Hugging Face's logo

v1tavitavita
/
lehome-residual-v4-global

Robotics
LeRobot
Safetensors
lehome-challenge
smolvla
residual-rl
Model card Files
xet
 Community
v1tavitavita commited on
Commit
25f8e90
·
verified ·
1 Parent(s): bb24889

Upload 4 files

Browse files
Files changed (4) hide show
  1. FINAL_SUMMARY.json +1923 -0
  2. README.md +129 -0
  3. RESULTS_SUMMARY.txt +13 -0
  4. residual_v4_global.py +131 -0
FINAL_SUMMARY.json ADDED
@@ -0,0 +1,1923 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "config": {
3
+ "policy_type": "residual_v4_global",
4
+ "policy_class": "ResidualV4GlobalPolicy",
5
+ "backbone_path": "/root/gpufree-data/lehome-dev/artifacts/models/smolvla_four_types/checkpoints/030000/pretrained_model",
6
+ "backbone_sha256": "7ff3915571622bf7530e9ba35540abf5c14f62d8c6a57491664b65a23869e6bc",
7
+ "residual_ckpt_path": "/root/gpufree-data/lehome-dev/artifacts/rl/four_types_sparse_v4/train/residual_averaged.pt",
8
+ "residual_ckpt_sha256": "9d695e278b4361509ac7e35f7d66eb251ec7e7f1f7c53878d453ef2b8aa0ce74",
9
+ "residual_scale": 0.03,
10
+ "task_description": "fold the garment on the table",
11
+ "deterministic": true,
12
+ "host": "lehome3 / 120.209.70.195:30239"
13
+ },
14
+ "by_type": {
15
+ "Top_Long": {
16
+ "total_ep": 60,
17
+ "total_success": 43,
18
+ "success_rate": 0.7166666666666667,
19
+ "seen": {
20
+ "ep": 50,
21
+ "success": 37,
22
+ "rate": 0.74
23
+ },
24
+ "unseen": {
25
+ "ep": 10,
26
+ "success": 6,
27
+ "rate": 0.6
28
+ },
29
+ "garments": {
30
+ "Top_Long_Seen_0": {
31
+ "n_ep": 5,
32
+ "n_success": 4,
33
+ "success_rate": 0.8,
34
+ "exit_code": "0",
35
+ "episodes": [
36
+ {
37
+ "idx": 1,
38
+ "return": 107.13,
39
+ "length": 600,
40
+ "success": false
41
+ },
42
+ {
43
+ "idx": 2,
44
+ "return": 119.28,
45
+ "length": 286,
46
+ "success": true
47
+ },
48
+ {
49
+ "idx": 3,
50
+ "return": 130.25,
51
+ "length": 293,
52
+ "success": true
53
+ },
54
+ {
55
+ "idx": 4,
56
+ "return": 132.55,
57
+ "length": 293,
58
+ "success": true
59
+ },
60
+ {
61
+ "idx": 5,
62
+ "return": 132.78,
63
+ "length": 293,
64
+ "success": true
65
+ }
66
+ ]
67
+ },
68
+ "Top_Long_Seen_1": {
69
+ "n_ep": 5,
70
+ "n_success": 4,
71
+ "success_rate": 0.8,
72
+ "exit_code": "0",
73
+ "episodes": [
74
+ {
75
+ "idx": 1,
76
+ "return": 146.39,
77
+ "length": 326,
78
+ "success": true
79
+ },
80
+ {
81
+ "idx": 2,
82
+ "return": 133.62,
83
+ "length": 293,
84
+ "success": true
85
+ },
86
+ {
87
+ "idx": 3,
88
+ "return": 139.13,
89
+ "length": 293,
90
+ "success": true
91
+ },
92
+ {
93
+ "idx": 4,
94
+ "return": 108.99,
95
+ "length": 293,
96
+ "success": true
97
+ },
98
+ {
99
+ "idx": 5,
100
+ "return": 128.61,
101
+ "length": 600,
102
+ "success": false
103
+ }
104
+ ]
105
+ },
106
+ "Top_Long_Seen_2": {
107
+ "n_ep": 5,
108
+ "n_success": 3,
109
+ "success_rate": 0.6,
110
+ "exit_code": "0",
111
+ "episodes": [
112
+ {
113
+ "idx": 1,
114
+ "return": 127.33,
115
+ "length": 326,
116
+ "success": true
117
+ },
118
+ {
119
+ "idx": 2,
120
+ "return": 151.7,
121
+ "length": 600,
122
+ "success": false
123
+ },
124
+ {
125
+ "idx": 3,
126
+ "return": 294.63,
127
+ "length": 600,
128
+ "success": false
129
+ },
130
+ {
131
+ "idx": 4,
132
+ "return": 101.97,
133
+ "length": 313,
134
+ "success": true
135
+ },
136
+ {
137
+ "idx": 5,
138
+ "return": 129.57,
139
+ "length": 293,
140
+ "success": true
141
+ }
142
+ ]
143
+ },
144
+ "Top_Long_Seen_3": {
145
+ "n_ep": 5,
146
+ "n_success": 4,
147
+ "success_rate": 0.8,
148
+ "exit_code": "0",
149
+ "episodes": [
150
+ {
151
+ "idx": 1,
152
+ "return": 139.86,
153
+ "length": 326,
154
+ "success": true
155
+ },
156
+ {
157
+ "idx": 2,
158
+ "return": 194.11,
159
+ "length": 600,
160
+ "success": false
161
+ },
162
+ {
163
+ "idx": 3,
164
+ "return": 124.3,
165
+ "length": 303,
166
+ "success": true
167
+ },
168
+ {
169
+ "idx": 4,
170
+ "return": 149.06,
171
+ "length": 343,
172
+ "success": true
173
+ },
174
+ {
175
+ "idx": 5,
176
+ "return": 147.65,
177
+ "length": 343,
178
+ "success": true
179
+ }
180
+ ]
181
+ },
182
+ "Top_Long_Seen_4": {
183
+ "n_ep": 5,
184
+ "n_success": 5,
185
+ "success_rate": 1.0,
186
+ "exit_code": "0",
187
+ "episodes": [
188
+ {
189
+ "idx": 1,
190
+ "return": 90.92,
191
+ "length": 276,
192
+ "success": true
193
+ },
194
+ {
195
+ "idx": 2,
196
+ "return": 139.21,
197
+ "length": 293,
198
+ "success": true
199
+ },
200
+ {
201
+ "idx": 3,
202
+ "return": 138.42,
203
+ "length": 293,
204
+ "success": true
205
+ },
206
+ {
207
+ "idx": 4,
208
+ "return": 94.25,
209
+ "length": 293,
210
+ "success": true
211
+ },
212
+ {
213
+ "idx": 5,
214
+ "return": 127.3,
215
+ "length": 293,
216
+ "success": true
217
+ }
218
+ ]
219
+ },
220
+ "Top_Long_Seen_5": {
221
+ "n_ep": 5,
222
+ "n_success": 5,
223
+ "success_rate": 1.0,
224
+ "exit_code": "0",
225
+ "episodes": [
226
+ {
227
+ "idx": 1,
228
+ "return": 84.64,
229
+ "length": 276,
230
+ "success": true
231
+ },
232
+ {
233
+ "idx": 2,
234
+ "return": 130.13,
235
+ "length": 293,
236
+ "success": true
237
+ },
238
+ {
239
+ "idx": 3,
240
+ "return": 141.21,
241
+ "length": 293,
242
+ "success": true
243
+ },
244
+ {
245
+ "idx": 4,
246
+ "return": 136.76,
247
+ "length": 343,
248
+ "success": true
249
+ },
250
+ {
251
+ "idx": 5,
252
+ "return": 143.51,
253
+ "length": 343,
254
+ "success": true
255
+ }
256
+ ]
257
+ },
258
+ "Top_Long_Seen_6": {
259
+ "n_ep": 5,
260
+ "n_success": 3,
261
+ "success_rate": 0.6,
262
+ "exit_code": "0",
263
+ "episodes": [
264
+ {
265
+ "idx": 1,
266
+ "return": 184.91,
267
+ "length": 526,
268
+ "success": true
269
+ },
270
+ {
271
+ "idx": 2,
272
+ "return": 153.59,
273
+ "length": 600,
274
+ "success": false
275
+ },
276
+ {
277
+ "idx": 3,
278
+ "return": 106.88,
279
+ "length": 303,
280
+ "success": true
281
+ },
282
+ {
283
+ "idx": 4,
284
+ "return": 301.33,
285
+ "length": 600,
286
+ "success": false
287
+ },
288
+ {
289
+ "idx": 5,
290
+ "return": 98.9,
291
+ "length": 303,
292
+ "success": true
293
+ }
294
+ ]
295
+ },
296
+ "Top_Long_Seen_7": {
297
+ "n_ep": 5,
298
+ "n_success": 2,
299
+ "success_rate": 0.4,
300
+ "exit_code": "0",
301
+ "episodes": [
302
+ {
303
+ "idx": 1,
304
+ "return": 271.85,
305
+ "length": 600,
306
+ "success": false
307
+ },
308
+ {
309
+ "idx": 2,
310
+ "return": 89.38,
311
+ "length": 286,
312
+ "success": true
313
+ },
314
+ {
315
+ "idx": 3,
316
+ "return": 306.21,
317
+ "length": 600,
318
+ "success": false
319
+ },
320
+ {
321
+ "idx": 4,
322
+ "return": 264.82,
323
+ "length": 600,
324
+ "success": false
325
+ },
326
+ {
327
+ "idx": 5,
328
+ "return": 85.21,
329
+ "length": 263,
330
+ "success": true
331
+ }
332
+ ]
333
+ },
334
+ "Top_Long_Seen_8": {
335
+ "n_ep": 5,
336
+ "n_success": 3,
337
+ "success_rate": 0.6,
338
+ "exit_code": "0",
339
+ "episodes": [
340
+ {
341
+ "idx": 1,
342
+ "return": 333.17,
343
+ "length": 600,
344
+ "success": false
345
+ },
346
+ {
347
+ "idx": 2,
348
+ "return": 91.8,
349
+ "length": 286,
350
+ "success": true
351
+ },
352
+ {
353
+ "idx": 3,
354
+ "return": 83.28,
355
+ "length": 243,
356
+ "success": true
357
+ },
358
+ {
359
+ "idx": 4,
360
+ "return": 127.02,
361
+ "length": 293,
362
+ "success": true
363
+ },
364
+ {
365
+ "idx": 5,
366
+ "return": 346.87,
367
+ "length": 600,
368
+ "success": false
369
+ }
370
+ ]
371
+ },
372
+ "Top_Long_Seen_9": {
373
+ "n_ep": 5,
374
+ "n_success": 4,
375
+ "success_rate": 0.8,
376
+ "exit_code": "0",
377
+ "episodes": [
378
+ {
379
+ "idx": 1,
380
+ "return": 134.93,
381
+ "length": 276,
382
+ "success": true
383
+ },
384
+ {
385
+ "idx": 2,
386
+ "return": 115.85,
387
+ "length": 293,
388
+ "success": true
389
+ },
390
+ {
391
+ "idx": 3,
392
+ "return": 126.17,
393
+ "length": 293,
394
+ "success": true
395
+ },
396
+ {
397
+ "idx": 4,
398
+ "return": 136.25,
399
+ "length": 293,
400
+ "success": true
401
+ },
402
+ {
403
+ "idx": 5,
404
+ "return": 335.05,
405
+ "length": 600,
406
+ "success": false
407
+ }
408
+ ]
409
+ },
410
+ "Top_Long_Unseen_0": {
411
+ "n_ep": 5,
412
+ "n_success": 5,
413
+ "success_rate": 1.0,
414
+ "exit_code": "0",
415
+ "episodes": [
416
+ {
417
+ "idx": 1,
418
+ "return": 134.37,
419
+ "length": 276,
420
+ "success": true
421
+ },
422
+ {
423
+ "idx": 2,
424
+ "return": 140.06,
425
+ "length": 293,
426
+ "success": true
427
+ },
428
+ {
429
+ "idx": 3,
430
+ "return": 138.81,
431
+ "length": 293,
432
+ "success": true
433
+ },
434
+ {
435
+ "idx": 4,
436
+ "return": 88.22,
437
+ "length": 243,
438
+ "success": true
439
+ },
440
+ {
441
+ "idx": 5,
442
+ "return": 141.7,
443
+ "length": 293,
444
+ "success": true
445
+ }
446
+ ]
447
+ },
448
+ "Top_Long_Unseen_1": {
449
+ "n_ep": 5,
450
+ "n_success": 1,
451
+ "success_rate": 0.2,
452
+ "exit_code": "0",
453
+ "episodes": [
454
+ {
455
+ "idx": 1,
456
+ "return": 107.97,
457
+ "length": 600,
458
+ "success": false
459
+ },
460
+ {
461
+ "idx": 2,
462
+ "return": 117.96,
463
+ "length": 286,
464
+ "success": true
465
+ },
466
+ {
467
+ "idx": 3,
468
+ "return": 98.82,
469
+ "length": 600,
470
+ "success": false
471
+ },
472
+ {
473
+ "idx": 4,
474
+ "return": 102.49,
475
+ "length": 600,
476
+ "success": false
477
+ },
478
+ {
479
+ "idx": 5,
480
+ "return": 173.82,
481
+ "length": 600,
482
+ "success": false
483
+ }
484
+ ]
485
+ }
486
+ }
487
+ },
488
+ "Top_Short": {
489
+ "total_ep": 60,
490
+ "total_success": 25,
491
+ "success_rate": 0.4166666666666667,
492
+ "seen": {
493
+ "ep": 50,
494
+ "success": 24,
495
+ "rate": 0.48
496
+ },
497
+ "unseen": {
498
+ "ep": 10,
499
+ "success": 1,
500
+ "rate": 0.1
501
+ },
502
+ "garments": {
503
+ "Top_Short_Seen_0": {
504
+ "n_ep": 5,
505
+ "n_success": 3,
506
+ "success_rate": 0.6,
507
+ "exit_code": "0",
508
+ "episodes": [
509
+ {
510
+ "idx": 1,
511
+ "return": 120.76,
512
+ "length": 276,
513
+ "success": true
514
+ },
515
+ {
516
+ "idx": 2,
517
+ "return": 317.46,
518
+ "length": 600,
519
+ "success": false
520
+ },
521
+ {
522
+ "idx": 3,
523
+ "return": 132.78,
524
+ "length": 303,
525
+ "success": true
526
+ },
527
+ {
528
+ "idx": 4,
529
+ "return": 294.51,
530
+ "length": 600,
531
+ "success": false
532
+ },
533
+ {
534
+ "idx": 5,
535
+ "return": 78.41,
536
+ "length": 253,
537
+ "success": true
538
+ }
539
+ ]
540
+ },
541
+ "Top_Short_Seen_1": {
542
+ "n_ep": 5,
543
+ "n_success": 3,
544
+ "success_rate": 0.6,
545
+ "exit_code": "0",
546
+ "episodes": [
547
+ {
548
+ "idx": 1,
549
+ "return": 347.09,
550
+ "length": 600,
551
+ "success": false
552
+ },
553
+ {
554
+ "idx": 2,
555
+ "return": 349.4,
556
+ "length": 600,
557
+ "success": false
558
+ },
559
+ {
560
+ "idx": 3,
561
+ "return": 80.24,
562
+ "length": 246,
563
+ "success": true
564
+ },
565
+ {
566
+ "idx": 4,
567
+ "return": 77.14,
568
+ "length": 243,
569
+ "success": true
570
+ },
571
+ {
572
+ "idx": 5,
573
+ "return": 78.58,
574
+ "length": 243,
575
+ "success": true
576
+ }
577
+ ]
578
+ },
579
+ "Top_Short_Seen_2": {
580
+ "n_ep": 5,
581
+ "n_success": 4,
582
+ "success_rate": 0.8,
583
+ "exit_code": "0",
584
+ "episodes": [
585
+ {
586
+ "idx": 1,
587
+ "return": 71.52,
588
+ "length": 226,
589
+ "success": true
590
+ },
591
+ {
592
+ "idx": 2,
593
+ "return": 134.23,
594
+ "length": 293,
595
+ "success": true
596
+ },
597
+ {
598
+ "idx": 3,
599
+ "return": 80.48,
600
+ "length": 243,
601
+ "success": true
602
+ },
603
+ {
604
+ "idx": 4,
605
+ "return": 131.94,
606
+ "length": 293,
607
+ "success": true
608
+ },
609
+ {
610
+ "idx": 5,
611
+ "return": 241.39,
612
+ "length": 600,
613
+ "success": false
614
+ }
615
+ ]
616
+ },
617
+ "Top_Short_Seen_3": {
618
+ "n_ep": 5,
619
+ "n_success": 2,
620
+ "success_rate": 0.4,
621
+ "exit_code": "0",
622
+ "episodes": [
623
+ {
624
+ "idx": 1,
625
+ "return": 236.26,
626
+ "length": 600,
627
+ "success": false
628
+ },
629
+ {
630
+ "idx": 2,
631
+ "return": 155.19,
632
+ "length": 286,
633
+ "success": true
634
+ },
635
+ {
636
+ "idx": 3,
637
+ "return": 158.1,
638
+ "length": 293,
639
+ "success": true
640
+ },
641
+ {
642
+ "idx": 4,
643
+ "return": 211.93,
644
+ "length": 600,
645
+ "success": false
646
+ },
647
+ {
648
+ "idx": 5,
649
+ "return": 181.74,
650
+ "length": 600,
651
+ "success": false
652
+ }
653
+ ]
654
+ },
655
+ "Top_Short_Seen_4": {
656
+ "n_ep": 5,
657
+ "n_success": 5,
658
+ "success_rate": 1.0,
659
+ "exit_code": "0",
660
+ "episodes": [
661
+ {
662
+ "idx": 1,
663
+ "return": 143.04,
664
+ "length": 276,
665
+ "success": true
666
+ },
667
+ {
668
+ "idx": 2,
669
+ "return": 157.74,
670
+ "length": 293,
671
+ "success": true
672
+ },
673
+ {
674
+ "idx": 3,
675
+ "return": 159.61,
676
+ "length": 293,
677
+ "success": true
678
+ },
679
+ {
680
+ "idx": 4,
681
+ "return": 155.01,
682
+ "length": 293,
683
+ "success": true
684
+ },
685
+ {
686
+ "idx": 5,
687
+ "return": 123.99,
688
+ "length": 243,
689
+ "success": true
690
+ }
691
+ ]
692
+ },
693
+ "Top_Short_Seen_5": {
694
+ "n_ep": 5,
695
+ "n_success": 2,
696
+ "success_rate": 0.4,
697
+ "exit_code": "0",
698
+ "episodes": [
699
+ {
700
+ "idx": 1,
701
+ "return": 261.31,
702
+ "length": 600,
703
+ "success": false
704
+ },
705
+ {
706
+ "idx": 2,
707
+ "return": 280.72,
708
+ "length": 600,
709
+ "success": false
710
+ },
711
+ {
712
+ "idx": 3,
713
+ "return": 64.64,
714
+ "length": 296,
715
+ "success": true
716
+ },
717
+ {
718
+ "idx": 4,
719
+ "return": 109.69,
720
+ "length": 293,
721
+ "success": true
722
+ },
723
+ {
724
+ "idx": 5,
725
+ "return": 217.27,
726
+ "length": 600,
727
+ "success": false
728
+ }
729
+ ]
730
+ },
731
+ "Top_Short_Seen_6": {
732
+ "n_ep": 5,
733
+ "n_success": 2,
734
+ "success_rate": 0.4,
735
+ "exit_code": "0",
736
+ "episodes": [
737
+ {
738
+ "idx": 1,
739
+ "return": 93.42,
740
+ "length": 276,
741
+ "success": true
742
+ },
743
+ {
744
+ "idx": 2,
745
+ "return": 245.57,
746
+ "length": 600,
747
+ "success": false
748
+ },
749
+ {
750
+ "idx": 3,
751
+ "return": 108.79,
752
+ "length": 303,
753
+ "success": true
754
+ },
755
+ {
756
+ "idx": 4,
757
+ "return": 273.29,
758
+ "length": 600,
759
+ "success": false
760
+ },
761
+ {
762
+ "idx": 5,
763
+ "return": 302.15,
764
+ "length": 600,
765
+ "success": false
766
+ }
767
+ ]
768
+ },
769
+ "Top_Short_Seen_7": {
770
+ "n_ep": 5,
771
+ "n_success": 1,
772
+ "success_rate": 0.2,
773
+ "exit_code": "0",
774
+ "episodes": [
775
+ {
776
+ "idx": 1,
777
+ "return": 311.42,
778
+ "length": 600,
779
+ "success": false
780
+ },
781
+ {
782
+ "idx": 2,
783
+ "return": 124.72,
784
+ "length": 286,
785
+ "success": true
786
+ },
787
+ {
788
+ "idx": 3,
789
+ "return": 285.96,
790
+ "length": 600,
791
+ "success": false
792
+ },
793
+ {
794
+ "idx": 4,
795
+ "return": 104.8,
796
+ "length": 600,
797
+ "success": false
798
+ },
799
+ {
800
+ "idx": 5,
801
+ "return": 282.99,
802
+ "length": 600,
803
+ "success": false
804
+ }
805
+ ]
806
+ },
807
+ "Top_Short_Seen_8": {
808
+ "n_ep": 5,
809
+ "n_success": 2,
810
+ "success_rate": 0.4,
811
+ "exit_code": "0",
812
+ "episodes": [
813
+ {
814
+ "idx": 1,
815
+ "return": 173.59,
816
+ "length": 600,
817
+ "success": false
818
+ },
819
+ {
820
+ "idx": 2,
821
+ "return": 128.72,
822
+ "length": 286,
823
+ "success": true
824
+ },
825
+ {
826
+ "idx": 3,
827
+ "return": 130.46,
828
+ "length": 293,
829
+ "success": true
830
+ },
831
+ {
832
+ "idx": 4,
833
+ "return": 221.56,
834
+ "length": 600,
835
+ "success": false
836
+ },
837
+ {
838
+ "idx": 5,
839
+ "return": 175.49,
840
+ "length": 600,
841
+ "success": false
842
+ }
843
+ ]
844
+ },
845
+ "Top_Short_Seen_9": {
846
+ "n_ep": 5,
847
+ "n_success": 0,
848
+ "success_rate": 0.0,
849
+ "exit_code": "0",
850
+ "episodes": [
851
+ {
852
+ "idx": 1,
853
+ "return": 106.43,
854
+ "length": 600,
855
+ "success": false
856
+ },
857
+ {
858
+ "idx": 2,
859
+ "return": 191.3,
860
+ "length": 600,
861
+ "success": false
862
+ },
863
+ {
864
+ "idx": 3,
865
+ "return": 226.0,
866
+ "length": 600,
867
+ "success": false
868
+ },
869
+ {
870
+ "idx": 4,
871
+ "return": 202.59,
872
+ "length": 600,
873
+ "success": false
874
+ },
875
+ {
876
+ "idx": 5,
877
+ "return": 233.97,
878
+ "length": 600,
879
+ "success": false
880
+ }
881
+ ]
882
+ },
883
+ "Top_Short_Unseen_0": {
884
+ "n_ep": 5,
885
+ "n_success": 0,
886
+ "success_rate": 0.0,
887
+ "exit_code": "0",
888
+ "episodes": [
889
+ {
890
+ "idx": 1,
891
+ "return": 107.75,
892
+ "length": 600,
893
+ "success": false
894
+ },
895
+ {
896
+ "idx": 2,
897
+ "return": 109.55,
898
+ "length": 600,
899
+ "success": false
900
+ },
901
+ {
902
+ "idx": 3,
903
+ "return": 110.61,
904
+ "length": 600,
905
+ "success": false
906
+ },
907
+ {
908
+ "idx": 4,
909
+ "return": 105.2,
910
+ "length": 600,
911
+ "success": false
912
+ },
913
+ {
914
+ "idx": 5,
915
+ "return": 109.6,
916
+ "length": 600,
917
+ "success": false
918
+ }
919
+ ]
920
+ },
921
+ "Top_Short_Unseen_1": {
922
+ "n_ep": 5,
923
+ "n_success": 1,
924
+ "success_rate": 0.2,
925
+ "exit_code": "0",
926
+ "episodes": [
927
+ {
928
+ "idx": 1,
929
+ "return": 267.1,
930
+ "length": 600,
931
+ "success": false
932
+ },
933
+ {
934
+ "idx": 2,
935
+ "return": 151.76,
936
+ "length": 286,
937
+ "success": true
938
+ },
939
+ {
940
+ "idx": 3,
941
+ "return": 187.69,
942
+ "length": 600,
943
+ "success": false
944
+ },
945
+ {
946
+ "idx": 4,
947
+ "return": 180.23,
948
+ "length": 600,
949
+ "success": false
950
+ },
951
+ {
952
+ "idx": 5,
953
+ "return": 356.67,
954
+ "length": 600,
955
+ "success": false
956
+ }
957
+ ]
958
+ }
959
+ }
960
+ },
961
+ "Pant_Long": {
962
+ "total_ep": 60,
963
+ "total_success": 28,
964
+ "success_rate": 0.4666666666666667,
965
+ "seen": {
966
+ "ep": 50,
967
+ "success": 27,
968
+ "rate": 0.54
969
+ },
970
+ "unseen": {
971
+ "ep": 10,
972
+ "success": 1,
973
+ "rate": 0.1
974
+ },
975
+ "garments": {
976
+ "Pant_Long_Seen_0": {
977
+ "n_ep": 5,
978
+ "n_success": 3,
979
+ "success_rate": 0.6,
980
+ "exit_code": "0",
981
+ "episodes": [
982
+ {
983
+ "idx": 1,
984
+ "return": 112.57,
985
+ "length": 600,
986
+ "success": false
987
+ },
988
+ {
989
+ "idx": 2,
990
+ "return": 80.38,
991
+ "length": 236,
992
+ "success": true
993
+ },
994
+ {
995
+ "idx": 3,
996
+ "return": 88.66,
997
+ "length": 243,
998
+ "success": true
999
+ },
1000
+ {
1001
+ "idx": 4,
1002
+ "return": 113.27,
1003
+ "length": 243,
1004
+ "success": true
1005
+ },
1006
+ {
1007
+ "idx": 5,
1008
+ "return": 131.67,
1009
+ "length": 600,
1010
+ "success": false
1011
+ }
1012
+ ]
1013
+ },
1014
+ "Pant_Long_Seen_1": {
1015
+ "n_ep": 5,
1016
+ "n_success": 2,
1017
+ "success_rate": 0.4,
1018
+ "exit_code": "0",
1019
+ "episodes": [
1020
+ {
1021
+ "idx": 1,
1022
+ "return": 74.93,
1023
+ "length": 226,
1024
+ "success": true
1025
+ },
1026
+ {
1027
+ "idx": 2,
1028
+ "return": 144.94,
1029
+ "length": 600,
1030
+ "success": false
1031
+ },
1032
+ {
1033
+ "idx": 3,
1034
+ "return": 193.31,
1035
+ "length": 600,
1036
+ "success": false
1037
+ },
1038
+ {
1039
+ "idx": 4,
1040
+ "return": 125.21,
1041
+ "length": 263,
1042
+ "success": true
1043
+ },
1044
+ {
1045
+ "idx": 5,
1046
+ "return": 142.17,
1047
+ "length": 600,
1048
+ "success": false
1049
+ }
1050
+ ]
1051
+ },
1052
+ "Pant_Long_Seen_2": {
1053
+ "n_ep": 5,
1054
+ "n_success": 1,
1055
+ "success_rate": 0.2,
1056
+ "exit_code": "0",
1057
+ "episodes": [
1058
+ {
1059
+ "idx": 1,
1060
+ "return": 225.21,
1061
+ "length": 600,
1062
+ "success": false
1063
+ },
1064
+ {
1065
+ "idx": 2,
1066
+ "return": 246.38,
1067
+ "length": 600,
1068
+ "success": false
1069
+ },
1070
+ {
1071
+ "idx": 3,
1072
+ "return": 251.02,
1073
+ "length": 600,
1074
+ "success": false
1075
+ },
1076
+ {
1077
+ "idx": 4,
1078
+ "return": 110.12,
1079
+ "length": 256,
1080
+ "success": true
1081
+ },
1082
+ {
1083
+ "idx": 5,
1084
+ "return": 117.5,
1085
+ "length": 600,
1086
+ "success": false
1087
+ }
1088
+ ]
1089
+ },
1090
+ "Pant_Long_Seen_3": {
1091
+ "n_ep": 5,
1092
+ "n_success": 3,
1093
+ "success_rate": 0.6,
1094
+ "exit_code": "0",
1095
+ "episodes": [
1096
+ {
1097
+ "idx": 1,
1098
+ "return": 124.02,
1099
+ "length": 600,
1100
+ "success": false
1101
+ },
1102
+ {
1103
+ "idx": 2,
1104
+ "return": 284.23,
1105
+ "length": 436,
1106
+ "success": true
1107
+ },
1108
+ {
1109
+ "idx": 3,
1110
+ "return": 271.73,
1111
+ "length": 600,
1112
+ "success": false
1113
+ },
1114
+ {
1115
+ "idx": 4,
1116
+ "return": 70.95,
1117
+ "length": 203,
1118
+ "success": true
1119
+ },
1120
+ {
1121
+ "idx": 5,
1122
+ "return": 121.35,
1123
+ "length": 243,
1124
+ "success": true
1125
+ }
1126
+ ]
1127
+ },
1128
+ "Pant_Long_Seen_4": {
1129
+ "n_ep": 5,
1130
+ "n_success": 4,
1131
+ "success_rate": 0.8,
1132
+ "exit_code": "0",
1133
+ "episodes": [
1134
+ {
1135
+ "idx": 1,
1136
+ "return": 71.02,
1137
+ "length": 226,
1138
+ "success": true
1139
+ },
1140
+ {
1141
+ "idx": 2,
1142
+ "return": 78.52,
1143
+ "length": 243,
1144
+ "success": true
1145
+ },
1146
+ {
1147
+ "idx": 3,
1148
+ "return": 76.16,
1149
+ "length": 243,
1150
+ "success": true
1151
+ },
1152
+ {
1153
+ "idx": 4,
1154
+ "return": 79.02,
1155
+ "length": 243,
1156
+ "success": true
1157
+ },
1158
+ {
1159
+ "idx": 5,
1160
+ "return": 289.31,
1161
+ "length": 600,
1162
+ "success": false
1163
+ }
1164
+ ]
1165
+ },
1166
+ "Pant_Long_Seen_5": {
1167
+ "n_ep": 5,
1168
+ "n_success": 3,
1169
+ "success_rate": 0.6,
1170
+ "exit_code": "0",
1171
+ "episodes": [
1172
+ {
1173
+ "idx": 1,
1174
+ "return": 71.14,
1175
+ "length": 226,
1176
+ "success": true
1177
+ },
1178
+ {
1179
+ "idx": 2,
1180
+ "return": 282.8,
1181
+ "length": 600,
1182
+ "success": false
1183
+ },
1184
+ {
1185
+ "idx": 3,
1186
+ "return": 106.07,
1187
+ "length": 253,
1188
+ "success": true
1189
+ },
1190
+ {
1191
+ "idx": 4,
1192
+ "return": 118.34,
1193
+ "length": 243,
1194
+ "success": true
1195
+ },
1196
+ {
1197
+ "idx": 5,
1198
+ "return": 112.33,
1199
+ "length": 600,
1200
+ "success": false
1201
+ }
1202
+ ]
1203
+ },
1204
+ "Pant_Long_Seen_6": {
1205
+ "n_ep": 5,
1206
+ "n_success": 4,
1207
+ "success_rate": 0.8,
1208
+ "exit_code": "0",
1209
+ "episodes": [
1210
+ {
1211
+ "idx": 1,
1212
+ "return": 212.99,
1213
+ "length": 600,
1214
+ "success": false
1215
+ },
1216
+ {
1217
+ "idx": 2,
1218
+ "return": 109.8,
1219
+ "length": 236,
1220
+ "success": true
1221
+ },
1222
+ {
1223
+ "idx": 3,
1224
+ "return": 118.22,
1225
+ "length": 243,
1226
+ "success": true
1227
+ },
1228
+ {
1229
+ "idx": 4,
1230
+ "return": 104.0,
1231
+ "length": 243,
1232
+ "success": true
1233
+ },
1234
+ {
1235
+ "idx": 5,
1236
+ "return": 140.5,
1237
+ "length": 443,
1238
+ "success": true
1239
+ }
1240
+ ]
1241
+ },
1242
+ "Pant_Long_Seen_7": {
1243
+ "n_ep": 5,
1244
+ "n_success": 4,
1245
+ "success_rate": 0.8,
1246
+ "exit_code": "0",
1247
+ "episodes": [
1248
+ {
1249
+ "idx": 1,
1250
+ "return": 144.45,
1251
+ "length": 600,
1252
+ "success": false
1253
+ },
1254
+ {
1255
+ "idx": 2,
1256
+ "return": 113.33,
1257
+ "length": 236,
1258
+ "success": true
1259
+ },
1260
+ {
1261
+ "idx": 3,
1262
+ "return": 111.85,
1263
+ "length": 243,
1264
+ "success": true
1265
+ },
1266
+ {
1267
+ "idx": 4,
1268
+ "return": 65.29,
1269
+ "length": 193,
1270
+ "success": true
1271
+ },
1272
+ {
1273
+ "idx": 5,
1274
+ "return": 119.77,
1275
+ "length": 243,
1276
+ "success": true
1277
+ }
1278
+ ]
1279
+ },
1280
+ "Pant_Long_Seen_8": {
1281
+ "n_ep": 5,
1282
+ "n_success": 2,
1283
+ "success_rate": 0.4,
1284
+ "exit_code": "0",
1285
+ "episodes": [
1286
+ {
1287
+ "idx": 1,
1288
+ "return": 97.96,
1289
+ "length": 600,
1290
+ "success": false
1291
+ },
1292
+ {
1293
+ "idx": 2,
1294
+ "return": 95.88,
1295
+ "length": 186,
1296
+ "success": true
1297
+ },
1298
+ {
1299
+ "idx": 3,
1300
+ "return": 120.89,
1301
+ "length": 243,
1302
+ "success": true
1303
+ },
1304
+ {
1305
+ "idx": 4,
1306
+ "return": 225.67,
1307
+ "length": 600,
1308
+ "success": false
1309
+ },
1310
+ {
1311
+ "idx": 5,
1312
+ "return": 104.63,
1313
+ "length": 600,
1314
+ "success": false
1315
+ }
1316
+ ]
1317
+ },
1318
+ "Pant_Long_Seen_9": {
1319
+ "n_ep": 5,
1320
+ "n_success": 1,
1321
+ "success_rate": 0.2,
1322
+ "exit_code": "0",
1323
+ "episodes": [
1324
+ {
1325
+ "idx": 1,
1326
+ "return": 190.31,
1327
+ "length": 600,
1328
+ "success": false
1329
+ },
1330
+ {
1331
+ "idx": 2,
1332
+ "return": 244.67,
1333
+ "length": 386,
1334
+ "success": true
1335
+ },
1336
+ {
1337
+ "idx": 3,
1338
+ "return": 341.02,
1339
+ "length": 600,
1340
+ "success": false
1341
+ },
1342
+ {
1343
+ "idx": 4,
1344
+ "return": 334.68,
1345
+ "length": 600,
1346
+ "success": false
1347
+ },
1348
+ {
1349
+ "idx": 5,
1350
+ "return": 381.72,
1351
+ "length": 600,
1352
+ "success": false
1353
+ }
1354
+ ]
1355
+ },
1356
+ "Pant_Long_Unseen_0": {
1357
+ "n_ep": 5,
1358
+ "n_success": 1,
1359
+ "success_rate": 0.2,
1360
+ "exit_code": "0",
1361
+ "episodes": [
1362
+ {
1363
+ "idx": 1,
1364
+ "return": 191.58,
1365
+ "length": 600,
1366
+ "success": false
1367
+ },
1368
+ {
1369
+ "idx": 2,
1370
+ "return": 116.62,
1371
+ "length": 236,
1372
+ "success": true
1373
+ },
1374
+ {
1375
+ "idx": 3,
1376
+ "return": 253.81,
1377
+ "length": 600,
1378
+ "success": false
1379
+ },
1380
+ {
1381
+ "idx": 4,
1382
+ "return": 181.91,
1383
+ "length": 600,
1384
+ "success": false
1385
+ },
1386
+ {
1387
+ "idx": 5,
1388
+ "return": 110.3,
1389
+ "length": 600,
1390
+ "success": false
1391
+ }
1392
+ ]
1393
+ },
1394
+ "Pant_Long_Unseen_1": {
1395
+ "n_ep": 5,
1396
+ "n_success": 0,
1397
+ "success_rate": 0.0,
1398
+ "exit_code": "0",
1399
+ "episodes": [
1400
+ {
1401
+ "idx": 1,
1402
+ "return": 110.98,
1403
+ "length": 600,
1404
+ "success": false
1405
+ },
1406
+ {
1407
+ "idx": 2,
1408
+ "return": 113.03,
1409
+ "length": 600,
1410
+ "success": false
1411
+ },
1412
+ {
1413
+ "idx": 3,
1414
+ "return": 103.66,
1415
+ "length": 600,
1416
+ "success": false
1417
+ },
1418
+ {
1419
+ "idx": 4,
1420
+ "return": 141.18,
1421
+ "length": 600,
1422
+ "success": false
1423
+ },
1424
+ {
1425
+ "idx": 5,
1426
+ "return": 127.11,
1427
+ "length": 600,
1428
+ "success": false
1429
+ }
1430
+ ]
1431
+ }
1432
+ }
1433
+ },
1434
+ "Pant_Short": {
1435
+ "total_ep": 60,
1436
+ "total_success": 54,
1437
+ "success_rate": 0.9,
1438
+ "seen": {
1439
+ "ep": 50,
1440
+ "success": 48,
1441
+ "rate": 0.96
1442
+ },
1443
+ "unseen": {
1444
+ "ep": 10,
1445
+ "success": 6,
1446
+ "rate": 0.6
1447
+ },
1448
+ "garments": {
1449
+ "Pant_Short_Seen_0": {
1450
+ "n_ep": 5,
1451
+ "n_success": 4,
1452
+ "success_rate": 0.8,
1453
+ "exit_code": "0",
1454
+ "episodes": [
1455
+ {
1456
+ "idx": 1,
1457
+ "return": 419.57,
1458
+ "length": 600,
1459
+ "success": false
1460
+ },
1461
+ {
1462
+ "idx": 2,
1463
+ "return": 105.87,
1464
+ "length": 136,
1465
+ "success": true
1466
+ },
1467
+ {
1468
+ "idx": 3,
1469
+ "return": 111.86,
1470
+ "length": 143,
1471
+ "success": true
1472
+ },
1473
+ {
1474
+ "idx": 4,
1475
+ "return": 104.08,
1476
+ "length": 143,
1477
+ "success": true
1478
+ },
1479
+ {
1480
+ "idx": 5,
1481
+ "return": 116.43,
1482
+ "length": 143,
1483
+ "success": true
1484
+ }
1485
+ ]
1486
+ },
1487
+ "Pant_Short_Seen_1": {
1488
+ "n_ep": 5,
1489
+ "n_success": 5,
1490
+ "success_rate": 1.0,
1491
+ "exit_code": "0",
1492
+ "episodes": [
1493
+ {
1494
+ "idx": 1,
1495
+ "return": 98.68,
1496
+ "length": 126,
1497
+ "success": true
1498
+ },
1499
+ {
1500
+ "idx": 2,
1501
+ "return": 115.22,
1502
+ "length": 143,
1503
+ "success": true
1504
+ },
1505
+ {
1506
+ "idx": 3,
1507
+ "return": 70.88,
1508
+ "length": 93,
1509
+ "success": true
1510
+ },
1511
+ {
1512
+ "idx": 4,
1513
+ "return": 113.0,
1514
+ "length": 143,
1515
+ "success": true
1516
+ },
1517
+ {
1518
+ "idx": 5,
1519
+ "return": 117.76,
1520
+ "length": 143,
1521
+ "success": true
1522
+ }
1523
+ ]
1524
+ },
1525
+ "Pant_Short_Seen_2": {
1526
+ "n_ep": 5,
1527
+ "n_success": 5,
1528
+ "success_rate": 1.0,
1529
+ "exit_code": "0",
1530
+ "episodes": [
1531
+ {
1532
+ "idx": 1,
1533
+ "return": 105.46,
1534
+ "length": 126,
1535
+ "success": true
1536
+ },
1537
+ {
1538
+ "idx": 2,
1539
+ "return": 108.81,
1540
+ "length": 143,
1541
+ "success": true
1542
+ },
1543
+ {
1544
+ "idx": 3,
1545
+ "return": 66.98,
1546
+ "length": 93,
1547
+ "success": true
1548
+ },
1549
+ {
1550
+ "idx": 4,
1551
+ "return": 112.69,
1552
+ "length": 143,
1553
+ "success": true
1554
+ },
1555
+ {
1556
+ "idx": 5,
1557
+ "return": 119.93,
1558
+ "length": 143,
1559
+ "success": true
1560
+ }
1561
+ ]
1562
+ },
1563
+ "Pant_Short_Seen_3": {
1564
+ "n_ep": 5,
1565
+ "n_success": 5,
1566
+ "success_rate": 1.0,
1567
+ "exit_code": "0",
1568
+ "episodes": [
1569
+ {
1570
+ "idx": 1,
1571
+ "return": 126.03,
1572
+ "length": 126,
1573
+ "success": true
1574
+ },
1575
+ {
1576
+ "idx": 2,
1577
+ "return": 126.69,
1578
+ "length": 143,
1579
+ "success": true
1580
+ },
1581
+ {
1582
+ "idx": 3,
1583
+ "return": 128.13,
1584
+ "length": 143,
1585
+ "success": true
1586
+ },
1587
+ {
1588
+ "idx": 4,
1589
+ "return": 133.18,
1590
+ "length": 143,
1591
+ "success": true
1592
+ },
1593
+ {
1594
+ "idx": 5,
1595
+ "return": 81.05,
1596
+ "length": 93,
1597
+ "success": true
1598
+ }
1599
+ ]
1600
+ },
1601
+ "Pant_Short_Seen_4": {
1602
+ "n_ep": 5,
1603
+ "n_success": 4,
1604
+ "success_rate": 0.8,
1605
+ "exit_code": "0",
1606
+ "episodes": [
1607
+ {
1608
+ "idx": 1,
1609
+ "return": 280.47,
1610
+ "length": 600,
1611
+ "success": false
1612
+ },
1613
+ {
1614
+ "idx": 2,
1615
+ "return": 111.85,
1616
+ "length": 136,
1617
+ "success": true
1618
+ },
1619
+ {
1620
+ "idx": 3,
1621
+ "return": 126.84,
1622
+ "length": 143,
1623
+ "success": true
1624
+ },
1625
+ {
1626
+ "idx": 4,
1627
+ "return": 78.51,
1628
+ "length": 93,
1629
+ "success": true
1630
+ },
1631
+ {
1632
+ "idx": 5,
1633
+ "return": 83.36,
1634
+ "length": 93,
1635
+ "success": true
1636
+ }
1637
+ ]
1638
+ },
1639
+ "Pant_Short_Seen_5": {
1640
+ "n_ep": 5,
1641
+ "n_success": 5,
1642
+ "success_rate": 1.0,
1643
+ "exit_code": "0",
1644
+ "episodes": [
1645
+ {
1646
+ "idx": 1,
1647
+ "return": 181.73,
1648
+ "length": 276,
1649
+ "success": true
1650
+ },
1651
+ {
1652
+ "idx": 2,
1653
+ "return": 121.7,
1654
+ "length": 143,
1655
+ "success": true
1656
+ },
1657
+ {
1658
+ "idx": 3,
1659
+ "return": 123.96,
1660
+ "length": 143,
1661
+ "success": true
1662
+ },
1663
+ {
1664
+ "idx": 4,
1665
+ "return": 126.77,
1666
+ "length": 143,
1667
+ "success": true
1668
+ },
1669
+ {
1670
+ "idx": 5,
1671
+ "return": 136.98,
1672
+ "length": 143,
1673
+ "success": true
1674
+ }
1675
+ ]
1676
+ },
1677
+ "Pant_Short_Seen_6": {
1678
+ "n_ep": 5,
1679
+ "n_success": 5,
1680
+ "success_rate": 1.0,
1681
+ "exit_code": "0",
1682
+ "episodes": [
1683
+ {
1684
+ "idx": 1,
1685
+ "return": 108.79,
1686
+ "length": 126,
1687
+ "success": true
1688
+ },
1689
+ {
1690
+ "idx": 2,
1691
+ "return": 125.49,
1692
+ "length": 143,
1693
+ "success": true
1694
+ },
1695
+ {
1696
+ "idx": 3,
1697
+ "return": 125.42,
1698
+ "length": 143,
1699
+ "success": true
1700
+ },
1701
+ {
1702
+ "idx": 4,
1703
+ "return": 122.69,
1704
+ "length": 143,
1705
+ "success": true
1706
+ },
1707
+ {
1708
+ "idx": 5,
1709
+ "return": 119.36,
1710
+ "length": 143,
1711
+ "success": true
1712
+ }
1713
+ ]
1714
+ },
1715
+ "Pant_Short_Seen_7": {
1716
+ "n_ep": 5,
1717
+ "n_success": 5,
1718
+ "success_rate": 1.0,
1719
+ "exit_code": "0",
1720
+ "episodes": [
1721
+ {
1722
+ "idx": 1,
1723
+ "return": 121.96,
1724
+ "length": 126,
1725
+ "success": true
1726
+ },
1727
+ {
1728
+ "idx": 2,
1729
+ "return": 130.41,
1730
+ "length": 143,
1731
+ "success": true
1732
+ },
1733
+ {
1734
+ "idx": 3,
1735
+ "return": 129.93,
1736
+ "length": 143,
1737
+ "success": true
1738
+ },
1739
+ {
1740
+ "idx": 4,
1741
+ "return": 131.86,
1742
+ "length": 143,
1743
+ "success": true
1744
+ },
1745
+ {
1746
+ "idx": 5,
1747
+ "return": 127.93,
1748
+ "length": 143,
1749
+ "success": true
1750
+ }
1751
+ ]
1752
+ },
1753
+ "Pant_Short_Seen_8": {
1754
+ "n_ep": 5,
1755
+ "n_success": 5,
1756
+ "success_rate": 1.0,
1757
+ "exit_code": "0",
1758
+ "episodes": [
1759
+ {
1760
+ "idx": 1,
1761
+ "return": 125.91,
1762
+ "length": 126,
1763
+ "success": true
1764
+ },
1765
+ {
1766
+ "idx": 2,
1767
+ "return": 146.37,
1768
+ "length": 143,
1769
+ "success": true
1770
+ },
1771
+ {
1772
+ "idx": 3,
1773
+ "return": 146.8,
1774
+ "length": 143,
1775
+ "success": true
1776
+ },
1777
+ {
1778
+ "idx": 4,
1779
+ "return": 91.86,
1780
+ "length": 93,
1781
+ "success": true
1782
+ },
1783
+ {
1784
+ "idx": 5,
1785
+ "return": 101.16,
1786
+ "length": 93,
1787
+ "success": true
1788
+ }
1789
+ ]
1790
+ },
1791
+ "Pant_Short_Seen_9": {
1792
+ "n_ep": 5,
1793
+ "n_success": 5,
1794
+ "success_rate": 1.0,
1795
+ "exit_code": "0",
1796
+ "episodes": [
1797
+ {
1798
+ "idx": 1,
1799
+ "return": 126.53,
1800
+ "length": 126,
1801
+ "success": true
1802
+ },
1803
+ {
1804
+ "idx": 2,
1805
+ "return": 142.1,
1806
+ "length": 143,
1807
+ "success": true
1808
+ },
1809
+ {
1810
+ "idx": 3,
1811
+ "return": 142.54,
1812
+ "length": 143,
1813
+ "success": true
1814
+ },
1815
+ {
1816
+ "idx": 4,
1817
+ "return": 140.78,
1818
+ "length": 143,
1819
+ "success": true
1820
+ },
1821
+ {
1822
+ "idx": 5,
1823
+ "return": 159.01,
1824
+ "length": 143,
1825
+ "success": true
1826
+ }
1827
+ ]
1828
+ },
1829
+ "Pant_Short_Unseen_0": {
1830
+ "n_ep": 5,
1831
+ "n_success": 1,
1832
+ "success_rate": 0.2,
1833
+ "exit_code": "0",
1834
+ "episodes": [
1835
+ {
1836
+ "idx": 1,
1837
+ "return": 253.11,
1838
+ "length": 600,
1839
+ "success": false
1840
+ },
1841
+ {
1842
+ "idx": 2,
1843
+ "return": 315.71,
1844
+ "length": 600,
1845
+ "success": false
1846
+ },
1847
+ {
1848
+ "idx": 3,
1849
+ "return": 223.21,
1850
+ "length": 600,
1851
+ "success": false
1852
+ },
1853
+ {
1854
+ "idx": 4,
1855
+ "return": 281.93,
1856
+ "length": 600,
1857
+ "success": false
1858
+ },
1859
+ {
1860
+ "idx": 5,
1861
+ "return": 272.08,
1862
+ "length": 366,
1863
+ "success": true
1864
+ }
1865
+ ]
1866
+ },
1867
+ "Pant_Short_Unseen_1": {
1868
+ "n_ep": 5,
1869
+ "n_success": 5,
1870
+ "success_rate": 1.0,
1871
+ "exit_code": "0",
1872
+ "episodes": [
1873
+ {
1874
+ "idx": 1,
1875
+ "return": 106.5,
1876
+ "length": 126,
1877
+ "success": true
1878
+ },
1879
+ {
1880
+ "idx": 2,
1881
+ "return": 98.05,
1882
+ "length": 143,
1883
+ "success": true
1884
+ },
1885
+ {
1886
+ "idx": 3,
1887
+ "return": 100.37,
1888
+ "length": 143,
1889
+ "success": true
1890
+ },
1891
+ {
1892
+ "idx": 4,
1893
+ "return": 106.63,
1894
+ "length": 143,
1895
+ "success": true
1896
+ },
1897
+ {
1898
+ "idx": 5,
1899
+ "return": 108.08,
1900
+ "length": 143,
1901
+ "success": true
1902
+ }
1903
+ ]
1904
+ }
1905
+ }
1906
+ }
1907
+ },
1908
+ "overall": {
1909
+ "total_ep": 240,
1910
+ "total_success": 150,
1911
+ "success_rate": 0.625,
1912
+ "seen": {
1913
+ "ep": 200,
1914
+ "success": 136,
1915
+ "rate": 0.68
1916
+ },
1917
+ "unseen": {
1918
+ "ep": 40,
1919
+ "success": 14,
1920
+ "rate": 0.35
1921
+ }
1922
+ }
1923
+ }
README.md ADDED
@@ -0,0 +1,129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # LeHome Challenge 2026 — Submission
2
+
3
+ **Method**: SmolVLA (frozen four_types 30K backbone) + state-only residual MLP, trained with sparse-reward residual RL on 40 Seen garments. Single model, deterministic inference.
4
+
5
+ ## File Layout
6
+
7
+ ```
8
+ submission_v4_global/
9
+ ├── README.md # this file
10
+ ├── residual_v4_global.py # the policy module (drop into eval_policy/)
11
+ └── submission_models/
12
+ ├── vla_backbone/ # SmolVLA four_types 30K (LeRobot pretrained_model)
13
+ ├── residual_averaged.pt # 40-garment averaged residual MLP
14
+ └── dataset_meta/ # LeRobot dataset metadata (stats.json etc.)
15
+ ```
16
+
17
+ ## How To Run (evaluator side)
18
+
19
+ 1. Drop `residual_v4_global.py` into `/opt/lehome-challenge/scripts/eval_policy/`.
20
+ 2. Add to `scripts/eval_policy/__init__.py`:
21
+ ```python
22
+ from .residual_v4_global import ResidualV4GlobalPolicy
23
+ ```
24
+ 3. Set environment variables:
25
+ ```bash
26
+ export LEHOME_VLA_POLICY_PATH=<path to submission_models/vla_backbone>
27
+ export LEHOME_VLA_DATASET_ROOT=<path to submission_models/dataset_meta or any LeRobot dataset>
28
+ export LEHOME_RESIDUAL_CHECKPOINT=<path to submission_models/residual_averaged.pt>
29
+ export LEHOME_RESIDUAL_SCALE=0.03
30
+ export HF_HUB_OFFLINE=1
31
+ export TRANSFORMERS_OFFLINE=1
32
+ ```
33
+ 4. Invoke evaluator:
34
+ ```bash
35
+ python -m scripts.eval \
36
+ --policy_type residual_v4_global \
37
+ --policy_path "$LEHOME_VLA_POLICY_PATH" \
38
+ --dataset_root "$LEHOME_VLA_DATASET_ROOT" \
39
+ --garment_type <top_long|top_short|pant_long|pant_short> \
40
+ --num_episodes 5 --max_steps 600 \
41
+ --enable_cameras --device cpu --headless
42
+ ```
43
+
44
+ ## Method Summary
45
+
46
+ - **Backbone**: SmolVLA, jointly-trained on 4 garment types for 30K steps. Frozen during residual RL.
47
+ - **Residual**: small state-only MLP (state_dim=12 → 256 → 256 → action_dim=12, 3 Linear+ReLU layers).
48
+ - **Final action**: `clip(base_action + 0.03 * residual_mlp(state))`.
49
+ - **Training signal**: sparse reward (1 if folding success at episode end, else 0).
50
+ - **Training data**: 40 Seen garments (10 per type × 4 types), 30 episodes per garment, on-policy PPO updates.
51
+ - **Aggregation**: weights averaged across 40 per-garment training runs to get a single global residual.
52
+ - **Inference**: deterministic — no exploration noise, no online updates.
53
+
54
+ ## Key Hyperparameters
55
+
56
+ | Parameter | Value |
57
+ |---|---|
58
+ | residual hidden dims | (256, 256) |
59
+ | residual scale | 0.03 |
60
+ | state_dim | 12 |
61
+ | action_dim | 12 |
62
+ | training reward | sparse (1 on success) |
63
+ | episodes per garment | 30 |
64
+ | training garments | 40 (10 Seen × 4 types) |
65
+
66
+ ## Evaluation Results
67
+
68
+ Run on `lehome3 / 120.209.70.195:30239`, 4× NVIDIA L40S, 4-GPU parallel.
69
+ 48 garments × 5 episodes = 240 episodes total.
70
+
71
+ | Metric | Value |
72
+ |---|---|
73
+ | **Total** | **150/240 = 62.50%** |
74
+ | Seen (40 garments × 5 ep) | 136/200 = 68.00% |
75
+ | Unseen (8 garments × 5 ep) | 14/40 = 35.00% |
76
+ | Top_Long | 43/60 = 71.67% (seen 74.0%, unseen 60.0%) |
77
+ | Top_Short | 25/60 = 41.67% (seen 48.0%, unseen 10.0%) |
78
+ | Pant_Long | 28/60 = 46.67% (seen 54.0%, unseen 10.0%) |
79
+ | **Pant_Short** | **54/60 = 90.00%** (seen 96.0%, unseen 60.0%) |
80
+
81
+ ### Reference baselines
82
+
83
+ | Method | Total | Notes |
84
+ |---|---|---|
85
+ | **This submission** (v4 global, deterministic) | **62.50%** | 240 ep, 4 types, single model |
86
+ | SmolVLA four_types 30K (no residual) | 60.42% | 96 ep, baseline backbone alone |
87
+ | Historical v4 global with `explore=True` | 58.75% | 240 ep, non-deterministic |
88
+ | ACT (single-type, top_long only) | 87.50% | 24 ep, not comparable across types |
89
+
90
+ The +1.71pp gain over the SmolVLA backbone alone confirms the residual carries useful signal.
91
+ The +3.75pp gain over the historical noisy run confirms determinism matters.
92
+
93
+ ### Notes on the run
94
+
95
+ - One garment (`Top_Long_Seen_9`) initially failed in the main parallel sweep with an Isaac Sim
96
+ `TiledCamera._annotators` AttributeError (unrelated to the policy). It was retried in a
97
+ fresh single-garment process and produced 4/5 success — that retry is included in the
98
+ `150/240` figure above.
99
+ - Episode-level data per garment is in `FINAL_SUMMARY.json` and per-garment stdout logs
100
+ are in `eval_raw/`.
101
+
102
+ ## Artefact Hashes
103
+
104
+ | Artefact | sha256 |
105
+ |---|---|
106
+ | residual_averaged.pt | `9d695e278b4361509ac7e35f7d66eb251ec7e7f1f7c53878d453ef2b8aa0ce74` |
107
+ | vla_backbone/model.safetensors | `7ff3915571622bf7530e9ba35540abf5c14f62d8c6a57491664b65a23869e6bc` |
108
+
109
+ ## Reproducibility
110
+
111
+ Inference is fully deterministic. Two runs with the same backbone, residual checkpoint, and `seed=42` (default) yield identical action sequences. Variability across runs comes only from Isaac Sim particle initialization (seeded by `--seed`).
112
+
113
+ ## Notes
114
+
115
+ - The `residual_averaged.pt` follows the format:
116
+ ```python
117
+ torch.save({
118
+ "state_dim": 12,
119
+ "action_dim": 12,
120
+ "hidden_dims": (256, 256),
121
+ "model_state_dict": <state-only MLP weights>,
122
+ }, path)
123
+ ```
124
+ - This submission is a **single model** (one residual checkpoint) handling all four garment types — not a per-type specialist ensemble.
125
+ - Inference path: `LeRobotPolicy.select_action(observation)` → adds `0.03 * residual_mlp(observation['observation.state'])`.
126
+
127
+ ## Contact
128
+
129
+ vita / realvitacai@gmail.com
RESULTS_SUMMARY.txt ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ v4 global residual sparse RL — single model deterministic eval
2
+ Total: 150/240 = 62.50%
3
+ Seen: 136/200 = 68.00%
4
+ Unseen: 14/40 = 35.00%
5
+
6
+ Per type:
7
+ Top_Long: 43/60 = 71.67% (seen 74.0% / unseen 60.0%)
8
+ Top_Short: 25/60 = 41.67% (seen 48.0% / unseen 10.0%)
9
+ Pant_Long: 28/60 = 46.67% (seen 54.0% / unseen 10.0%)
10
+ Pant_Short: 54/60 = 90.00% (seen 96.0% / unseen 60.0%)
11
+ Reference baseline:
12
+ SmolVLA four_types 30K (no residual): 60.42% (96 ep, 4 types)
13
+ Historical v4 global with explore=True: 58.75% (240 ep)
residual_v4_global.py ADDED
@@ -0,0 +1,131 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """LeHome Challenge 2026 - single-model residual policy (v4 global).
2
+
3
+ Loads SmolVLA backbone (frozen) and applies a state-only residual MLP
4
+ averaged from 40 Seen garments under sparse-reward residual RL.
5
+ Inference is deterministic: no exploration noise, no online updates.
6
+
7
+ Required env vars:
8
+ LEHOME_VLA_POLICY_PATH SmolVLA backbone directory
9
+ LEHOME_VLA_DATASET_ROOT LeRobot dataset root
10
+ LEHOME_RESIDUAL_CHECKPOINT path to residual_averaged.pt
11
+ Optional:
12
+ LEHOME_RESIDUAL_SCALE default 0.03
13
+ LEHOME_TASK_DESCRIPTION default fold the garment on the table
14
+ LEHOME_POLICY_DEVICE default cuda:0
15
+ """
16
+ from __future__ import annotations
17
+
18
+ import os
19
+ from pathlib import Path
20
+ from typing import Dict, Optional
21
+
22
+ import numpy as np
23
+ import torch
24
+ from torch import nn
25
+
26
+ from .base_policy import BasePolicy
27
+ from .lerobot_policy import LeRobotPolicy
28
+ from .registry import PolicyRegistry
29
+
30
+
31
+ class _ResidualActor(nn.Module):
32
+ def __init__(self, state_dim: int, action_dim: int, hidden_dims=(256, 256)):
33
+ super().__init__()
34
+ dims = [state_dim, *hidden_dims, action_dim]
35
+ layers: list[nn.Module] = []
36
+ for i in range(len(dims) - 2):
37
+ layers.append(nn.Linear(dims[i], dims[i + 1]))
38
+ layers.append(nn.ReLU())
39
+ layers.append(nn.Linear(dims[-2], dims[-1]))
40
+ self.network = nn.Sequential(*layers)
41
+
42
+ def forward(self, state: torch.Tensor) -> torch.Tensor:
43
+ return self.network(state)
44
+
45
+
46
+ @PolicyRegistry.register("residual_v4_global")
47
+ class ResidualV4GlobalPolicy(BasePolicy):
48
+ """SmolVLA + state-only residual (single model, deterministic)."""
49
+
50
+ def __init__(
51
+ self,
52
+ policy_path: Optional[str] = None,
53
+ model_path: Optional[str] = None,
54
+ dataset_root: Optional[str] = None,
55
+ task_description: str = "fold the garment on the table",
56
+ device: str = "cuda",
57
+ **kwargs,
58
+ ):
59
+ super().__init__(**kwargs)
60
+
61
+ backbone_path = (
62
+ os.environ.get("LEHOME_VLA_POLICY_PATH")
63
+ or policy_path
64
+ or model_path
65
+ or ""
66
+ )
67
+ ds_root = os.environ.get("LEHOME_VLA_DATASET_ROOT") or dataset_root or ""
68
+ residual_ckpt = os.environ.get("LEHOME_RESIDUAL_CHECKPOINT", "")
69
+ residual_scale = float(os.environ.get("LEHOME_RESIDUAL_SCALE", "0.03"))
70
+ task_desc = os.environ.get("LEHOME_TASK_DESCRIPTION", task_description)
71
+ policy_device = os.environ.get(
72
+ "LEHOME_POLICY_DEVICE",
73
+ "cuda:0" if torch.cuda.is_available() else "cpu",
74
+ )
75
+
76
+ if not backbone_path or not Path(backbone_path).exists():
77
+ raise FileNotFoundError(
78
+ f"VLA backbone not found (set LEHOME_VLA_POLICY_PATH): {backbone_path}"
79
+ )
80
+ if not ds_root or not Path(ds_root).exists():
81
+ raise FileNotFoundError(
82
+ f"Dataset root not found (set LEHOME_VLA_DATASET_ROOT): {ds_root}"
83
+ )
84
+ if not residual_ckpt or not Path(residual_ckpt).exists():
85
+ raise FileNotFoundError(
86
+ f"Residual checkpoint not found (set LEHOME_RESIDUAL_CHECKPOINT): {residual_ckpt}"
87
+ )
88
+
89
+ self.policy_device = torch.device(policy_device)
90
+ self.residual_scale = residual_scale
91
+
92
+ self.base_policy = LeRobotPolicy(
93
+ policy_path=backbone_path,
94
+ dataset_root=ds_root,
95
+ task_description=task_desc,
96
+ device=str(self.policy_device),
97
+ )
98
+
99
+ payload = torch.load(residual_ckpt, map_location="cpu")
100
+ self.state_dim = int(payload["state_dim"])
101
+ self.action_dim = int(payload["action_dim"])
102
+ hidden_dims = tuple(payload["hidden_dims"])
103
+ self.actor = _ResidualActor(
104
+ state_dim=self.state_dim,
105
+ action_dim=self.action_dim,
106
+ hidden_dims=hidden_dims,
107
+ ).to(self.policy_device)
108
+ self.actor.load_state_dict(payload["model_state_dict"])
109
+ self.actor.eval()
110
+
111
+ print(
112
+ f"[ResidualV4GlobalPolicy] backbone={backbone_path} "
113
+ f"residual={residual_ckpt} scale={self.residual_scale} "
114
+ f"state_dim={self.state_dim} action_dim={self.action_dim} hidden={hidden_dims}"
115
+ )
116
+
117
+ def reset(self):
118
+ self.base_policy.reset()
119
+
120
+ def select_action(self, observation: Dict[str, np.ndarray]) -> np.ndarray:
121
+ base_action = self.base_policy.select_action(observation).astype(np.float32, copy=False)
122
+
123
+ state_np = observation.get(
124
+ "observation.state", np.zeros(self.state_dim, dtype=np.float32)
125
+ ).astype(np.float32, copy=False)
126
+ state = torch.from_numpy(state_np).to(self.policy_device)
127
+
128
+ with torch.inference_mode():
129
+ residual = self.actor(state.unsqueeze(0)).squeeze(0).cpu().numpy().astype(np.float32)
130
+
131
+ return (base_action + self.residual_scale * residual).astype(np.float32, copy=False)