret": 1.3369984101918009,
483
+ "policy_rollout_progress": 0.4047339239054256,
484
+ "policy_rollout_success_rate": 0.14444444444444443,
485
+ "restore_max_error": 2.2351741790771484e-07
486
+ }
487
+ }
488
+ }
489
+ ]
490
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_wgmargin0p03_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_wgpickpull_wgmargin0p03_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% +/- 1.59%
11
+ Gain vs h=16 rank checkpoint: +6.26%
12
+ Mean progress: 57.39%
13
+ Mean action MSE to best: 0.417
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 margins residual_wrong_gripper=0.03@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 | 35.65% | 55.92% | 85.74% | n/a | n/a | 0.396 |
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 margins residual_wrong_gripper=0.03@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% | 57.19% | 86.96% | n/a | n/a | 0.408 |
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 margins residual_wrong_gripper=0.03@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 | 37.74% | 59.04% | 87.65% | n/a | n/a | 0.446 |
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_wgmargin0p05_challenger0p01_summary.json ADDED
@@ -0,0 +1,490 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_wgpickpull_wgmargin0p05_challenger0p01.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.35884057971014494,
9
+ "std_success": 0.015780472566743406,
10
+ "mean_progress": 0.5730632333303599,
11
+ "mean_action_mse_to_best": 0.41403828773242624,
12
+ "gain_vs_h4": 0.062140579710144916,
13
+ "gain_vs_h16_rank_checkpoint": 0.06144927536231887,
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_wgpickpull_wgmargin0p05_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
+ "residual_wrong_gripper": 0.05
58
+ },
59
+ "retrieval_residual_challenger_tasks": [],
60
+ "retrieval_residual_challenger_type_tasks": {
61
+ "residual_wrong_gripper": [
62
+ "PickCube-v1",
63
+ "PullCube-v1"
64
+ ]
65
+ },
66
+ "lattice_exclude_types": [
67
+ "residual_random_negative",
68
+ "residual_wrong_direction",
69
+ "residual_near_miss+residual_no_op"
70
+ ],
71
+ "lattice_exclude_type_tasks": {},
72
+ "candidate_type_bonuses": {
73
+ "retrieval_residual_no_op": 0.03
74
+ },
75
+ "candidate_type_bonus_components": false,
76
+ "candidate_oracle_rollouts": 0,
77
+ "candidate_oracle_unique_tolerance": 1e-06,
78
+ "candidate_oracle_success_rate": null,
79
+ "candidate_oracle_progress": null,
80
+ "candidate_oracle_score": null,
81
+ "candidate_oracle_regret": null,
82
+ "candidate_oracle_score_gain_over_selected": null,
83
+ "candidate_oracle_unique_count": null,
84
+ "candidate_oracle_improvement_rate": null,
85
+ "candidate_oracle_selected_branch_success_rate": null,
86
+ "candidate_oracle_selected_branch_progress": null,
87
+ "candidate_oracle_type_counts": {},
88
+ "candidate_oracle_best_branch_rank": null,
89
+ "candidate_oracle_best_branch_rank_counts": {},
90
+ "candidate_oracle_branch_success_rates": [],
91
+ "candidate_oracle_branch_progress": [],
92
+ "candidate_oracle_branch_score_gains_over_selected": [],
93
+ "selected_residual_scale_counts": {
94
+ "0.35": 518,
95
+ "0.45": 52,
96
+ "0.4": 5
97
+ },
98
+ "policy_rollout_success_rate": 0.3565217391304348,
99
+ "policy_rollout_progress": 0.5592912306045384,
100
+ "oracle_success_rate": 0.8573913043478261,
101
+ "action_mse_to_best": 0.3946219221810284,
102
+ "best_policy_val": {
103
+ "bc_loss": 0.13721593966086706,
104
+ "field_effect_loss": 0.009290305380192068,
105
+ "field_potential_loss": 0.2666468388504452,
106
+ "field_preference_loss": 0.5130573478009965,
107
+ "lattice_edges": 3833.3333333333335,
108
+ "progress_mae": 0.1933159919248687,
109
+ "rank_acc": 0.8265031774838766,
110
+ "rank_loss": 0.5130523675017886,
111
+ "regret_mae": 0.3756548762321472,
112
+ "success_accuracy": 0.8773836526605818,
113
+ "total_loss": 1.5581054819954767
114
+ },
115
+ "per_task": {
116
+ "LiftPegUpright-v1": {
117
+ "action_mse_to_best": 0.32276735697709713,
118
+ "expert_success_rate": 0.8865979381443299,
119
+ "num_groups": 97,
120
+ "oracle_success_rate": 0.9278350515463918,
121
+ "policy_expert_regret": 1.0783334703175063,
122
+ "policy_oracle_regret": 1.1047489480259491,
123
+ "policy_rollout_progress": 0.5714634232914325,
124
+ "policy_rollout_success_rate": 0.2268041237113402,
125
+ "restore_max_error": 2.384185791015625e-07
126
+ },
127
+ "PickCube-v1": {
128
+ "action_mse_to_best": 0.3026962882409302,
129
+ "expert_success_rate": 0.9375,
130
+ "num_groups": 208,
131
+ "oracle_success_rate": 0.9471153846153846,
132
+ "policy_expert_regret": 0.9403875692246052,
133
+ "policy_oracle_regret": 0.9538104366511106,
134
+ "policy_rollout_progress": 0.622455997679096,
135
+ "policy_rollout_success_rate": 0.3557692307692308,
136
+ "restore_max_error": 2.384185791015625e-07
137
+ },
138
+ "PullCube-v1": {
139
+ "action_mse_to_best": 0.6013313270621485,
140
+ "expert_success_rate": 0.19480519480519481,
141
+ "num_groups": 77,
142
+ "oracle_success_rate": 0.36363636363636365,
143
+ "policy_expert_regret": 0.2471449282322708,
144
+ "policy_oracle_regret": 0.420613325388449,
145
+ "policy_rollout_progress": 0.3190435618827386,
146
+ "policy_rollout_success_rate": 0.22077922077922077,
147
+ "restore_max_error": 2.3096799850463867e-07
148
+ },
149
+ "PushCube-v1": {
150
+ "action_mse_to_best": 0.3675119048578942,
151
+ "expert_success_rate": 0.8617021276595744,
152
+ "num_groups": 94,
153
+ "oracle_success_rate": 0.9893617021276596,
154
+ "policy_expert_regret": 0.5508309629686335,
155
+ "policy_oracle_regret": 0.628381609441118,
156
+ "policy_rollout_progress": 0.7014056246014352,
157
+ "policy_rollout_success_rate": 0.6702127659574468,
158
+ "restore_max_error": 2.384185791015625e-07
159
+ },
160
+ "StackCube-v1": {
161
+ "action_mse_to_best": 0.5231284084118376,
162
+ "expert_success_rate": 0.6767676767676768,
163
+ "num_groups": 99,
164
+ "oracle_success_rate": 0.8585858585858586,
165
+ "policy_expert_regret": 0.9252789386294105,
166
+ "policy_oracle_regret": 1.0729591972900159,
167
+ "policy_rollout_progress": 0.46657752569275673,
168
+ "policy_rollout_success_rate": 0.29292929292929293,
169
+ "restore_max_error": 2.384185791015625e-07
170
+ }
171
+ }
172
+ },
173
+ {
174
+ "seed": 1,
175
+ "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_wgpickpull_wgmargin0p05_challenger0p01.json",
176
+ "num_groups": 575,
177
+ "selection_mode": "retrieval_residual",
178
+ "num_candidates": 48,
179
+ "candidate_sigma": 0.0,
180
+ "proposal_types": [],
181
+ "proposal_lattice_types": [],
182
+ "selection_margin": 0.2,
183
+ "prepend_policy_candidate": false,
184
+ "field_optim_steps": 0,
185
+ "field_optim_step_size": 0.0,
186
+ "field_optim_trust_radius": 0.0,
187
+ "field_optim_l2_penalty": 0.0,
188
+ "retrieval_neighbors": 4,
189
+ "retrieval_metric": "raw",
190
+ "retrieval_type_min_success": 0.0,
191
+ "retrieval_type_success_bonus_scale": 0.0,
192
+ "retrieval_residual_consensus_penalty_scale": 0.0,
193
+ "retrieval_residual_min_source_progress": 0.0,
194
+ "retrieval_residual_min_source_advantage": -1000000000.0,
195
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
196
+ "retrieval_residual_source_score_bonus_scale": 0.0,
197
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
198
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
199
+ "retrieval_residual_scale": 1.0,
200
+ "retrieval_residual_scales": [
201
+ 0.35,
202
+ 0.4,
203
+ 0.45
204
+ ],
205
+ "retrieval_residual_anchor": "expert",
206
+ "retrieval_residual_direction": "candidate_minus_anchor",
207
+ "retrieval_residual_reduce": "compose_mean_by_type",
208
+ "retrieval_residual_challenger_types": [
209
+ "residual_near_miss",
210
+ "residual_wrong_gripper"
211
+ ],
212
+ "retrieval_residual_challenger_scales": [],
213
+ "retrieval_residual_challenger_margin": 0.01,
214
+ "retrieval_residual_challenger_type_margins": {
215
+ "residual_wrong_gripper": 0.05
216
+ },
217
+ "retrieval_residual_challenger_tasks": [],
218
+ "retrieval_residual_challenger_type_tasks": {
219
+ "residual_wrong_gripper": [
220
+ "PickCube-v1",
221
+ "PullCube-v1"
222
+ ]
223
+ },
224
+ "lattice_exclude_types": [
225
+ "residual_random_negative",
226
+ "residual_wrong_direction",
227
+ "residual_near_miss+residual_no_op"
228
+ ],
229
+ "lattice_exclude_type_tasks": {},
230
+ "candidate_type_bonuses": {
231
+ "retrieval_residual_no_op": 0.03
232
+ },
233
+ "candidate_type_bonus_components": false,
234
+ "candidate_oracle_rollouts": 0,
235
+ "candidate_oracle_unique_tolerance": 1e-06,
236
+ "candidate_oracle_success_rate": null,
237
+ "candidate_oracle_progress": null,
238
+ "candidate_oracle_score": null,
239
+ "candidate_oracle_regret": null,
240
+ "candidate_oracle_score_gain_over_selected": null,
241
+ "candidate_oracle_unique_count": null,
242
+ "candidate_oracle_improvement_rate": null,
243
+ "candidate_oracle_selected_branch_success_rate": null,
244
+ "candidate_oracle_selected_branch_progress": null,
245
+ "candidate_oracle_type_counts": {},
246
+ "candidate_oracle_best_branch_rank": null,
247
+ "candidate_oracle_best_branch_rank_counts": {},
248
+ "candidate_oracle_branch_success_rates": [],
249
+ "candidate_oracle_branch_progress": [],
250
+ "candidate_oracle_branch_score_gains_over_selected": [],
251
+ "selected_residual_scale_counts": {
252
+ "0.35": 524,
253
+ "0.45": 45,
254
+ "0.4": 6
255
+ },
256
+ "policy_rollout_success_rate": 0.3443478260869565,
257
+ "policy_rollout_progress": 0.5702725335210562,
258
+ "oracle_success_rate": 0.8695652173913043,
259
+ "action_mse_to_best": 0.4068148876516067,
260
+ "best_policy_val": {
261
+ "bc_loss": 0.1315989200439718,
262
+ "field_effect_loss": 0.010110370814800262,
263
+ "field_potential_loss": 0.31546536915832096,
264
+ "field_preference_loss": 0.4960637440284093,
265
+ "lattice_edges": 3833.3333333333335,
266
+ "progress_mae": 0.19023226698239645,
267
+ "rank_acc": 0.8255469501018524,
268
+ "rank_loss": 0.4960531195004781,
269
+ "regret_mae": 0.42156532737943864,
270
+ "success_accuracy": 0.8807473679383596,
271
+ "total_loss": 1.561984618504842
272
+ },
273
+ "per_task": {
274
+ "LiftPegUpright-v1": {
275
+ "action_mse_to_best": 0.3281630301287612,
276
+ "expert_success_rate": 0.8584070796460177,
277
+ "num_groups": 113,
278
+ "oracle_success_rate": 0.9380530973451328,
279
+ "policy_expert_regret": 0.987445328615408,
280
+ "policy_oracle_regret": 1.0666473587002374,
281
+ "policy_rollout_progress": 0.6354643670858535,
282
+ "policy_rollout_success_rate": 0.23008849557522124,
283
+ "restore_max_error": 2.384185791015625e-07
284
+ },
285
+ "PickCube-v1": {
286
+ "action_mse_to_best": 0.34687355154639354,
287
+ "expert_success_rate": 0.9402173913043478,
288
+ "num_groups": 184,
289
+ "oracle_success_rate": 0.9456521739130435,
290
+ "policy_expert_regret": 1.0885244157145042,
291
+ "policy_oracle_regret": 1.0885484201894344,
292
+ "policy_rollout_progress": 0.5675469987276618,
293
+ "policy_rollout_success_rate": 0.2826086956521739,
294
+ "restore_max_error": 2.384185791015625e-07
295
+ },
296
+ "PullCube-v1": {
297
+ "action_mse_to_best": 0.594155711464976,
298
+ "expert_success_rate": 0.25,
299
+ "num_groups": 76,
300
+ "oracle_success_rate": 0.40789473684210525,
301
+ "policy_expert_regret": 0.2993461788680993,
302
+ "policy_oracle_regret": 0.5232914631303988,
303
+ "policy_rollout_progress": 0.2899509874221526,
304
+ "policy_rollout_success_rate": 0.18421052631578946,
305
+ "restore_max_error": 2.384185791015625e-07
306
+ },
307
+ "PushCube-v1": {
308
+ "action_mse_to_best": 0.3595503710948669,
309
+ "expert_success_rate": 0.8198198198198198,
310
+ "num_groups": 111,
311
+ "oracle_success_rate": 1.0,
312
+ "policy_expert_regret": 0.3128753544779511,
313
+ "policy_oracle_regret": 0.33004292161078064,
314
+ "policy_rollout_progress": 0.8411282495603906,
315
+ "policy_rollout_success_rate": 0.8288288288288288,
316
+ "restore_max_error": 2.3795291781425476e-07
317
+ },
318
+ "StackCube-v1": {
319
+ "action_mse_to_best": 0.526873398326583,
320
+ "expert_success_rate": 0.6923076923076923,
321
+ "num_groups": 91,
322
+ "oracle_success_rate": 0.8571428571428571,
323
+ "policy_expert_regret": 1.132382218654339,
324
+ "policy_oracle_regret": 1.2920440658108219,
325
+ "policy_rollout_progress": 0.398561261348672,
326
+ "policy_rollout_success_rate": 0.15384615384615385,
327
+ "restore_max_error": 1.9371509552001953e-07
328
+ }
329
+ }
330
+ },
331
+ {
332
+ "seed": 2,
333
+ "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_wgpickpull_wgmargin0p05_challenger0p01.json",
334
+ "num_groups": 575,
335
+ "selection_mode": "retrieval_residual",
336
+ "num_candidates": 48,
337
+ "candidate_sigma": 0.0,
338
+ "proposal_types": [],
339
+ "proposal_lattice_types": [],
340
+ "selection_margin": 0.2,
341
+ "prepend_policy_candidate": false,
342
+ "field_optim_steps": 0,
343
+ "field_optim_step_size": 0.0,
344
+ "field_optim_trust_radius": 0.0,
345
+ "field_optim_l2_penalty": 0.0,
346
+ "retrieval_neighbors": 4,
347
+ "retrieval_metric": "raw",
348
+ "retrieval_type_min_success": 0.0,
349
+ "retrieval_type_success_bonus_scale": 0.0,
350
+ "retrieval_residual_consensus_penalty_scale": 0.0,
351
+ "retrieval_residual_min_source_progress": 0.0,
352
+ "retrieval_residual_min_source_advantage": -1000000000.0,
353
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
354
+ "retrieval_residual_source_score_bonus_scale": 0.0,
355
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
356
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
357
+ "retrieval_residual_scale": 1.0,
358
+ "retrieval_residual_scales": [
359
+ 0.35,
360
+ 0.4,
361
+ 0.45
362
+ ],
363
+ "retrieval_residual_anchor": "expert",
364
+ "retrieval_residual_direction": "candidate_minus_anchor",
365
+ "retrieval_residual_reduce": "compose_mean_by_type",
366
+ "retrieval_residual_challenger_types": [
367
+ "residual_near_miss",
368
+ "residual_wrong_gripper"
369
+ ],
370
+ "retrieval_residual_challenger_scales": [],
371
+ "retrieval_residual_challenger_margin": 0.01,
372
+ "retrieval_residual_challenger_type_margins": {
373
+ "residual_wrong_gripper": 0.05
374
+ },
375
+ "retrieval_residual_challenger_tasks": [],
376
+ "retrieval_residual_challenger_type_tasks": {
377
+ "residual_wrong_gripper": [
378
+ "PickCube-v1",
379
+ "PullCube-v1"
380
+ ]
381
+ },
382
+ "lattice_exclude_types": [
383
+ "residual_random_negative",
384
+ "residual_wrong_direction",
385
+ "residual_near_miss+residual_no_op"
386
+ ],
387
+ "lattice_exclude_type_tasks": {},
388
+ "candidate_type_bonuses": {
389
+ "retrieval_residual_no_op": 0.03
390
+ },
391
+ "candidate_type_bonus_components": false,
392
+ "candidate_oracle_rollouts": 0,
393
+ "candidate_oracle_unique_tolerance": 1e-06,
394
+ "candidate_oracle_success_rate": null,
395
+ "candidate_oracle_progress": null,
396
+ "candidate_oracle_score": null,
397
+ "candidate_oracle_regret": null,
398
+ "candidate_oracle_score_gain_over_selected": null,
399
+ "candidate_oracle_unique_count": null,
400
+ "candidate_oracle_improvement_rate": null,
401
+ "candidate_oracle_selected_branch_success_rate": null,
402
+ "candidate_oracle_selected_branch_progress": null,
403
+ "candidate_oracle_type_counts": {},
404
+ "candidate_oracle_best_branch_rank": null,
405
+ "candidate_oracle_best_branch_rank_counts": {},
406
+ "candidate_oracle_branch_success_rates": [],
407
+ "candidate_oracle_branch_progress": [],
408
+ "candidate_oracle_branch_score_gains_over_selected": [],
409
+ "selected_residual_scale_counts": {
410
+ "0.35": 508,
411
+ "0.45": 52,
412
+ "0.4": 15
413
+ },
414
+ "policy_rollout_success_rate": 0.37565217391304345,
415
+ "policy_rollout_progress": 0.5896259358654852,
416
+ "oracle_success_rate": 0.8765217391304347,
417
+ "action_mse_to_best": 0.4406780533646436,
418
+ "best_policy_val": {
419
+ "bc_loss": 0.11367896075050037,
420
+ "field_effect_loss": 0.009670218582161598,
421
+ "field_potential_loss": 0.2641640139950646,
422
+ "field_preference_loss": 0.5130490180518892,
423
+ "lattice_edges": 3833.3333333333335,
424
+ "progress_mae": 0.2021729110015763,
425
+ "rank_acc": 0.8333857821093665,
426
+ "rank_loss": 0.5130119257503085,
427
+ "regret_mae": 0.3958987047274907,
428
+ "success_accuracy": 0.8680730561415354,
429
+ "total_loss": 1.4394984311527677
430
+ },
431
+ "per_task": {
432
+ "LiftPegUpright-v1": {
433
+ "action_mse_to_best": 0.35602021238203935,
434
+ "expert_success_rate": 0.8229166666666666,
435
+ "num_groups": 96,
436
+ "oracle_success_rate": 0.9270833333333334,
437
+ "policy_expert_regret": 0.8476849501021206,
438
+ "policy_oracle_regret": 0.9539546399998168,
439
+ "policy_rollout_progress": 0.6463205499264101,
440
+ "policy_rollout_success_rate": 0.3333333333333333,
441
+ "restore_max_error": 1.955777406692505e-07
442
+ },
443
+ "PickCube-v1": {
444
+ "action_mse_to_best": 0.3756813048149901,
445
+ "expert_success_rate": 0.9444444444444444,
446
+ "num_groups": 198,
447
+ "oracle_success_rate": 0.9595959595959596,
448
+ "policy_expert_regret": 0.973482538541459,
449
+ "policy_oracle_regret": 0.9805659890701675,
450
+ "policy_rollout_progress": 0.6379556701415114,
451
+ "policy_rollout_success_rate": 0.3434343434343434,
452
+ "restore_max_error": 2.384185791015625e-07
453
+ },
454
+ "PullCube-v1": {
455
+ "action_mse_to_best": 0.6409675161871645,
456
+ "expert_success_rate": 0.24444444444444444,
457
+ "num_groups": 90,
458
+ "oracle_success_rate": 0.4666666666666667,
459
+ "policy_expert_regret": 0.26891600628279977,
460
+ "policy_oracle_regret": 0.45518461333380805,
461
+ "policy_rollout_progress": 0.3699289129012161,
462
+ "policy_rollout_success_rate": 0.26666666666666666,
463
+ "restore_max_error": 2.384185791015625e-07
464
+ },
465
+ "PushCube-v1": {
466
+ "action_mse_to_best": 0.397049247997232,
467
+ "expert_success_rate": 0.8514851485148515,
468
+ "num_groups": 101,
469
+ "oracle_success_rate": 1.0,
470
+ "policy_expert_regret": 0.3614020819711213,
471
+ "policy_oracle_regret": 0.41630433337523204,
472
+ "policy_rollout_progress": 0.8015174488029858,
473
+ "policy_rollout_success_rate": 0.7821782178217822,
474
+ "restore_max_error": 2.384185791015625e-07
475
+ },
476
+ "StackCube-v1": {
477
+ "action_mse_to_best": 0.5226443493117888,
478
+ "expert_success_rate": 0.7666666666666667,
479
+ "num_groups": 90,
480
+ "oracle_success_rate": 0.9111111111111111,
481
+ "policy_expert_regret": 1.1748778588242001,
482
+ "policy_oracle_regret": 1.3369984101918009,
483
+ "policy_rollout_progress": 0.4047339239054256,
484
+ "policy_rollout_success_rate": 0.14444444444444443,
485
+ "restore_max_error": 2.2351741790771484e-07
486
+ }
487
+ }
488
+ }
489
+ ]
490
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_wgmargin0p05_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_wgpickpull_wgmargin0p05_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: 35.88% +/- 1.58%
11
+ Gain vs h=16 rank checkpoint: +6.14%
12
+ Mean progress: 57.31%
13
+ Mean action MSE to best: 0.414
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 margins residual_wrong_gripper=0.05@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 | 35.65% | 55.93% | 85.74% | n/a | n/a | 0.395 |
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 margins residual_wrong_gripper=0.05@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% | 57.03% | 86.96% | n/a | n/a | 0.407 |
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 margins residual_wrong_gripper=0.05@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% | 58.96% | 87.65% | n/a | n/a | 0.441 |