hmkang commited on
Commit
641edb2
·
verified ·
1 Parent(s): dd441ae

metrics: RoboCasa Kitchen 24-task success rate, ckpt 60k + 80k (EMA, seed42av, N_ENVS=1)

Browse files
Files changed (1) hide show
  1. metrics/eval_metrics.json +317 -0
metrics/eval_metrics.json ADDED
@@ -0,0 +1,317 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "protocol": {
3
+ "benchmark": "RoboCasa Kitchen, 24 tasks x 50 episodes (1200 total), env layout 5, N_ENVS=1",
4
+ "weights": "EVAL_WEIGHTS=ema -- EMA video DiT + EMA action head, loaded by key-swap (843 + 247 tensors, src==dst verified)",
5
+ "seeds": "WAM_EVAL_ACTION_SEED=42 WAM_EVAL_VIDEO_SEED=42 (seed42av cell)",
6
+ "budget": "50 episodes/task; dropcur WAM_DITACTION_EXCLUDE_CURRENT=1, matching the training yaml",
7
+ "code": "branch ctxpool-bmethod @ eb321c8 (pinned eval tree). That commit also carries the attach_ctx_pool fix WITHOUT which no pooled checkpoint loads at eval at all (meta-device + fp32-on-bf16).",
8
+ "render_bug": "N_ENVS=1 deliberately: at N_ENVS=5 the EGL buffer-swap bug mis-serves 5.9% of render calls vs 0.05% at 1. huiwon's detector scanned the 60k cell: 4/1200 episodes contaminated (0.3%), ALL of them already failures, so the reported number can only move UP by at most 0.33pp.",
9
+ "measured": "2026-09-13 / 2026-09-14, A100-80GB, kakao"
10
+ },
11
+ "robocasa_kitchen": {
12
+ "60000": {
13
+ "overall_success_rate": 0.6192,
14
+ "successes": 743,
15
+ "episodes": 1200,
16
+ "tasks_complete": 24,
17
+ "per_task": {
18
+ "TurnSinkSpout": {
19
+ "task_index": 0,
20
+ "successes": 31,
21
+ "episodes": 50,
22
+ "success_rate": 0.62
23
+ },
24
+ "TurnOnStove": {
25
+ "task_index": 1,
26
+ "successes": 13,
27
+ "episodes": 50,
28
+ "success_rate": 0.26
29
+ },
30
+ "TurnOnSinkFaucet": {
31
+ "task_index": 2,
32
+ "successes": 34,
33
+ "episodes": 50,
34
+ "success_rate": 0.68
35
+ },
36
+ "TurnOnMicrowave": {
37
+ "task_index": 3,
38
+ "successes": 44,
39
+ "episodes": 50,
40
+ "success_rate": 0.88
41
+ },
42
+ "TurnOffStove": {
43
+ "task_index": 4,
44
+ "successes": 10,
45
+ "episodes": 50,
46
+ "success_rate": 0.2
47
+ },
48
+ "TurnOffSinkFaucet": {
49
+ "task_index": 5,
50
+ "successes": 47,
51
+ "episodes": 50,
52
+ "success_rate": 0.94
53
+ },
54
+ "TurnOffMicrowave": {
55
+ "task_index": 6,
56
+ "successes": 50,
57
+ "episodes": 50,
58
+ "success_rate": 1.0
59
+ },
60
+ "PnPStoveToCounter": {
61
+ "task_index": 7,
62
+ "successes": 23,
63
+ "episodes": 50,
64
+ "success_rate": 0.46
65
+ },
66
+ "PnPSinkToCounter": {
67
+ "task_index": 8,
68
+ "successes": 31,
69
+ "episodes": 50,
70
+ "success_rate": 0.62
71
+ },
72
+ "PnPMicrowaveToCounter": {
73
+ "task_index": 9,
74
+ "successes": 19,
75
+ "episodes": 50,
76
+ "success_rate": 0.38
77
+ },
78
+ "PnPCounterToStove": {
79
+ "task_index": 10,
80
+ "successes": 20,
81
+ "episodes": 50,
82
+ "success_rate": 0.4
83
+ },
84
+ "PnPCounterToSink": {
85
+ "task_index": 11,
86
+ "successes": 28,
87
+ "episodes": 50,
88
+ "success_rate": 0.56
89
+ },
90
+ "PnPCounterToMicrowave": {
91
+ "task_index": 12,
92
+ "successes": 16,
93
+ "episodes": 50,
94
+ "success_rate": 0.32
95
+ },
96
+ "PnPCounterToCab": {
97
+ "task_index": 13,
98
+ "successes": 25,
99
+ "episodes": 50,
100
+ "success_rate": 0.5
101
+ },
102
+ "PnPCabToCounter": {
103
+ "task_index": 14,
104
+ "successes": 25,
105
+ "episodes": 50,
106
+ "success_rate": 0.5
107
+ },
108
+ "OpenSingleDoor": {
109
+ "task_index": 15,
110
+ "successes": 33,
111
+ "episodes": 50,
112
+ "success_rate": 0.66
113
+ },
114
+ "OpenDrawer": {
115
+ "task_index": 16,
116
+ "successes": 34,
117
+ "episodes": 50,
118
+ "success_rate": 0.68
119
+ },
120
+ "OpenDoubleDoor": {
121
+ "task_index": 17,
122
+ "successes": 27,
123
+ "episodes": 50,
124
+ "success_rate": 0.54
125
+ },
126
+ "CoffeeSetupMug": {
127
+ "task_index": 18,
128
+ "successes": 18,
129
+ "episodes": 50,
130
+ "success_rate": 0.36
131
+ },
132
+ "CoffeeServeMug": {
133
+ "task_index": 19,
134
+ "successes": 38,
135
+ "episodes": 50,
136
+ "success_rate": 0.76
137
+ },
138
+ "CoffeePressButton": {
139
+ "task_index": 20,
140
+ "successes": 42,
141
+ "episodes": 50,
142
+ "success_rate": 0.84
143
+ },
144
+ "CloseSingleDoor": {
145
+ "task_index": 21,
146
+ "successes": 49,
147
+ "episodes": 50,
148
+ "success_rate": 0.98
149
+ },
150
+ "CloseDrawer": {
151
+ "task_index": 22,
152
+ "successes": 49,
153
+ "episodes": 50,
154
+ "success_rate": 0.98
155
+ },
156
+ "CloseDoubleDoor": {
157
+ "task_index": 23,
158
+ "successes": 37,
159
+ "episodes": 50,
160
+ "success_rate": 0.74
161
+ }
162
+ }
163
+ },
164
+ "80000": {
165
+ "overall_success_rate": 0.645,
166
+ "successes": 774,
167
+ "episodes": 1200,
168
+ "tasks_complete": 24,
169
+ "per_task": {
170
+ "TurnSinkSpout": {
171
+ "task_index": 0,
172
+ "successes": 40,
173
+ "episodes": 50,
174
+ "success_rate": 0.8
175
+ },
176
+ "TurnOnStove": {
177
+ "task_index": 1,
178
+ "successes": 13,
179
+ "episodes": 50,
180
+ "success_rate": 0.26
181
+ },
182
+ "TurnOnSinkFaucet": {
183
+ "task_index": 2,
184
+ "successes": 42,
185
+ "episodes": 50,
186
+ "success_rate": 0.84
187
+ },
188
+ "TurnOnMicrowave": {
189
+ "task_index": 3,
190
+ "successes": 45,
191
+ "episodes": 50,
192
+ "success_rate": 0.9
193
+ },
194
+ "TurnOffStove": {
195
+ "task_index": 4,
196
+ "successes": 9,
197
+ "episodes": 50,
198
+ "success_rate": 0.18
199
+ },
200
+ "TurnOffSinkFaucet": {
201
+ "task_index": 5,
202
+ "successes": 48,
203
+ "episodes": 50,
204
+ "success_rate": 0.96
205
+ },
206
+ "TurnOffMicrowave": {
207
+ "task_index": 6,
208
+ "successes": 50,
209
+ "episodes": 50,
210
+ "success_rate": 1.0
211
+ },
212
+ "PnPStoveToCounter": {
213
+ "task_index": 7,
214
+ "successes": 23,
215
+ "episodes": 50,
216
+ "success_rate": 0.46
217
+ },
218
+ "PnPSinkToCounter": {
219
+ "task_index": 8,
220
+ "successes": 31,
221
+ "episodes": 50,
222
+ "success_rate": 0.62
223
+ },
224
+ "PnPMicrowaveToCounter": {
225
+ "task_index": 9,
226
+ "successes": 10,
227
+ "episodes": 50,
228
+ "success_rate": 0.2
229
+ },
230
+ "PnPCounterToStove": {
231
+ "task_index": 10,
232
+ "successes": 27,
233
+ "episodes": 50,
234
+ "success_rate": 0.54
235
+ },
236
+ "PnPCounterToSink": {
237
+ "task_index": 11,
238
+ "successes": 28,
239
+ "episodes": 50,
240
+ "success_rate": 0.56
241
+ },
242
+ "PnPCounterToMicrowave": {
243
+ "task_index": 12,
244
+ "successes": 18,
245
+ "episodes": 50,
246
+ "success_rate": 0.36
247
+ },
248
+ "PnPCounterToCab": {
249
+ "task_index": 13,
250
+ "successes": 23,
251
+ "episodes": 50,
252
+ "success_rate": 0.46
253
+ },
254
+ "PnPCabToCounter": {
255
+ "task_index": 14,
256
+ "successes": 27,
257
+ "episodes": 50,
258
+ "success_rate": 0.54
259
+ },
260
+ "OpenSingleDoor": {
261
+ "task_index": 15,
262
+ "successes": 36,
263
+ "episodes": 50,
264
+ "success_rate": 0.72
265
+ },
266
+ "OpenDrawer": {
267
+ "task_index": 16,
268
+ "successes": 40,
269
+ "episodes": 50,
270
+ "success_rate": 0.8
271
+ },
272
+ "OpenDoubleDoor": {
273
+ "task_index": 17,
274
+ "successes": 32,
275
+ "episodes": 50,
276
+ "success_rate": 0.64
277
+ },
278
+ "CoffeeSetupMug": {
279
+ "task_index": 18,
280
+ "successes": 11,
281
+ "episodes": 50,
282
+ "success_rate": 0.22
283
+ },
284
+ "CoffeeServeMug": {
285
+ "task_index": 19,
286
+ "successes": 34,
287
+ "episodes": 50,
288
+ "success_rate": 0.68
289
+ },
290
+ "CoffeePressButton": {
291
+ "task_index": 20,
292
+ "successes": 46,
293
+ "episodes": 50,
294
+ "success_rate": 0.92
295
+ },
296
+ "CloseSingleDoor": {
297
+ "task_index": 21,
298
+ "successes": 48,
299
+ "episodes": 50,
300
+ "success_rate": 0.96
301
+ },
302
+ "CloseDrawer": {
303
+ "task_index": 22,
304
+ "successes": 50,
305
+ "episodes": 50,
306
+ "success_rate": 1.0
307
+ },
308
+ "CloseDoubleDoor": {
309
+ "task_index": 23,
310
+ "successes": 43,
311
+ "episodes": 50,
312
+ "success_rate": 0.86
313
+ }
314
+ }
315
+ }
316
+ }
317
+ }