HyeonseokE commited on
Commit
5f96189
·
verified ·
1 Parent(s): e0b0d46

Upload UR7e Sort RGBBlock DCT SmolVLA Phase2 artifacts

Browse files
MANIFEST.json ADDED
@@ -0,0 +1,292 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "checkpoint": {
3
+ "batch_size": 16,
4
+ "dataset_repo_id": "CoRL2026-CSI/UR7e_phase1_Sort_RGBblock_10fps",
5
+ "empty_cameras": 1,
6
+ "gradient_accumulation_steps": 1,
7
+ "local_bundle_path": "pretrained_model",
8
+ "output_features": {
9
+ "action": {
10
+ "shape": [
11
+ 7
12
+ ],
13
+ "type": "ACTION"
14
+ }
15
+ },
16
+ "rename_map": {
17
+ "observation.images.realsense_topview": "observation.images.camera2",
18
+ "observation.images.realsense_wrist": "observation.images.camera1"
19
+ },
20
+ "skill_dct_parquet": "/home/csiwoo/CORL2026/ur7e_autodata_workspace/runtime/AutoDataCollector-ur7e/results/skill_dct/UR7e_phase1_Sort_RGBblock_10fps.canonical.parquet",
21
+ "train_steps": 3000
22
+ },
23
+ "created_at_unix": 1779939970,
24
+ "files": [
25
+ {
26
+ "bytes": 4519,
27
+ "path": "config/phase2_config.yaml",
28
+ "sha256": "d3221b62e48e829828a3d45535d87a3942513758493b77689262c665fb3ea746"
29
+ },
30
+ {
31
+ "bytes": 984,
32
+ "path": "config/recording_config_ur7e_phase2.yaml",
33
+ "sha256": "f7915a2bdebbf67e2e90f0f32710fc539b422f1b0a33031d478af85cf30a0c7e"
34
+ },
35
+ {
36
+ "bytes": 2645,
37
+ "path": "pretrained_model/config.json",
38
+ "sha256": "f6f994dcaea938d2602dac3bbe77aff1170ac5029446a24483a58a2878f412a2"
39
+ },
40
+ {
41
+ "bytes": 906712520,
42
+ "path": "pretrained_model/model.safetensors",
43
+ "sha256": "b3f36316a140b608f1404d823bc51c300858332200ec2496b4079d2b565d4f3b"
44
+ },
45
+ {
46
+ "bytes": 660,
47
+ "path": "pretrained_model/policy_postprocessor.json",
48
+ "sha256": "3b51f092c70c710ce0213ee1b63bf51b4878ec67828dd2d67daf9ef51081a41a"
49
+ },
50
+ {
51
+ "bytes": 15124,
52
+ "path": "pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors",
53
+ "sha256": "9bb5dbd493da25bfa7f225edcf47d85c85788fad965ebd86aa87877cb9d674e5"
54
+ },
55
+ {
56
+ "bytes": 2246,
57
+ "path": "pretrained_model/policy_preprocessor.json",
58
+ "sha256": "9f2f42a4a8ef6c09a5e453dc9fc8bdcff550c95d6d5b8fded8fcc343267da6ee"
59
+ },
60
+ {
61
+ "bytes": 15124,
62
+ "path": "pretrained_model/policy_preprocessor_step_5_normalizer_processor.safetensors",
63
+ "sha256": "9bb5dbd493da25bfa7f225edcf47d85c85788fad965ebd86aa87877cb9d674e5"
64
+ },
65
+ {
66
+ "bytes": 7427,
67
+ "path": "pretrained_model/train_config.json",
68
+ "sha256": "c6cb73590ed5ea777ef779318338256e268787ae1ee76499c0bfdadbd7434dc7"
69
+ },
70
+ {
71
+ "bytes": 1167,
72
+ "path": "session/dataset_judge_rejected_skips.json",
73
+ "sha256": "f40f273b4f93de4dc9d9d82881eb27373547c3e82b18aa4bf6ba725956a0ef2e"
74
+ },
75
+ {
76
+ "bytes": 506,
77
+ "path": "session/resume_append_status.json",
78
+ "sha256": "aec2fd8dc6f9398740decfe66b6bf51e97872f6f53c2cb5e1c1230126e030bb7"
79
+ },
80
+ {
81
+ "bytes": 44251,
82
+ "path": "session/session_summary.json",
83
+ "sha256": "a910888c0415da922aea4c5832829319af67039cc21b22cb6bd293a6e3947576"
84
+ },
85
+ {
86
+ "bytes": 302162,
87
+ "path": "session/subgoal_buffer.npz",
88
+ "sha256": "0f11ec4ac1aca73d9766e9af8611a5e2fe8bfac3389a4928c11e30e7acfd1697"
89
+ },
90
+ {
91
+ "bytes": 1361616,
92
+ "path": "skill_dct/UR7e_phase1_Sort_RGBblock_10fps.canonical.parquet",
93
+ "sha256": "6a3a60175a2e1462ea85b75f66e5e4ba599a249b81d084a0309821866ea1770f"
94
+ },
95
+ {
96
+ "bytes": 1358948,
97
+ "path": "skill_dct/UR7e_phase1_Sort_RGBblock_10fps.parquet",
98
+ "sha256": "d3fb466912f11a750c736c7b8cc29fb751775e9c15945b2db3dd10e1ec26b77a"
99
+ },
100
+ {
101
+ "bytes": 7102690,
102
+ "path": "vector_db/server_skill_wise_vector_db.npz",
103
+ "sha256": "9748b78a48c19e5bd87f865f220d7fa9d6141a7777b53855752d522926cce0ab"
104
+ },
105
+ {
106
+ "bytes": 7102690,
107
+ "path": "vector_db/skill_wise_vector_db.npz",
108
+ "sha256": "9748b78a48c19e5bd87f865f220d7fa9d6141a7777b53855752d522926cce0ab"
109
+ }
110
+ ],
111
+ "hf_model_repo": "CoRL2026-CSI/UR7e_Sort_RGBblock_DCT_SmolVLA_3000s",
112
+ "name": "UR7e Sort RGBBlock DCT-tuned SmolVLA Phase2 Artifacts",
113
+ "phase2": {
114
+ "buffer": {
115
+ "filename": "server_skill_wise_vector_db.npz",
116
+ "save_dir": "grpc_server/buffer"
117
+ },
118
+ "mi_selection": {
119
+ "coverage_key_mode": "proprio",
120
+ "k_min": 3,
121
+ "min_covered_windows": 1,
122
+ "q1_chosen_strategy": "argq25",
123
+ "radius_k": 8,
124
+ "radius_quantile": 0.9,
125
+ "selection_mode": "Q1"
126
+ },
127
+ "phase1_dataset_path": "CoRL2026-CSI/UR7e_phase1_Sort_RGBblock_10fps",
128
+ "phase1_trained_vla_path": "results/method3_embed_vla/smolvla_dct_ur7e_sort_rgbblock_canonical_3000s_20260528_050642/checkpoints/003000/pretrained_model",
129
+ "selector_skill_dct_parquet": "results/skill_dct/UR7e_phase1_Sort_RGBblock_10fps.canonical.parquet",
130
+ "transport": {
131
+ "address": "127.0.0.1:50061",
132
+ "debug_verbose": true,
133
+ "timeout_s": 60.0
134
+ }
135
+ },
136
+ "server_command_hint": "SKIP_SETUP=1 GPU_ID=0 ENV_NAME=gpu_server HOST=0.0.0.0 PORT=50061 RECORDING_CONFIG=pipeline_config/recording_config_ur7e_phase2.yaml URDF=assets/urdf/ur7e.urdf bash grpc_server/run_server.sh",
137
+ "source_dataset": "CoRL2026-CSI/UR7e_phase1_Sort_RGBblock_10fps",
138
+ "task": "Sort each colored block onto the dish of the matching color.",
139
+ "vector_db": {
140
+ "counts_per_skill": {
141
+ "skill_0": 20,
142
+ "skill_1": 20,
143
+ "skill_10": 20,
144
+ "skill_11": 20,
145
+ "skill_12": 20,
146
+ "skill_13": 20,
147
+ "skill_14": 20,
148
+ "skill_15": 20,
149
+ "skill_16": 20,
150
+ "skill_17": 20,
151
+ "skill_18": 20,
152
+ "skill_19": 20,
153
+ "skill_2": 20,
154
+ "skill_20": 20,
155
+ "skill_21": 20,
156
+ "skill_22": 20,
157
+ "skill_23": 20,
158
+ "skill_24": 20,
159
+ "skill_25": 20,
160
+ "skill_26": 20,
161
+ "skill_27": 20,
162
+ "skill_28": 20,
163
+ "skill_3": 20,
164
+ "skill_4": 20,
165
+ "skill_5": 20,
166
+ "skill_6": 20,
167
+ "skill_7": 20,
168
+ "skill_8": 20,
169
+ "skill_9": 20
170
+ },
171
+ "entries": 580,
172
+ "key_shapes_per_skill": {
173
+ "skill_0": [
174
+ 20,
175
+ 966
176
+ ],
177
+ "skill_1": [
178
+ 20,
179
+ 966
180
+ ],
181
+ "skill_10": [
182
+ 20,
183
+ 966
184
+ ],
185
+ "skill_11": [
186
+ 20,
187
+ 966
188
+ ],
189
+ "skill_12": [
190
+ 20,
191
+ 966
192
+ ],
193
+ "skill_13": [
194
+ 20,
195
+ 966
196
+ ],
197
+ "skill_14": [
198
+ 20,
199
+ 966
200
+ ],
201
+ "skill_15": [
202
+ 20,
203
+ 966
204
+ ],
205
+ "skill_16": [
206
+ 20,
207
+ 966
208
+ ],
209
+ "skill_17": [
210
+ 20,
211
+ 966
212
+ ],
213
+ "skill_18": [
214
+ 20,
215
+ 966
216
+ ],
217
+ "skill_19": [
218
+ 20,
219
+ 966
220
+ ],
221
+ "skill_2": [
222
+ 20,
223
+ 966
224
+ ],
225
+ "skill_20": [
226
+ 20,
227
+ 966
228
+ ],
229
+ "skill_21": [
230
+ 20,
231
+ 966
232
+ ],
233
+ "skill_22": [
234
+ 20,
235
+ 966
236
+ ],
237
+ "skill_23": [
238
+ 20,
239
+ 966
240
+ ],
241
+ "skill_24": [
242
+ 20,
243
+ 966
244
+ ],
245
+ "skill_25": [
246
+ 20,
247
+ 966
248
+ ],
249
+ "skill_26": [
250
+ 20,
251
+ 966
252
+ ],
253
+ "skill_27": [
254
+ 20,
255
+ 966
256
+ ],
257
+ "skill_28": [
258
+ 20,
259
+ 966
260
+ ],
261
+ "skill_3": [
262
+ 20,
263
+ 966
264
+ ],
265
+ "skill_4": [
266
+ 20,
267
+ 966
268
+ ],
269
+ "skill_5": [
270
+ 20,
271
+ 966
272
+ ],
273
+ "skill_6": [
274
+ 20,
275
+ 966
276
+ ],
277
+ "skill_7": [
278
+ 20,
279
+ 966
280
+ ],
281
+ "skill_8": [
282
+ 20,
283
+ 966
284
+ ],
285
+ "skill_9": [
286
+ 20,
287
+ 966
288
+ ]
289
+ },
290
+ "skills": 29
291
+ }
292
+ }
README.md ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - robotics
5
+ - ur7e
6
+ - smolvla
7
+ - method3
8
+ - dct
9
+ - phase2
10
+ pipeline_tag: robotics
11
+ ---
12
+
13
+ # UR7e Sort RGBBlock DCT-tuned SmolVLA 3000-step Checkpoint
14
+
15
+ This repository preserves the DCT-tuned SmolVLA checkpoint and Method3 Phase2 server artifacts for the UR7e Sort RGBBlock task.
16
+
17
+ ## Source
18
+
19
+ - Phase1 dataset: `CoRL2026-CSI/UR7e_phase1_Sort_RGBblock_10fps`
20
+ - Task: Sort each colored block onto the dish of the matching color.
21
+ - Training: `steps=3000`, `batch_size=16`, `gradient_accumulation_steps=1`, `empty_cameras=1`
22
+ - Camera rename: `realsense_wrist -> camera1`, `realsense_topview -> camera2`
23
+
24
+ ## Contents
25
+
26
+ - `pretrained_model/`: LeRobot/SmolVLA `from_pretrained` checkpoint.
27
+ - `skill_dct/`: raw and canonical skill DCT sidecar parquet files.
28
+ - `vector_db/`: Phase2 P_phase1 vector DB and server buffer DB.
29
+ - `session/`: minimal Phase1 session artifacts needed by UR7e Phase2 replay.
30
+ - `config/`: Phase2 and recording configs used for server startup.
31
+ - `MANIFEST.json`: file hashes, training metadata, and DB summary.
32
+
33
+ ## Verified DB Summary
34
+
35
+ - Skills: `29`
36
+ - Entries: `580`
37
+ - Per-skill entries: `20` to `20`
38
+ - Key shape example: `[20, 966]`
39
+
40
+ ## Server Startup Hint
41
+
42
+ ```bash
43
+ cd /home/csiwoo/CORL2026/ur7e_autodata_workspace/runtime/AutoDataCollector-ur7e
44
+ PATH=/home/csiwoo/robobridge/nvidia_580_126_09/usr/bin:$PATH \
45
+ LD_LIBRARY_PATH=/home/csiwoo/robobridge/nvidia_580_126_09/usr/lib/x86_64-linux-gnu:${LD_LIBRARY_PATH:-} \
46
+ SKIP_SETUP=1 GPU_ID=0 ENV_NAME=gpu_server HOST=0.0.0.0 PORT=50061 \
47
+ RECORDING_CONFIG=pipeline_config/recording_config_ur7e_phase2.yaml \
48
+ URDF=assets/urdf/ur7e.urdf \
49
+ bash grpc_server/run_server.sh
50
+ ```
config/phase2_config.yaml ADDED
@@ -0,0 +1,155 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # ============================================================
2
+ # Phase2 Config — UR7e Method3 MI-based Acquisition
3
+ # ============================================================
4
+
5
+ skill_planner_transport:
6
+ mode: grpc
7
+
8
+ phase1_trained_vla_path: "results/method3_embed_vla/smolvla_dct_ur7e_sort_rgbblock_canonical_3000s_20260528_050642/checkpoints/003000/pretrained_model"
9
+ phase1_dataset_path: "CoRL2026-CSI/UR7e_phase1_Sort_RGBblock_10fps"
10
+
11
+ budget: 100
12
+ phase1_min: 40
13
+ phase1_max: 60
14
+ tau_ready: 0.7
15
+ saturation_window: 10
16
+ enable_phase2_early_stop: true
17
+ dct_coeffs: 3
18
+ verbose: false
19
+
20
+ readiness:
21
+ k_min: 3
22
+ radius_k: 5
23
+ radius_quantile: 0.7
24
+
25
+ mi_selection:
26
+ selection_mode: Q1
27
+ q1_chosen_strategy: argq25
28
+ tau_U_strategy: null
29
+ k_nn_a: 5
30
+ q_quantile: 0.5
31
+ # UR7e Curobo candidates share observation-level VLA keys within a plan_batch.
32
+ # Gate support by trajectory proprio; keep VLA for U_VLA ranking.
33
+ coverage_key_mode: proprio
34
+ # Normalize M_MI over supported candidates only; under-covered candidates
35
+ # are still rejected and should not lift the Useful-OOD threshold.
36
+ mi_norm_scope: supported
37
+ # Median/MAD keeps the supported-candidate threshold robust to MI outliers.
38
+ mi_norm_center: median
39
+ mi_norm_scale: mad
40
+ mi_norm_mad_consistency: true
41
+ k_min: 3
42
+ radius_k: 8
43
+ radius_quantile: 0.9
44
+ s_min_a: 0.001
45
+ eps: 1.0e-6
46
+ beta: 1.0
47
+ lambda: 1.0
48
+ tau_MI: 0.0
49
+ amb_agg: mean
50
+ min_covered_windows: 1
51
+ debug_verbose: true
52
+ arm_dof: 6
53
+ # UR7e Phase1 DB state/action descriptors are already arm-only 6D.
54
+ # Keep VLA action-head dimensionality separate via METHOD3_VLA_ACTION_DOF=7.
55
+ full_dof: 6
56
+ dct_L0: 50
57
+
58
+ vla_informativeness:
59
+ enabled: true
60
+ scorer_type: dct
61
+ R: 4
62
+ agg: mean
63
+ sigma: 0.5
64
+
65
+ skill_perturbation:
66
+ enabled_forward: true
67
+ enabled_reset: false
68
+ n_candidates: 64
69
+ # This file must be supplied before PHASE=phase2 can run on hardware.
70
+ curobo_robot_cfg_path: robot_configs/curobo/ur7e.yml
71
+ curobo_num_trajopt_seeds: 4
72
+ curobo_num_ik_seeds: 16
73
+ # Dynamic scene collision updates the Curobo world per skill request. CUDA
74
+ # graph capture can retain stale collision-world state, so keep it disabled.
75
+ curobo_use_cuda_graph: false
76
+ curobo_via_offset_mag: 0.12
77
+ curobo_junction_smooth_k: 5
78
+ curobo_max_vias_per_candidate: 2
79
+ # Do not perturb Curobo joint waypoints after collision-aware planning; doing
80
+ # so can invalidate the checked trajectory geometry.
81
+ curobo_joint_perturb_k: 0
82
+ curobo_joint_perturb_mag: 0.0
83
+ dynamic_collision_enabled: true
84
+ dynamic_collision_fail_closed: true
85
+ dynamic_collision_max_cuboids: 16
86
+ dynamic_collision_include_table: true
87
+ dynamic_collision_table_size: [1.2, 1.2, 0.05]
88
+ dynamic_collision_table_y_max_base_link: -0.18
89
+ dynamic_collision_table_z_floor_enabled: true
90
+ dynamic_collision_table_z_clearance_m: 0.02
91
+ dynamic_collision_table_z_check_stride: 1
92
+ dynamic_collision_disable_joint_perturbation: true
93
+ fixed_joint_indices: []
94
+ arm_joint_count: 6
95
+ # Curobo returns UR7e arm-only joint waypoints. The VLA scorer pads to
96
+ # METHOD3_VLA_ACTION_DOF=7 when a policy action head expects gripper state.
97
+ dct_target_dof: 6
98
+ ee_arm_dof: 6
99
+ servo_calibration_file: null
100
+
101
+ reembedding:
102
+ skip_invalid: false
103
+ show_progress: true
104
+ frame_stride: 1
105
+ batch_size: 32
106
+ subgoal_filter_radius_m: 0.03
107
+ subgoal_filter_min_keep: 30
108
+ decode_workers: 4
109
+
110
+ camera_rename:
111
+ realsense_wrist: camera1
112
+ realsense_topview: camera2
113
+
114
+ transport:
115
+ address: "127.0.0.1:50061"
116
+ timeout_s: 60.0
117
+ debug_verbose: true
118
+
119
+ server_build_phase1: false
120
+
121
+ policy:
122
+ family: smolvla
123
+ device: "cuda"
124
+ autocast_dtype: bfloat16
125
+ state_weight: 1.0
126
+
127
+ selector:
128
+ action_horizon: 50
129
+ max_T_eval: null
130
+ skill_dct_parquet: "results/skill_dct/UR7e_phase1_Sort_RGBblock_10fps.canonical.parquet"
131
+
132
+ buffer:
133
+ save_dir: "grpc_server/buffer"
134
+ filename: "server_skill_wise_vector_db.npz"
135
+
136
+ remote:
137
+ ssh_alias: ""
138
+ user: "csiwoo"
139
+ hostname: "61.107.202.230"
140
+ port: 22
141
+ identity_file: ""
142
+ # Clone root of this superproject on the remote server.
143
+ workspace_path: "/home/csiwoo/CORL2026/ur7e_autodata_workspace"
144
+ # Materialized patched AutoDataCollector runtime under workspace_path.
145
+ project_path: "/home/csiwoo/CORL2026/ur7e_autodata_workspace/runtime/AutoDataCollector-ur7e"
146
+ conda_env: "gpu_server"
147
+ gpu_id: 0
148
+ tmux_session: "phase2_server_ur7e"
149
+ git_pull_before_start: true
150
+ ready_timeout_s: 360
151
+ nvidia_backport_path: ""
152
+
153
+ server_runtime:
154
+ recording_config: "pipeline_config/recording_config_ur7e_phase2.yaml"
155
+ urdf: "assets/urdf/ur7e.urdf"
config/recording_config_ur7e_phase2.yaml ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # UR7e Phase2 gRPC server recording config.
2
+ #
3
+ # This file is intentionally minimal. Phase2 Method3 server settings such as
4
+ # VLA checkpoint, Curobo config, selector, buffer, and camera key remapping live
5
+ # in pipeline_config/phase2_config.yaml. The gRPC server still requires a
6
+ # recording config path for legacy compatibility and reads only a few generic
7
+ # values from it.
8
+
9
+ dataset_repo_id: "CoRL2026-CSI/UR7e_Pnp_phase2_server"
10
+ recording_fps: 10
11
+ robot_type: "ur7e"
12
+
13
+ # Server-side Phase2 does not write reset datasets. Local robot recording uses
14
+ # run_forward_and_reset_ur7e_method.sh and its selected dataset repo.
15
+ enable_reset_dataset_recording: false
16
+
17
+ # Keep transit pitch unconstrained for UR7e Curobo candidates unless explicitly
18
+ # overridden by an experiment-specific recording config.
19
+ transit_pitch_max_deg: null
20
+
21
+ # Legacy compatibility only. The actual Method3 Phase2 source of truth is
22
+ # pipeline_config/phase2_config.yaml.
23
+ preselective_filter: {}
24
+ perturbation:
25
+ skill: {}
pretrained_model/config.json ADDED
@@ -0,0 +1,109 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "smolvla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 6
9
+ ]
10
+ },
11
+ "observation.images.camera1": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 256,
16
+ 256
17
+ ]
18
+ },
19
+ "observation.images.camera2": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 256,
24
+ 256
25
+ ]
26
+ },
27
+ "observation.images.camera3": {
28
+ "type": "VISUAL",
29
+ "shape": [
30
+ 3,
31
+ 256,
32
+ 256
33
+ ]
34
+ },
35
+ "observation.images.empty_camera_0": {
36
+ "type": "VISUAL",
37
+ "shape": [
38
+ 3,
39
+ 480,
40
+ 640
41
+ ]
42
+ }
43
+ },
44
+ "output_features": {
45
+ "action": {
46
+ "type": "ACTION",
47
+ "shape": [
48
+ 7
49
+ ]
50
+ }
51
+ },
52
+ "device": "cuda",
53
+ "use_amp": false,
54
+ "use_peft": false,
55
+ "push_to_hub": false,
56
+ "repo_id": null,
57
+ "private": null,
58
+ "tags": null,
59
+ "license": null,
60
+ "pretrained_path": "lerobot/smolvla_base",
61
+ "chunk_size": 50,
62
+ "n_action_steps": 50,
63
+ "normalization_mapping": {
64
+ "VISUAL": "IDENTITY",
65
+ "STATE": "MEAN_STD",
66
+ "ACTION": "MEAN_STD"
67
+ },
68
+ "max_state_dim": 32,
69
+ "max_action_dim": 32,
70
+ "resize_imgs_with_padding": [
71
+ 512,
72
+ 512
73
+ ],
74
+ "empty_cameras": 1,
75
+ "adapt_to_pi_aloha": false,
76
+ "use_delta_joint_actions_aloha": false,
77
+ "tokenizer_max_length": 48,
78
+ "num_steps": 10,
79
+ "use_cache": true,
80
+ "freeze_vision_encoder": true,
81
+ "train_expert_only": true,
82
+ "train_state_proj": true,
83
+ "optimizer_lr": 0.0001,
84
+ "optimizer_betas": [
85
+ 0.9,
86
+ 0.95
87
+ ],
88
+ "optimizer_eps": 1e-08,
89
+ "optimizer_weight_decay": 1e-10,
90
+ "optimizer_grad_clip_norm": 10.0,
91
+ "scheduler_warmup_steps": 1000,
92
+ "scheduler_decay_steps": 30000,
93
+ "scheduler_decay_lr": 2.5e-06,
94
+ "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
95
+ "load_vlm_weights": true,
96
+ "add_image_special_tokens": false,
97
+ "attention_mode": "cross_attn",
98
+ "prefix_length": 0,
99
+ "pad_language_to": "max_length",
100
+ "num_expert_layers": 0,
101
+ "num_vlm_layers": 16,
102
+ "self_attn_every_n_layers": 2,
103
+ "expert_width_multiplier": 0.75,
104
+ "min_period": 0.004,
105
+ "max_period": 4.0,
106
+ "rtc_config": null,
107
+ "compile_model": false,
108
+ "compile_mode": "max-autotune"
109
+ }
pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3f36316a140b608f1404d823bc51c300858332200ec2496b4079d2b565d4f3b
3
+ size 906712520
pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 7
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "IDENTITY",
18
+ "STATE": "MEAN_STD",
19
+ "ACTION": "MEAN_STD"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "device_processor",
26
+ "config": {
27
+ "device": "cpu",
28
+ "float_dtype": null
29
+ }
30
+ }
31
+ ]
32
+ }
pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bb5dbd493da25bfa7f225edcf47d85c85788fad965ebd86aa87877cb9d674e5
3
+ size 15124
pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,98 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {
8
+ "observation.images.realsense_wrist": "observation.images.camera1",
9
+ "observation.images.realsense_topview": "observation.images.camera2"
10
+ }
11
+ }
12
+ },
13
+ {
14
+ "registry_name": "to_batch_processor",
15
+ "config": {}
16
+ },
17
+ {
18
+ "registry_name": "smolvla_new_line_processor",
19
+ "config": {}
20
+ },
21
+ {
22
+ "registry_name": "tokenizer_processor",
23
+ "config": {
24
+ "max_length": 48,
25
+ "task_key": "task",
26
+ "padding_side": "right",
27
+ "padding": "max_length",
28
+ "truncation": true,
29
+ "tokenizer_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct"
30
+ }
31
+ },
32
+ {
33
+ "registry_name": "device_processor",
34
+ "config": {
35
+ "device": "cuda",
36
+ "float_dtype": null
37
+ }
38
+ },
39
+ {
40
+ "registry_name": "normalizer_processor",
41
+ "config": {
42
+ "eps": 1e-08,
43
+ "features": {
44
+ "observation.state": {
45
+ "type": "STATE",
46
+ "shape": [
47
+ 6
48
+ ]
49
+ },
50
+ "observation.images.camera1": {
51
+ "type": "VISUAL",
52
+ "shape": [
53
+ 3,
54
+ 256,
55
+ 256
56
+ ]
57
+ },
58
+ "observation.images.camera2": {
59
+ "type": "VISUAL",
60
+ "shape": [
61
+ 3,
62
+ 256,
63
+ 256
64
+ ]
65
+ },
66
+ "observation.images.camera3": {
67
+ "type": "VISUAL",
68
+ "shape": [
69
+ 3,
70
+ 256,
71
+ 256
72
+ ]
73
+ },
74
+ "observation.images.empty_camera_0": {
75
+ "type": "VISUAL",
76
+ "shape": [
77
+ 3,
78
+ 480,
79
+ 640
80
+ ]
81
+ },
82
+ "action": {
83
+ "type": "ACTION",
84
+ "shape": [
85
+ 7
86
+ ]
87
+ }
88
+ },
89
+ "norm_map": {
90
+ "VISUAL": "IDENTITY",
91
+ "STATE": "MEAN_STD",
92
+ "ACTION": "MEAN_STD"
93
+ }
94
+ },
95
+ "state_file": "policy_preprocessor_step_5_normalizer_processor.safetensors"
96
+ }
97
+ ]
98
+ }
pretrained_model/policy_preprocessor_step_5_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bb5dbd493da25bfa7f225edcf47d85c85788fad965ebd86aa87877cb9d674e5
3
+ size 15124
pretrained_model/train_config.json ADDED
@@ -0,0 +1,251 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "CoRL2026-CSI/UR7e_phase1_Sort_RGBblock_10fps",
4
+ "root": null,
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": true,
8
+ "max_num_transforms": 3,
9
+ "random_order": true,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 1.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.8,
17
+ 1.2
18
+ ]
19
+ }
20
+ },
21
+ "contrast": {
22
+ "weight": 1.0,
23
+ "type": "ColorJitter",
24
+ "kwargs": {
25
+ "contrast": [
26
+ 0.8,
27
+ 1.2
28
+ ]
29
+ }
30
+ },
31
+ "saturation": {
32
+ "weight": 1.0,
33
+ "type": "ColorJitter",
34
+ "kwargs": {
35
+ "saturation": [
36
+ 0.5,
37
+ 1.5
38
+ ]
39
+ }
40
+ },
41
+ "hue": {
42
+ "weight": 1.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.05,
47
+ 0.05
48
+ ]
49
+ }
50
+ },
51
+ "sharpness": {
52
+ "weight": 1.0,
53
+ "type": "SharpnessJitter",
54
+ "kwargs": {
55
+ "sharpness": [
56
+ 0.5,
57
+ 1.5
58
+ ]
59
+ }
60
+ },
61
+ "affine": {
62
+ "weight": 1.0,
63
+ "type": "RandomAffine",
64
+ "kwargs": {
65
+ "degrees": [
66
+ -5.0,
67
+ 5.0
68
+ ],
69
+ "translate": [
70
+ 0.05,
71
+ 0.05
72
+ ]
73
+ }
74
+ }
75
+ }
76
+ },
77
+ "revision": "v3.0",
78
+ "use_imagenet_stats": true,
79
+ "video_backend": "pyav",
80
+ "streaming": false,
81
+ "skill_dct_parquet": "/home/csiwoo/CORL2026/ur7e_autodata_workspace/runtime/AutoDataCollector-ur7e/results/skill_dct/UR7e_phase1_Sort_RGBblock_10fps.canonical.parquet"
82
+ },
83
+ "env": null,
84
+ "policy": {
85
+ "type": "smolvla",
86
+ "n_obs_steps": 1,
87
+ "input_features": {
88
+ "observation.state": {
89
+ "type": "STATE",
90
+ "shape": [
91
+ 6
92
+ ]
93
+ },
94
+ "observation.images.camera1": {
95
+ "type": "VISUAL",
96
+ "shape": [
97
+ 3,
98
+ 256,
99
+ 256
100
+ ]
101
+ },
102
+ "observation.images.camera2": {
103
+ "type": "VISUAL",
104
+ "shape": [
105
+ 3,
106
+ 256,
107
+ 256
108
+ ]
109
+ },
110
+ "observation.images.camera3": {
111
+ "type": "VISUAL",
112
+ "shape": [
113
+ 3,
114
+ 256,
115
+ 256
116
+ ]
117
+ },
118
+ "observation.images.empty_camera_0": {
119
+ "type": "VISUAL",
120
+ "shape": [
121
+ 3,
122
+ 480,
123
+ 640
124
+ ]
125
+ }
126
+ },
127
+ "output_features": {
128
+ "action": {
129
+ "type": "ACTION",
130
+ "shape": [
131
+ 7
132
+ ]
133
+ }
134
+ },
135
+ "device": "cuda",
136
+ "use_amp": false,
137
+ "use_peft": false,
138
+ "push_to_hub": false,
139
+ "repo_id": null,
140
+ "private": null,
141
+ "tags": null,
142
+ "license": null,
143
+ "pretrained_path": "lerobot/smolvla_base",
144
+ "chunk_size": 50,
145
+ "n_action_steps": 50,
146
+ "normalization_mapping": {
147
+ "VISUAL": "IDENTITY",
148
+ "STATE": "MEAN_STD",
149
+ "ACTION": "MEAN_STD"
150
+ },
151
+ "max_state_dim": 32,
152
+ "max_action_dim": 32,
153
+ "resize_imgs_with_padding": [
154
+ 512,
155
+ 512
156
+ ],
157
+ "empty_cameras": 1,
158
+ "adapt_to_pi_aloha": false,
159
+ "use_delta_joint_actions_aloha": false,
160
+ "tokenizer_max_length": 48,
161
+ "num_steps": 10,
162
+ "use_cache": true,
163
+ "freeze_vision_encoder": true,
164
+ "train_expert_only": true,
165
+ "train_state_proj": true,
166
+ "optimizer_lr": 0.0001,
167
+ "optimizer_betas": [
168
+ 0.9,
169
+ 0.95
170
+ ],
171
+ "optimizer_eps": 1e-08,
172
+ "optimizer_weight_decay": 1e-10,
173
+ "optimizer_grad_clip_norm": 10.0,
174
+ "scheduler_warmup_steps": 1000,
175
+ "scheduler_decay_steps": 30000,
176
+ "scheduler_decay_lr": 2.5e-06,
177
+ "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
178
+ "load_vlm_weights": true,
179
+ "add_image_special_tokens": false,
180
+ "attention_mode": "cross_attn",
181
+ "prefix_length": 0,
182
+ "pad_language_to": "max_length",
183
+ "num_expert_layers": 0,
184
+ "num_vlm_layers": 16,
185
+ "self_attn_every_n_layers": 2,
186
+ "expert_width_multiplier": 0.75,
187
+ "min_period": 0.004,
188
+ "max_period": 4.0,
189
+ "rtc_config": null,
190
+ "compile_model": false,
191
+ "compile_mode": "max-autotune"
192
+ },
193
+ "output_dir": "/home/csiwoo/CORL2026/ur7e_autodata_workspace/runtime/AutoDataCollector-ur7e/results/method3_embed_vla/smolvla_dct_ur7e_sort_rgbblock_canonical_3000s_20260528_050642",
194
+ "job_name": "smolvla_dct_ur7e_sort_rgbblock_canonical_3000s_20260528_050642",
195
+ "resume": false,
196
+ "seed": 1000,
197
+ "cudnn_deterministic": false,
198
+ "num_workers": 8,
199
+ "batch_size": 16,
200
+ "gradient_accumulation_steps": 1,
201
+ "steps": 3000,
202
+ "eval_freq": 0,
203
+ "log_freq": 25,
204
+ "tolerance_s": 0.0001,
205
+ "save_checkpoint": true,
206
+ "save_freq": 3000,
207
+ "use_policy_training_preset": true,
208
+ "optimizer": {
209
+ "type": "adamw",
210
+ "lr": 0.0001,
211
+ "weight_decay": 1e-10,
212
+ "grad_clip_norm": 10.0,
213
+ "betas": [
214
+ 0.9,
215
+ 0.95
216
+ ],
217
+ "eps": 1e-08
218
+ },
219
+ "scheduler": {
220
+ "type": "cosine_decay_with_warmup",
221
+ "num_warmup_steps": 1000,
222
+ "num_decay_steps": 30000,
223
+ "peak_lr": 0.0001,
224
+ "decay_lr": 2.5e-06
225
+ },
226
+ "eval": {
227
+ "n_episodes": 50,
228
+ "batch_size": 50,
229
+ "use_async_envs": false
230
+ },
231
+ "wandb": {
232
+ "enable": false,
233
+ "disable_artifact": false,
234
+ "project": "lerobot-smolvla-dct",
235
+ "entity": null,
236
+ "notes": null,
237
+ "run_id": null,
238
+ "mode": null
239
+ },
240
+ "peft": null,
241
+ "use_rabc": false,
242
+ "rabc_progress_path": null,
243
+ "rabc_kappa": 0.01,
244
+ "rabc_epsilon": 1e-06,
245
+ "rabc_head_mode": "sparse",
246
+ "rename_map": {
247
+ "observation.images.realsense_wrist": "observation.images.camera1",
248
+ "observation.images.realsense_topview": "observation.images.camera2"
249
+ },
250
+ "checkpoint_path": null
251
+ }
session/dataset_judge_rejected_skips.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "forward_raw_kept_conversion_skipped": [
3
+ {
4
+ "episode": 6,
5
+ "raw_dir": "/home/franka/hscho_workspace/ur7e_autodata_workspace/results/ur7e_method3/session_20260528_011649/episode_06/recording_raw",
6
+ "judge": "FALSE",
7
+ "reason": "forward_judge_not_true",
8
+ "timestamp": "20260528_014136"
9
+ },
10
+ {
11
+ "episode": 7,
12
+ "raw_dir": "/home/franka/hscho_workspace/ur7e_autodata_workspace/results/ur7e_method3/session_20260528_011649/episode_07/recording_raw",
13
+ "judge": "FALSE",
14
+ "reason": "forward_judge_not_true",
15
+ "timestamp": "20260528_014547"
16
+ },
17
+ {
18
+ "episode": 12,
19
+ "raw_dir": "/home/franka/hscho_workspace/ur7e_autodata_workspace/results/ur7e_method3/session_20260528_011649/episode_12/recording_raw",
20
+ "judge": "FALSE",
21
+ "reason": "forward_judge_not_true",
22
+ "timestamp": "20260528_023625"
23
+ },
24
+ {
25
+ "episode": 19,
26
+ "raw_dir": "/home/franka/hscho_workspace/ur7e_autodata_workspace/results/ur7e_method3/session_20260528_011649/episode_19/recording_raw",
27
+ "judge": "FALSE",
28
+ "reason": "forward_judge_not_true",
29
+ "timestamp": "20260528_032638"
30
+ }
31
+ ]
32
+ }
session/resume_append_status.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "session_dir": "/home/franka/hscho_workspace/ur7e_autodata_workspace/results/ur7e_method3/session_20260528_011649",
3
+ "new_episode_numbers": [
4
+ 20,
5
+ 21,
6
+ 22,
7
+ 23,
8
+ 24,
9
+ 25,
10
+ 26,
11
+ 27
12
+ ],
13
+ "phase": "phase1",
14
+ "excluded_episodes": [],
15
+ "target_success_episodes": 20,
16
+ "clean_forward_judge_true": 20,
17
+ "missing_recording_forward_true_episodes": [],
18
+ "effective_num_episodes": 27,
19
+ "raw_num_episodes": 27,
20
+ "phase2_stop_reason": null,
21
+ "phase2_saturation_last_mean": null
22
+ }
session/session_summary.json ADDED
@@ -0,0 +1,1410 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "num_episodes": 150,
3
+ "effective_num_episodes": 27,
4
+ "target_success_episodes": 20,
5
+ "excluded_episodes": [],
6
+ "instruction": "Sort each colored block onto the dish of the matching color.",
7
+ "objects": [
8
+ "red block",
9
+ "green block",
10
+ "blue block",
11
+ "red dish",
12
+ "green dish",
13
+ "blue dish"
14
+ ],
15
+ "phase": "phase1",
16
+ "timestamp": "20260528_043210",
17
+ "summary": {
18
+ "forward_success": 8,
19
+ "forward_judge_true": 20,
20
+ "forward_judge_false": 4,
21
+ "reset_success": 8,
22
+ "reset_judge_true": 17,
23
+ "reset_judge_false": 5
24
+ },
25
+ "raw_summary": {
26
+ "forward_success": 8,
27
+ "forward_judge_true": 20,
28
+ "forward_judge_false": 4,
29
+ "reset_success": 8,
30
+ "reset_judge_true": 17,
31
+ "reset_judge_false": 5
32
+ },
33
+ "clean_recorded_forward_judge_true": 20,
34
+ "missing_recording_forward_true_episodes": [],
35
+ "episodes": [
36
+ {
37
+ "episode": 1,
38
+ "success": true,
39
+ "error": null,
40
+ "forward_success": false,
41
+ "forward_judge": "TRUE",
42
+ "reset_success": false,
43
+ "reset_judge": "TRUE",
44
+ "forward_judge_input": {
45
+ "judge_mode": "standard_forward_vlm",
46
+ "model": "gemini-3.1-flash-lite",
47
+ "use_server": false,
48
+ "temperature": 0.0,
49
+ "max_tokens": 10000,
50
+ "image_resolution": [
51
+ 640,
52
+ 480
53
+ ],
54
+ "image_inputs": [
55
+ {
56
+ "role": "initial",
57
+ "label": "Image 1: initial top-view before task execution",
58
+ "path": "episode_01/forward/initial_state.jpg"
59
+ },
60
+ {
61
+ "role": "final",
62
+ "label": "Image 2: final top-view after task execution",
63
+ "path": "episode_01/forward/final_state.jpg"
64
+ }
65
+ ],
66
+ "stack_status": "not_stack",
67
+ "metadata_status": "complete",
68
+ "metadata_errors": [],
69
+ "prompt": {
70
+ "path": "episode_01/forward/judge_prompt.txt",
71
+ "sha256": "589a86cdbef4a9d9837804c392ed9ce2ab25667fbaff06d4d4c5394a0f12b314",
72
+ "chars": 3076
73
+ },
74
+ "raw_response": {
75
+ "path": "episode_01/forward/judge_raw_response.txt",
76
+ "sha256": "64b50a57728be7ae89c9011e3005e8fbe6aa4740303b53c1d901d4a506885850",
77
+ "chars": 2170
78
+ },
79
+ "usage_status": "available",
80
+ "usage": {
81
+ "model": "gemini-3.1-flash-lite",
82
+ "inference_time_s": 3.48,
83
+ "input_tokens": 2899,
84
+ "output_tokens": 589,
85
+ "total_tokens": 3488
86
+ }
87
+ },
88
+ "excluded": false,
89
+ "forward_raw_exists": true,
90
+ "reset_raw_exists": false,
91
+ "forward_recording_required": true,
92
+ "forward_recording_valid": true,
93
+ "target_forward_success": true
94
+ },
95
+ {
96
+ "episode": 2,
97
+ "success": true,
98
+ "error": null,
99
+ "forward_success": false,
100
+ "forward_judge": "TRUE",
101
+ "reset_success": false,
102
+ "reset_judge": "TRUE",
103
+ "forward_judge_input": {
104
+ "judge_mode": "standard_forward_vlm",
105
+ "model": "gemini-3.1-flash-lite",
106
+ "use_server": false,
107
+ "temperature": 0.0,
108
+ "max_tokens": 10000,
109
+ "image_resolution": [
110
+ 640,
111
+ 480
112
+ ],
113
+ "image_inputs": [
114
+ {
115
+ "role": "initial",
116
+ "label": "Image 1: initial top-view before task execution",
117
+ "path": "episode_02/forward/initial_state.jpg"
118
+ },
119
+ {
120
+ "role": "final",
121
+ "label": "Image 2: final top-view after task execution",
122
+ "path": "episode_02/forward/final_state.jpg"
123
+ }
124
+ ],
125
+ "stack_status": "not_stack",
126
+ "metadata_status": "complete",
127
+ "metadata_errors": [],
128
+ "prompt": {
129
+ "path": "episode_02/forward/judge_prompt.txt",
130
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
131
+ "chars": 3076
132
+ },
133
+ "raw_response": {
134
+ "path": "episode_02/forward/judge_raw_response.txt",
135
+ "sha256": "2a8e9a7a2c9931dfd0e9272e4069e18369708111df9e4a8c84cd6cb3b04c810f",
136
+ "chars": 2132
137
+ },
138
+ "usage_status": "available",
139
+ "usage": {
140
+ "model": "gemini-3.1-flash-lite",
141
+ "inference_time_s": 3.42,
142
+ "input_tokens": 2899,
143
+ "output_tokens": 573,
144
+ "total_tokens": 3472
145
+ }
146
+ },
147
+ "excluded": false,
148
+ "forward_raw_exists": true,
149
+ "reset_raw_exists": false,
150
+ "forward_recording_required": true,
151
+ "forward_recording_valid": true,
152
+ "target_forward_success": true
153
+ },
154
+ {
155
+ "episode": 3,
156
+ "success": true,
157
+ "error": null,
158
+ "forward_success": false,
159
+ "forward_judge": "TRUE",
160
+ "reset_success": false,
161
+ "reset_judge": "FALSE",
162
+ "forward_judge_input": {
163
+ "judge_mode": "standard_forward_vlm",
164
+ "model": "gemini-3.1-flash-lite",
165
+ "use_server": false,
166
+ "temperature": 0.0,
167
+ "max_tokens": 10000,
168
+ "image_resolution": [
169
+ 640,
170
+ 480
171
+ ],
172
+ "image_inputs": [
173
+ {
174
+ "role": "initial",
175
+ "label": "Image 1: initial top-view before task execution",
176
+ "path": "episode_03/forward/initial_state.jpg"
177
+ },
178
+ {
179
+ "role": "final",
180
+ "label": "Image 2: final top-view after task execution",
181
+ "path": "episode_03/forward/final_state.jpg"
182
+ }
183
+ ],
184
+ "stack_status": "not_stack",
185
+ "metadata_status": "complete",
186
+ "metadata_errors": [],
187
+ "prompt": {
188
+ "path": "episode_03/forward/judge_prompt.txt",
189
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
190
+ "chars": 3076
191
+ },
192
+ "raw_response": {
193
+ "path": "episode_03/forward/judge_raw_response.txt",
194
+ "sha256": "6de1199d54d47a06c13c0c91d9eab58c4190b684dd4c8aa804ec00f67c0315d8",
195
+ "chars": 2269
196
+ },
197
+ "usage_status": "available",
198
+ "usage": {
199
+ "model": "gemini-3.1-flash-lite",
200
+ "inference_time_s": 3.29,
201
+ "input_tokens": 2899,
202
+ "output_tokens": 595,
203
+ "total_tokens": 3494
204
+ }
205
+ },
206
+ "excluded": false,
207
+ "forward_raw_exists": true,
208
+ "reset_raw_exists": false,
209
+ "forward_recording_required": true,
210
+ "forward_recording_valid": true,
211
+ "target_forward_success": true
212
+ },
213
+ {
214
+ "episode": 4,
215
+ "success": true,
216
+ "error": null,
217
+ "forward_success": false,
218
+ "forward_judge": "TRUE",
219
+ "reset_success": false,
220
+ "reset_judge": "TRUE",
221
+ "forward_judge_input": {
222
+ "judge_mode": "standard_forward_vlm",
223
+ "model": "gemini-3.1-flash-lite",
224
+ "use_server": false,
225
+ "temperature": 0.0,
226
+ "max_tokens": 10000,
227
+ "image_resolution": [
228
+ 640,
229
+ 480
230
+ ],
231
+ "image_inputs": [
232
+ {
233
+ "role": "initial",
234
+ "label": "Image 1: initial top-view before task execution",
235
+ "path": "episode_04/forward/initial_state.jpg"
236
+ },
237
+ {
238
+ "role": "final",
239
+ "label": "Image 2: final top-view after task execution",
240
+ "path": "episode_04/forward/final_state.jpg"
241
+ }
242
+ ],
243
+ "stack_status": "not_stack",
244
+ "metadata_status": "complete",
245
+ "metadata_errors": [],
246
+ "prompt": {
247
+ "path": "episode_04/forward/judge_prompt.txt",
248
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
249
+ "chars": 3076
250
+ },
251
+ "raw_response": {
252
+ "path": "episode_04/forward/judge_raw_response.txt",
253
+ "sha256": "8a183af81d93af91c5ae3db4394e162e8d56aed5fd85f40e087ce32f628053a3",
254
+ "chars": 2181
255
+ },
256
+ "usage_status": "available",
257
+ "usage": {
258
+ "model": "gemini-3.1-flash-lite",
259
+ "inference_time_s": 2.97,
260
+ "input_tokens": 2899,
261
+ "output_tokens": 580,
262
+ "total_tokens": 3479
263
+ }
264
+ },
265
+ "excluded": false,
266
+ "forward_raw_exists": true,
267
+ "reset_raw_exists": false,
268
+ "forward_recording_required": true,
269
+ "forward_recording_valid": true,
270
+ "target_forward_success": true
271
+ },
272
+ {
273
+ "episode": 5,
274
+ "success": true,
275
+ "error": null,
276
+ "forward_success": false,
277
+ "forward_judge": "TRUE",
278
+ "reset_success": false,
279
+ "reset_judge": "TRUE",
280
+ "forward_judge_input": {
281
+ "judge_mode": "standard_forward_vlm",
282
+ "model": "gemini-3.1-flash-lite",
283
+ "use_server": false,
284
+ "temperature": 0.0,
285
+ "max_tokens": 10000,
286
+ "image_resolution": [
287
+ 640,
288
+ 480
289
+ ],
290
+ "image_inputs": [
291
+ {
292
+ "role": "initial",
293
+ "label": "Image 1: initial top-view before task execution",
294
+ "path": "episode_05/forward/initial_state.jpg"
295
+ },
296
+ {
297
+ "role": "final",
298
+ "label": "Image 2: final top-view after task execution",
299
+ "path": "episode_05/forward/final_state.jpg"
300
+ }
301
+ ],
302
+ "stack_status": "not_stack",
303
+ "metadata_status": "complete",
304
+ "metadata_errors": [],
305
+ "prompt": {
306
+ "path": "episode_05/forward/judge_prompt.txt",
307
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
308
+ "chars": 3076
309
+ },
310
+ "raw_response": {
311
+ "path": "episode_05/forward/judge_raw_response.txt",
312
+ "sha256": "21b761139ab2185d5899f215a5e41b7e74a7f33d1a64b6815a5ecc7268b56bae",
313
+ "chars": 1941
314
+ },
315
+ "usage_status": "available",
316
+ "usage": {
317
+ "model": "gemini-3.1-flash-lite",
318
+ "inference_time_s": 3.25,
319
+ "input_tokens": 2899,
320
+ "output_tokens": 515,
321
+ "total_tokens": 3414
322
+ }
323
+ },
324
+ "excluded": false,
325
+ "forward_raw_exists": true,
326
+ "reset_raw_exists": false,
327
+ "forward_recording_required": true,
328
+ "forward_recording_valid": true,
329
+ "target_forward_success": true
330
+ },
331
+ {
332
+ "episode": 6,
333
+ "success": true,
334
+ "error": null,
335
+ "forward_success": false,
336
+ "forward_judge": "FALSE",
337
+ "reset_success": false,
338
+ "reset_judge": "TRUE",
339
+ "forward_judge_input": {
340
+ "judge_mode": "standard_forward_vlm",
341
+ "model": "gemini-3.1-flash-lite",
342
+ "use_server": false,
343
+ "temperature": 0.0,
344
+ "max_tokens": 10000,
345
+ "image_resolution": [
346
+ 640,
347
+ 480
348
+ ],
349
+ "image_inputs": [
350
+ {
351
+ "role": "initial",
352
+ "label": "Image 1: initial top-view before task execution",
353
+ "path": "episode_06/forward/initial_state.jpg"
354
+ },
355
+ {
356
+ "role": "final",
357
+ "label": "Image 2: final top-view after task execution",
358
+ "path": "episode_06/forward/final_state.jpg"
359
+ }
360
+ ],
361
+ "stack_status": "not_stack",
362
+ "metadata_status": "complete",
363
+ "metadata_errors": [],
364
+ "prompt": {
365
+ "path": "episode_06/forward/judge_prompt.txt",
366
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
367
+ "chars": 3076
368
+ },
369
+ "raw_response": {
370
+ "path": "episode_06/forward/judge_raw_response.txt",
371
+ "sha256": "bb694ad795cb4f1be1adf3d3d560f0a191fe5e502e8c1895bb5b1b476a5f647b",
372
+ "chars": 2304
373
+ },
374
+ "usage_status": "available",
375
+ "usage": {
376
+ "model": "gemini-3.1-flash-lite",
377
+ "inference_time_s": 4.73,
378
+ "input_tokens": 2899,
379
+ "output_tokens": 617,
380
+ "total_tokens": 3516
381
+ }
382
+ },
383
+ "excluded": false,
384
+ "forward_raw_exists": true,
385
+ "reset_raw_exists": false,
386
+ "forward_recording_required": true,
387
+ "forward_recording_valid": true,
388
+ "target_forward_success": false
389
+ },
390
+ {
391
+ "episode": 7,
392
+ "success": true,
393
+ "error": null,
394
+ "forward_success": false,
395
+ "forward_judge": "FALSE",
396
+ "reset_success": false,
397
+ "reset_judge": "TRUE",
398
+ "forward_judge_input": {
399
+ "judge_mode": "standard_forward_vlm",
400
+ "model": "gemini-3.1-flash-lite",
401
+ "use_server": false,
402
+ "temperature": 0.0,
403
+ "max_tokens": 10000,
404
+ "image_resolution": [
405
+ 640,
406
+ 480
407
+ ],
408
+ "image_inputs": [
409
+ {
410
+ "role": "initial",
411
+ "label": "Image 1: initial top-view before task execution",
412
+ "path": "episode_07/forward/initial_state.jpg"
413
+ },
414
+ {
415
+ "role": "final",
416
+ "label": "Image 2: final top-view after task execution",
417
+ "path": "episode_07/forward/final_state.jpg"
418
+ }
419
+ ],
420
+ "stack_status": "not_stack",
421
+ "metadata_status": "complete",
422
+ "metadata_errors": [],
423
+ "prompt": {
424
+ "path": "episode_07/forward/judge_prompt.txt",
425
+ "sha256": "51e1a2a0c8f5446b06717507e0ab31dcbaf5ac6eca0d54fac7223b00ed6b6fd6",
426
+ "chars": 3076
427
+ },
428
+ "raw_response": {
429
+ "path": "episode_07/forward/judge_raw_response.txt",
430
+ "sha256": "60f0aead44dc35f4b199714227e61afe1c8be97f3e19bd485be4ff9976c6dd2f",
431
+ "chars": 2172
432
+ },
433
+ "usage_status": "available",
434
+ "usage": {
435
+ "model": "gemini-3.1-flash-lite",
436
+ "inference_time_s": 4.49,
437
+ "input_tokens": 2905,
438
+ "output_tokens": 610,
439
+ "total_tokens": 3515
440
+ }
441
+ },
442
+ "excluded": false,
443
+ "forward_raw_exists": true,
444
+ "reset_raw_exists": false,
445
+ "forward_recording_required": true,
446
+ "forward_recording_valid": true,
447
+ "target_forward_success": false
448
+ },
449
+ {
450
+ "episode": 8,
451
+ "success": false,
452
+ "error": null,
453
+ "forward_success": false,
454
+ "forward_judge": "N/A",
455
+ "reset_success": false,
456
+ "reset_judge": "N/A",
457
+ "excluded": false,
458
+ "forward_raw_exists": false,
459
+ "reset_raw_exists": false,
460
+ "forward_recording_required": true,
461
+ "forward_recording_valid": false,
462
+ "target_forward_success": false
463
+ },
464
+ {
465
+ "episode": 9,
466
+ "success": true,
467
+ "error": null,
468
+ "forward_success": false,
469
+ "forward_judge": "TRUE",
470
+ "reset_success": false,
471
+ "reset_judge": "FALSE",
472
+ "forward_judge_input": {
473
+ "judge_mode": "standard_forward_vlm",
474
+ "model": "gemini-3.1-flash-lite",
475
+ "use_server": false,
476
+ "temperature": 0.0,
477
+ "max_tokens": 10000,
478
+ "image_resolution": [
479
+ 640,
480
+ 480
481
+ ],
482
+ "image_inputs": [
483
+ {
484
+ "role": "initial",
485
+ "label": "Image 1: initial top-view before task execution",
486
+ "path": "episode_09/forward/initial_state.jpg"
487
+ },
488
+ {
489
+ "role": "final",
490
+ "label": "Image 2: final top-view after task execution",
491
+ "path": "episode_09/forward/final_state.jpg"
492
+ }
493
+ ],
494
+ "stack_status": "not_stack",
495
+ "metadata_status": "complete",
496
+ "metadata_errors": [],
497
+ "prompt": {
498
+ "path": "episode_09/forward/judge_prompt.txt",
499
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
500
+ "chars": 3076
501
+ },
502
+ "raw_response": {
503
+ "path": "episode_09/forward/judge_raw_response.txt",
504
+ "sha256": "91da804a7b1f94e09ecdd1ac5602814ad7f1d7823568fc65d1da4e096c432889",
505
+ "chars": 1987
506
+ },
507
+ "usage_status": "available",
508
+ "usage": {
509
+ "model": "gemini-3.1-flash-lite",
510
+ "inference_time_s": 2.67,
511
+ "input_tokens": 2899,
512
+ "output_tokens": 527,
513
+ "total_tokens": 3426
514
+ }
515
+ },
516
+ "excluded": false,
517
+ "forward_raw_exists": true,
518
+ "reset_raw_exists": false,
519
+ "forward_recording_required": true,
520
+ "forward_recording_valid": true,
521
+ "target_forward_success": true
522
+ },
523
+ {
524
+ "episode": 10,
525
+ "success": true,
526
+ "error": null,
527
+ "forward_success": false,
528
+ "forward_judge": "TRUE",
529
+ "reset_success": false,
530
+ "reset_judge": "FALSE",
531
+ "forward_judge_input": {
532
+ "judge_mode": "standard_forward_vlm",
533
+ "model": "gemini-3.1-flash-lite",
534
+ "use_server": false,
535
+ "temperature": 0.0,
536
+ "max_tokens": 10000,
537
+ "image_resolution": [
538
+ 640,
539
+ 480
540
+ ],
541
+ "image_inputs": [
542
+ {
543
+ "role": "initial",
544
+ "label": "Image 1: initial top-view before task execution",
545
+ "path": "episode_10/forward/initial_state.jpg"
546
+ },
547
+ {
548
+ "role": "final",
549
+ "label": "Image 2: final top-view after task execution",
550
+ "path": "episode_10/forward/final_state.jpg"
551
+ }
552
+ ],
553
+ "stack_status": "not_stack",
554
+ "metadata_status": "complete",
555
+ "metadata_errors": [],
556
+ "prompt": {
557
+ "path": "episode_10/forward/judge_prompt.txt",
558
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
559
+ "chars": 3076
560
+ },
561
+ "raw_response": {
562
+ "path": "episode_10/forward/judge_raw_response.txt",
563
+ "sha256": "42744ab6a07d69aa4e9086ef3f912a5fcd232eaca5cf454ee26b6d8550a0b1c3",
564
+ "chars": 2067
565
+ },
566
+ "usage_status": "available",
567
+ "usage": {
568
+ "model": "gemini-3.1-flash-lite",
569
+ "inference_time_s": 4.01,
570
+ "input_tokens": 2899,
571
+ "output_tokens": 558,
572
+ "total_tokens": 3457
573
+ }
574
+ },
575
+ "excluded": false,
576
+ "forward_raw_exists": true,
577
+ "reset_raw_exists": false,
578
+ "forward_recording_required": true,
579
+ "forward_recording_valid": true,
580
+ "target_forward_success": true
581
+ },
582
+ {
583
+ "episode": 11,
584
+ "success": true,
585
+ "error": null,
586
+ "forward_success": false,
587
+ "forward_judge": "TRUE",
588
+ "reset_success": false,
589
+ "reset_judge": "TRUE",
590
+ "forward_judge_input": {
591
+ "judge_mode": "standard_forward_vlm",
592
+ "model": "gemini-3.1-flash-lite",
593
+ "use_server": false,
594
+ "temperature": 0.0,
595
+ "max_tokens": 10000,
596
+ "image_resolution": [
597
+ 640,
598
+ 480
599
+ ],
600
+ "image_inputs": [
601
+ {
602
+ "role": "initial",
603
+ "label": "Image 1: initial top-view before task execution",
604
+ "path": "episode_11/forward/initial_state.jpg"
605
+ },
606
+ {
607
+ "role": "final",
608
+ "label": "Image 2: final top-view after task execution",
609
+ "path": "episode_11/forward/final_state.jpg"
610
+ }
611
+ ],
612
+ "stack_status": "not_stack",
613
+ "metadata_status": "complete",
614
+ "metadata_errors": [],
615
+ "prompt": {
616
+ "path": "episode_11/forward/judge_prompt.txt",
617
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
618
+ "chars": 3076
619
+ },
620
+ "raw_response": {
621
+ "path": "episode_11/forward/judge_raw_response.txt",
622
+ "sha256": "3605b1232dab27232cfa922251eefed6ede3ec621077b9b2d79e027070af26fe",
623
+ "chars": 1986
624
+ },
625
+ "usage_status": "available",
626
+ "usage": {
627
+ "model": "gemini-3.1-flash-lite",
628
+ "inference_time_s": 2.66,
629
+ "input_tokens": 2899,
630
+ "output_tokens": 531,
631
+ "total_tokens": 3430
632
+ }
633
+ },
634
+ "excluded": false,
635
+ "forward_raw_exists": true,
636
+ "reset_raw_exists": false,
637
+ "forward_recording_required": true,
638
+ "forward_recording_valid": true,
639
+ "target_forward_success": true
640
+ },
641
+ {
642
+ "episode": 12,
643
+ "success": true,
644
+ "error": null,
645
+ "forward_success": false,
646
+ "forward_judge": "FALSE",
647
+ "reset_success": false,
648
+ "reset_judge": "N/A",
649
+ "forward_judge_input": {
650
+ "judge_mode": "standard_forward_vlm",
651
+ "model": "gemini-3.1-flash-lite",
652
+ "use_server": false,
653
+ "temperature": 0.0,
654
+ "max_tokens": 10000,
655
+ "image_resolution": [
656
+ 640,
657
+ 480
658
+ ],
659
+ "image_inputs": [
660
+ {
661
+ "role": "initial",
662
+ "label": "Image 1: initial top-view before task execution",
663
+ "path": "episode_12/forward/initial_state.jpg"
664
+ },
665
+ {
666
+ "role": "final",
667
+ "label": "Image 2: final top-view after task execution",
668
+ "path": "episode_12/forward/final_state.jpg"
669
+ }
670
+ ],
671
+ "stack_status": "not_stack",
672
+ "metadata_status": "complete",
673
+ "metadata_errors": [],
674
+ "prompt": {
675
+ "path": "episode_12/forward/judge_prompt.txt",
676
+ "sha256": "ba267bed50d76fe0f2a25d8cd0b6e77e09dcf6d3d92c28a839385f749a486bdf",
677
+ "chars": 3076
678
+ },
679
+ "raw_response": {
680
+ "path": "episode_12/forward/judge_raw_response.txt",
681
+ "sha256": "063884d67865b23976fa1a4f474cb4d575b7247d75a18fb4bdcd7ebce54d6bf6",
682
+ "chars": 2191
683
+ },
684
+ "usage_status": "available",
685
+ "usage": {
686
+ "model": "gemini-3.1-flash-lite",
687
+ "inference_time_s": 4.74,
688
+ "input_tokens": 2899,
689
+ "output_tokens": 595,
690
+ "total_tokens": 3494
691
+ }
692
+ },
693
+ "excluded": false,
694
+ "forward_raw_exists": true,
695
+ "reset_raw_exists": false,
696
+ "forward_recording_required": true,
697
+ "forward_recording_valid": true,
698
+ "target_forward_success": false
699
+ },
700
+ {
701
+ "episode": 13,
702
+ "success": true,
703
+ "error": null,
704
+ "forward_success": false,
705
+ "forward_judge": "TRUE",
706
+ "reset_success": false,
707
+ "reset_judge": "TRUE",
708
+ "forward_judge_input": {
709
+ "judge_mode": "standard_forward_vlm",
710
+ "model": "gemini-3.1-flash-lite",
711
+ "use_server": false,
712
+ "temperature": 0.0,
713
+ "max_tokens": 10000,
714
+ "image_resolution": [
715
+ 640,
716
+ 480
717
+ ],
718
+ "image_inputs": [
719
+ {
720
+ "role": "initial",
721
+ "label": "Image 1: initial top-view before task execution",
722
+ "path": "episode_13/forward/initial_state.jpg"
723
+ },
724
+ {
725
+ "role": "final",
726
+ "label": "Image 2: final top-view after task execution",
727
+ "path": "episode_13/forward/final_state.jpg"
728
+ }
729
+ ],
730
+ "stack_status": "not_stack",
731
+ "metadata_status": "complete",
732
+ "metadata_errors": [],
733
+ "prompt": {
734
+ "path": "episode_13/forward/judge_prompt.txt",
735
+ "sha256": "589a86cdbef4a9d9837804c392ed9ce2ab25667fbaff06d4d4c5394a0f12b314",
736
+ "chars": 3076
737
+ },
738
+ "raw_response": {
739
+ "path": "episode_13/forward/judge_raw_response.txt",
740
+ "sha256": "f7f72610963f1d8c9adba6c6cdf63da291edd23dbd9770c2a82db1f772dfe999",
741
+ "chars": 1948
742
+ },
743
+ "usage_status": "available",
744
+ "usage": {
745
+ "model": "gemini-3.1-flash-lite",
746
+ "inference_time_s": 3.08,
747
+ "input_tokens": 2899,
748
+ "output_tokens": 523,
749
+ "total_tokens": 3422
750
+ }
751
+ },
752
+ "excluded": false,
753
+ "forward_raw_exists": true,
754
+ "reset_raw_exists": false,
755
+ "forward_recording_required": true,
756
+ "forward_recording_valid": true,
757
+ "target_forward_success": true
758
+ },
759
+ {
760
+ "episode": 14,
761
+ "success": true,
762
+ "error": null,
763
+ "forward_success": false,
764
+ "forward_judge": "TRUE",
765
+ "reset_success": false,
766
+ "reset_judge": "FALSE",
767
+ "forward_judge_input": {
768
+ "judge_mode": "standard_forward_vlm",
769
+ "model": "gemini-3.1-flash-lite",
770
+ "use_server": false,
771
+ "temperature": 0.0,
772
+ "max_tokens": 10000,
773
+ "image_resolution": [
774
+ 640,
775
+ 480
776
+ ],
777
+ "image_inputs": [
778
+ {
779
+ "role": "initial",
780
+ "label": "Image 1: initial top-view before task execution",
781
+ "path": "episode_14/forward/initial_state.jpg"
782
+ },
783
+ {
784
+ "role": "final",
785
+ "label": "Image 2: final top-view after task execution",
786
+ "path": "episode_14/forward/final_state.jpg"
787
+ }
788
+ ],
789
+ "stack_status": "not_stack",
790
+ "metadata_status": "complete",
791
+ "metadata_errors": [],
792
+ "prompt": {
793
+ "path": "episode_14/forward/judge_prompt.txt",
794
+ "sha256": "e8b0757569fcae3f3f09ac3ec70a33d356d073cc833a152d7e622f52f3729a55",
795
+ "chars": 3076
796
+ },
797
+ "raw_response": {
798
+ "path": "episode_14/forward/judge_raw_response.txt",
799
+ "sha256": "bc9b2664555476770dd8644fa33dd3e12725ecbf0cb53b04710eed3ab06e8a99",
800
+ "chars": 2023
801
+ },
802
+ "usage_status": "available",
803
+ "usage": {
804
+ "model": "gemini-3.1-flash-lite",
805
+ "inference_time_s": 3.97,
806
+ "input_tokens": 2899,
807
+ "output_tokens": 559,
808
+ "total_tokens": 3458
809
+ }
810
+ },
811
+ "excluded": false,
812
+ "forward_raw_exists": true,
813
+ "reset_raw_exists": false,
814
+ "forward_recording_required": true,
815
+ "forward_recording_valid": true,
816
+ "target_forward_success": true
817
+ },
818
+ {
819
+ "episode": 15,
820
+ "success": false,
821
+ "error": null,
822
+ "forward_success": false,
823
+ "forward_judge": "N/A",
824
+ "reset_success": false,
825
+ "reset_judge": "N/A",
826
+ "excluded": false,
827
+ "forward_raw_exists": true,
828
+ "reset_raw_exists": false,
829
+ "forward_recording_required": true,
830
+ "forward_recording_valid": true,
831
+ "target_forward_success": false
832
+ },
833
+ {
834
+ "episode": 16,
835
+ "success": false,
836
+ "error": null,
837
+ "forward_success": false,
838
+ "forward_judge": "PENDING",
839
+ "reset_success": false,
840
+ "reset_judge": "N/A",
841
+ "excluded": false,
842
+ "forward_raw_exists": true,
843
+ "reset_raw_exists": false,
844
+ "forward_recording_required": true,
845
+ "forward_recording_valid": true,
846
+ "target_forward_success": false
847
+ },
848
+ {
849
+ "episode": 17,
850
+ "success": true,
851
+ "error": null,
852
+ "forward_success": false,
853
+ "forward_judge": "TRUE",
854
+ "reset_success": false,
855
+ "reset_judge": "TRUE",
856
+ "forward_judge_input": {
857
+ "judge_mode": "standard_forward_vlm",
858
+ "model": "gemini-3.1-flash-lite",
859
+ "use_server": false,
860
+ "temperature": 0.0,
861
+ "max_tokens": 10000,
862
+ "image_resolution": [
863
+ 640,
864
+ 480
865
+ ],
866
+ "image_inputs": [
867
+ {
868
+ "role": "initial",
869
+ "label": "Image 1: initial top-view before task execution",
870
+ "path": "episode_17/forward/initial_state.jpg"
871
+ },
872
+ {
873
+ "role": "final",
874
+ "label": "Image 2: final top-view after task execution",
875
+ "path": "episode_17/forward/final_state.jpg"
876
+ }
877
+ ],
878
+ "stack_status": "not_stack",
879
+ "metadata_status": "complete",
880
+ "metadata_errors": [],
881
+ "prompt": {
882
+ "path": "episode_17/forward/judge_prompt.txt",
883
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
884
+ "chars": 3076
885
+ },
886
+ "raw_response": {
887
+ "path": "episode_17/forward/judge_raw_response.txt",
888
+ "sha256": "c1d6bdd7d50504889968416e59b2fd5518bf167619762ce6c5404eb47c6311cb",
889
+ "chars": 1930
890
+ },
891
+ "usage_status": "available",
892
+ "usage": {
893
+ "model": "gemini-3.1-flash-lite",
894
+ "inference_time_s": 3.39,
895
+ "input_tokens": 2899,
896
+ "output_tokens": 512,
897
+ "total_tokens": 3411
898
+ }
899
+ },
900
+ "excluded": false,
901
+ "forward_raw_exists": true,
902
+ "reset_raw_exists": false,
903
+ "forward_recording_required": true,
904
+ "forward_recording_valid": true,
905
+ "target_forward_success": true
906
+ },
907
+ {
908
+ "episode": 18,
909
+ "success": true,
910
+ "error": null,
911
+ "forward_success": false,
912
+ "forward_judge": "TRUE",
913
+ "reset_success": false,
914
+ "reset_judge": "TRUE",
915
+ "forward_judge_input": {
916
+ "judge_mode": "standard_forward_vlm",
917
+ "model": "gemini-3.1-flash-lite",
918
+ "use_server": false,
919
+ "temperature": 0.0,
920
+ "max_tokens": 10000,
921
+ "image_resolution": [
922
+ 640,
923
+ 480
924
+ ],
925
+ "image_inputs": [
926
+ {
927
+ "role": "initial",
928
+ "label": "Image 1: initial top-view before task execution",
929
+ "path": "episode_18/forward/initial_state.jpg"
930
+ },
931
+ {
932
+ "role": "final",
933
+ "label": "Image 2: final top-view after task execution",
934
+ "path": "episode_18/forward/final_state.jpg"
935
+ }
936
+ ],
937
+ "stack_status": "not_stack",
938
+ "metadata_status": "complete",
939
+ "metadata_errors": [],
940
+ "prompt": {
941
+ "path": "episode_18/forward/judge_prompt.txt",
942
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
943
+ "chars": 3076
944
+ },
945
+ "raw_response": {
946
+ "path": "episode_18/forward/judge_raw_response.txt",
947
+ "sha256": "183e311badfb169ff75dfc2b2461997b13038cd10724187826dd67bdb687201e",
948
+ "chars": 2062
949
+ },
950
+ "usage_status": "available",
951
+ "usage": {
952
+ "model": "gemini-3.1-flash-lite",
953
+ "inference_time_s": 3.77,
954
+ "input_tokens": 2899,
955
+ "output_tokens": 560,
956
+ "total_tokens": 3459
957
+ }
958
+ },
959
+ "excluded": false,
960
+ "forward_raw_exists": true,
961
+ "reset_raw_exists": false,
962
+ "forward_recording_required": true,
963
+ "forward_recording_valid": true,
964
+ "target_forward_success": true
965
+ },
966
+ {
967
+ "episode": 19,
968
+ "success": true,
969
+ "error": null,
970
+ "forward_success": false,
971
+ "forward_judge": "FALSE",
972
+ "reset_success": false,
973
+ "reset_judge": "N/A",
974
+ "forward_judge_input": {
975
+ "judge_mode": "standard_forward_vlm",
976
+ "model": "gemini-3.1-flash-lite",
977
+ "use_server": false,
978
+ "temperature": 0.0,
979
+ "max_tokens": 10000,
980
+ "image_resolution": [
981
+ 640,
982
+ 480
983
+ ],
984
+ "image_inputs": [
985
+ {
986
+ "role": "initial",
987
+ "label": "Image 1: initial top-view before task execution",
988
+ "path": "episode_19/forward/initial_state.jpg"
989
+ },
990
+ {
991
+ "role": "final",
992
+ "label": "Image 2: final top-view after task execution",
993
+ "path": "episode_19/forward/final_state.jpg"
994
+ }
995
+ ],
996
+ "stack_status": "not_stack",
997
+ "metadata_status": "complete",
998
+ "metadata_errors": [],
999
+ "prompt": {
1000
+ "path": "episode_19/forward/judge_prompt.txt",
1001
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
1002
+ "chars": 3076
1003
+ },
1004
+ "raw_response": {
1005
+ "path": "episode_19/forward/judge_raw_response.txt",
1006
+ "sha256": "9ccf51831edd490432ac11905f5179dcc67ad0df636fdafbca88d62290baadee",
1007
+ "chars": 2111
1008
+ },
1009
+ "usage_status": "available",
1010
+ "usage": {
1011
+ "model": "gemini-3.1-flash-lite",
1012
+ "inference_time_s": 3.2,
1013
+ "input_tokens": 2899,
1014
+ "output_tokens": 548,
1015
+ "total_tokens": 3447
1016
+ }
1017
+ },
1018
+ "excluded": false,
1019
+ "forward_raw_exists": true,
1020
+ "reset_raw_exists": false,
1021
+ "forward_recording_required": true,
1022
+ "forward_recording_valid": true,
1023
+ "target_forward_success": false
1024
+ },
1025
+ {
1026
+ "episode": 20,
1027
+ "success": true,
1028
+ "error": null,
1029
+ "forward_success": true,
1030
+ "forward_judge": "TRUE",
1031
+ "reset_success": true,
1032
+ "reset_judge": "TRUE",
1033
+ "forward_judge_input": {
1034
+ "judge_mode": "standard_forward_vlm",
1035
+ "model": "gemini-3.1-flash-lite",
1036
+ "image_inputs": [
1037
+ {
1038
+ "role": "initial",
1039
+ "label": "Image 1: initial top-view before task execution",
1040
+ "path": "episode_20/forward/initial_state.jpg"
1041
+ },
1042
+ {
1043
+ "role": "final",
1044
+ "label": "Image 2: final top-view after task execution",
1045
+ "path": "episode_20/forward/final_state.jpg"
1046
+ }
1047
+ ],
1048
+ "prompt": {
1049
+ "path": "episode_20/forward/judge_prompt.txt",
1050
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
1051
+ "chars": 3076
1052
+ },
1053
+ "raw_response": {
1054
+ "path": "episode_20/forward/judge_raw_response.txt",
1055
+ "sha256": "728aed0540e4901fa26864d672cb698263aa33099a55f20525e3b10c0e4abf91",
1056
+ "chars": 2249
1057
+ },
1058
+ "usage": {
1059
+ "model": "gemini-3.1-flash-lite",
1060
+ "inference_time_s": 3.29,
1061
+ "input_tokens": 2899,
1062
+ "output_tokens": 604,
1063
+ "total_tokens": 3503
1064
+ }
1065
+ },
1066
+ "excluded": false,
1067
+ "forward_raw_exists": true,
1068
+ "reset_raw_exists": false,
1069
+ "forward_recording_required": true,
1070
+ "forward_recording_valid": true,
1071
+ "target_forward_success": true
1072
+ },
1073
+ {
1074
+ "episode": 21,
1075
+ "success": true,
1076
+ "error": null,
1077
+ "forward_success": true,
1078
+ "forward_judge": "TRUE",
1079
+ "reset_success": true,
1080
+ "reset_judge": "FALSE",
1081
+ "forward_judge_input": {
1082
+ "judge_mode": "standard_forward_vlm",
1083
+ "model": "gemini-3.1-flash-lite",
1084
+ "image_inputs": [
1085
+ {
1086
+ "role": "initial",
1087
+ "label": "Image 1: initial top-view before task execution",
1088
+ "path": "episode_21/forward/initial_state.jpg"
1089
+ },
1090
+ {
1091
+ "role": "final",
1092
+ "label": "Image 2: final top-view after task execution",
1093
+ "path": "episode_21/forward/final_state.jpg"
1094
+ }
1095
+ ],
1096
+ "prompt": {
1097
+ "path": "episode_21/forward/judge_prompt.txt",
1098
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
1099
+ "chars": 3076
1100
+ },
1101
+ "raw_response": {
1102
+ "path": "episode_21/forward/judge_raw_response.txt",
1103
+ "sha256": "771c7ab6d84d7a9e75eb7652df66d73522d117ea6bf6ad5e9c10a4411c8bcf15",
1104
+ "chars": 2161
1105
+ },
1106
+ "usage": {
1107
+ "model": "gemini-3.1-flash-lite",
1108
+ "inference_time_s": 3.97,
1109
+ "input_tokens": 2899,
1110
+ "output_tokens": 568,
1111
+ "total_tokens": 3467
1112
+ }
1113
+ },
1114
+ "excluded": false,
1115
+ "forward_raw_exists": true,
1116
+ "reset_raw_exists": false,
1117
+ "forward_recording_required": true,
1118
+ "forward_recording_valid": true,
1119
+ "target_forward_success": true
1120
+ },
1121
+ {
1122
+ "episode": 22,
1123
+ "success": true,
1124
+ "error": null,
1125
+ "forward_success": true,
1126
+ "forward_judge": "TRUE",
1127
+ "reset_success": true,
1128
+ "reset_judge": "TRUE",
1129
+ "forward_judge_input": {
1130
+ "judge_mode": "standard_forward_vlm",
1131
+ "model": "gemini-3.1-flash-lite",
1132
+ "image_inputs": [
1133
+ {
1134
+ "role": "initial",
1135
+ "label": "Image 1: initial top-view before task execution",
1136
+ "path": "episode_22/forward/initial_state.jpg"
1137
+ },
1138
+ {
1139
+ "role": "final",
1140
+ "label": "Image 2: final top-view after task execution",
1141
+ "path": "episode_22/forward/final_state.jpg"
1142
+ }
1143
+ ],
1144
+ "prompt": {
1145
+ "path": "episode_22/forward/judge_prompt.txt",
1146
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
1147
+ "chars": 3076
1148
+ },
1149
+ "raw_response": {
1150
+ "path": "episode_22/forward/judge_raw_response.txt",
1151
+ "sha256": "d224c86e90579512877dd915f10b90ab990e336274e649d92c1d3b1dcaed8c81",
1152
+ "chars": 2067
1153
+ },
1154
+ "usage": {
1155
+ "model": "gemini-3.1-flash-lite",
1156
+ "inference_time_s": 2.92,
1157
+ "input_tokens": 2899,
1158
+ "output_tokens": 556,
1159
+ "total_tokens": 3455
1160
+ }
1161
+ },
1162
+ "excluded": false,
1163
+ "forward_raw_exists": true,
1164
+ "reset_raw_exists": false,
1165
+ "forward_recording_required": true,
1166
+ "forward_recording_valid": true,
1167
+ "target_forward_success": true
1168
+ },
1169
+ {
1170
+ "episode": 23,
1171
+ "success": true,
1172
+ "error": null,
1173
+ "forward_success": true,
1174
+ "forward_judge": "TRUE",
1175
+ "reset_success": true,
1176
+ "reset_judge": "TRUE",
1177
+ "forward_judge_input": {
1178
+ "judge_mode": "standard_forward_vlm",
1179
+ "model": "gemini-3.1-flash-lite",
1180
+ "image_inputs": [
1181
+ {
1182
+ "role": "initial",
1183
+ "label": "Image 1: initial top-view before task execution",
1184
+ "path": "episode_23/forward/initial_state.jpg"
1185
+ },
1186
+ {
1187
+ "role": "final",
1188
+ "label": "Image 2: final top-view after task execution",
1189
+ "path": "episode_23/forward/final_state.jpg"
1190
+ }
1191
+ ],
1192
+ "prompt": {
1193
+ "path": "episode_23/forward/judge_prompt.txt",
1194
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
1195
+ "chars": 3076
1196
+ },
1197
+ "raw_response": {
1198
+ "path": "episode_23/forward/judge_raw_response.txt",
1199
+ "sha256": "c003f659ab398fed4b196fa812fc37f918253cb4f944db841d94d3821bf1e996",
1200
+ "chars": 2034
1201
+ },
1202
+ "usage": {
1203
+ "model": "gemini-3.1-flash-lite",
1204
+ "inference_time_s": 7.41,
1205
+ "input_tokens": 2899,
1206
+ "output_tokens": 538,
1207
+ "total_tokens": 3437
1208
+ }
1209
+ },
1210
+ "excluded": false,
1211
+ "forward_raw_exists": true,
1212
+ "reset_raw_exists": false,
1213
+ "forward_recording_required": true,
1214
+ "forward_recording_valid": true,
1215
+ "target_forward_success": true
1216
+ },
1217
+ {
1218
+ "episode": 24,
1219
+ "success": true,
1220
+ "error": null,
1221
+ "forward_success": true,
1222
+ "forward_judge": "TRUE",
1223
+ "reset_success": true,
1224
+ "reset_judge": "TRUE",
1225
+ "forward_judge_input": {
1226
+ "judge_mode": "standard_forward_vlm",
1227
+ "model": "gemini-3.1-flash-lite",
1228
+ "image_inputs": [
1229
+ {
1230
+ "role": "initial",
1231
+ "label": "Image 1: initial top-view before task execution",
1232
+ "path": "episode_24/forward/initial_state.jpg"
1233
+ },
1234
+ {
1235
+ "role": "final",
1236
+ "label": "Image 2: final top-view after task execution",
1237
+ "path": "episode_24/forward/final_state.jpg"
1238
+ }
1239
+ ],
1240
+ "prompt": {
1241
+ "path": "episode_24/forward/judge_prompt.txt",
1242
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
1243
+ "chars": 3076
1244
+ },
1245
+ "raw_response": {
1246
+ "path": "episode_24/forward/judge_raw_response.txt",
1247
+ "sha256": "89625d929c63954d851f75009961b129f2a0d4f3fe5248228c40fce5f7a85d1e",
1248
+ "chars": 2196
1249
+ },
1250
+ "usage": {
1251
+ "model": "gemini-3.1-flash-lite",
1252
+ "inference_time_s": 2.89,
1253
+ "input_tokens": 2899,
1254
+ "output_tokens": 591,
1255
+ "total_tokens": 3490
1256
+ }
1257
+ },
1258
+ "excluded": false,
1259
+ "forward_raw_exists": true,
1260
+ "reset_raw_exists": false,
1261
+ "forward_recording_required": true,
1262
+ "forward_recording_valid": true,
1263
+ "target_forward_success": true
1264
+ },
1265
+ {
1266
+ "episode": 25,
1267
+ "success": true,
1268
+ "error": null,
1269
+ "forward_success": true,
1270
+ "forward_judge": "TRUE",
1271
+ "reset_success": true,
1272
+ "reset_judge": "TRUE",
1273
+ "forward_judge_input": {
1274
+ "judge_mode": "standard_forward_vlm",
1275
+ "model": "gemini-3.1-flash-lite",
1276
+ "image_inputs": [
1277
+ {
1278
+ "role": "initial",
1279
+ "label": "Image 1: initial top-view before task execution",
1280
+ "path": "episode_25/forward/initial_state.jpg"
1281
+ },
1282
+ {
1283
+ "role": "final",
1284
+ "label": "Image 2: final top-view after task execution",
1285
+ "path": "episode_25/forward/final_state.jpg"
1286
+ }
1287
+ ],
1288
+ "prompt": {
1289
+ "path": "episode_25/forward/judge_prompt.txt",
1290
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
1291
+ "chars": 3076
1292
+ },
1293
+ "raw_response": {
1294
+ "path": "episode_25/forward/judge_raw_response.txt",
1295
+ "sha256": "7796f3c82a2ddbeb54f399ec30e24815406f0e0387ae957c8e43b1a99258b34a",
1296
+ "chars": 1937
1297
+ },
1298
+ "usage": {
1299
+ "model": "gemini-3.1-flash-lite",
1300
+ "inference_time_s": 3.0,
1301
+ "input_tokens": 2899,
1302
+ "output_tokens": 514,
1303
+ "total_tokens": 3413
1304
+ }
1305
+ },
1306
+ "excluded": false,
1307
+ "forward_raw_exists": true,
1308
+ "reset_raw_exists": false,
1309
+ "forward_recording_required": true,
1310
+ "forward_recording_valid": true,
1311
+ "target_forward_success": true
1312
+ },
1313
+ {
1314
+ "episode": 26,
1315
+ "success": true,
1316
+ "error": null,
1317
+ "forward_success": true,
1318
+ "forward_judge": "TRUE",
1319
+ "reset_success": true,
1320
+ "reset_judge": "TRUE",
1321
+ "forward_judge_input": {
1322
+ "judge_mode": "standard_forward_vlm",
1323
+ "model": "gemini-3.1-flash-lite",
1324
+ "image_inputs": [
1325
+ {
1326
+ "role": "initial",
1327
+ "label": "Image 1: initial top-view before task execution",
1328
+ "path": "episode_26/forward/initial_state.jpg"
1329
+ },
1330
+ {
1331
+ "role": "final",
1332
+ "label": "Image 2: final top-view after task execution",
1333
+ "path": "episode_26/forward/final_state.jpg"
1334
+ }
1335
+ ],
1336
+ "prompt": {
1337
+ "path": "episode_26/forward/judge_prompt.txt",
1338
+ "sha256": "81884c824d596f57585b8c04ca0f88e7bc2dac2b9bb0c07f0641979a7fb01fb7",
1339
+ "chars": 3076
1340
+ },
1341
+ "raw_response": {
1342
+ "path": "episode_26/forward/judge_raw_response.txt",
1343
+ "sha256": "aa8b0f39cb772d4d9b61375e89f5fca9a6797802e6f3ff10e41a8f52f3411939",
1344
+ "chars": 2036
1345
+ },
1346
+ "usage": {
1347
+ "model": "gemini-3.1-flash-lite",
1348
+ "inference_time_s": 6.04,
1349
+ "input_tokens": 2899,
1350
+ "output_tokens": 559,
1351
+ "total_tokens": 3458
1352
+ }
1353
+ },
1354
+ "excluded": false,
1355
+ "forward_raw_exists": true,
1356
+ "reset_raw_exists": false,
1357
+ "forward_recording_required": true,
1358
+ "forward_recording_valid": true,
1359
+ "target_forward_success": true
1360
+ },
1361
+ {
1362
+ "episode": 27,
1363
+ "success": true,
1364
+ "error": null,
1365
+ "forward_success": true,
1366
+ "forward_judge": "TRUE",
1367
+ "reset_success": true,
1368
+ "reset_judge": "TRUE",
1369
+ "forward_judge_input": {
1370
+ "judge_mode": "standard_forward_vlm",
1371
+ "model": "gemini-3.1-flash-lite",
1372
+ "image_inputs": [
1373
+ {
1374
+ "role": "initial",
1375
+ "label": "Image 1: initial top-view before task execution",
1376
+ "path": "episode_27/forward/initial_state.jpg"
1377
+ },
1378
+ {
1379
+ "role": "final",
1380
+ "label": "Image 2: final top-view after task execution",
1381
+ "path": "episode_27/forward/final_state.jpg"
1382
+ }
1383
+ ],
1384
+ "prompt": {
1385
+ "path": "episode_27/forward/judge_prompt.txt",
1386
+ "sha256": "09bd9c7b270ed0954c83bb6f48345c04c48d4a597c1f0a9b2b76a052c7799977",
1387
+ "chars": 3076
1388
+ },
1389
+ "raw_response": {
1390
+ "path": "episode_27/forward/judge_raw_response.txt",
1391
+ "sha256": "f8c1ffe80b1b664a930706a6d6b3b62732f22a28d3ffdac7bbdde012b923a11a",
1392
+ "chars": 1989
1393
+ },
1394
+ "usage": {
1395
+ "model": "gemini-3.1-flash-lite",
1396
+ "inference_time_s": 4.13,
1397
+ "input_tokens": 2899,
1398
+ "output_tokens": 530,
1399
+ "total_tokens": 3429
1400
+ }
1401
+ },
1402
+ "excluded": false,
1403
+ "forward_raw_exists": true,
1404
+ "reset_raw_exists": false,
1405
+ "forward_recording_required": true,
1406
+ "forward_recording_valid": true,
1407
+ "target_forward_success": true
1408
+ }
1409
+ ]
1410
+ }
session/subgoal_buffer.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f11ec4ac1aca73d9766e9af8611a5e2fe8bfac3389a4928c11e30e7acfd1697
3
+ size 302162
skill_dct/UR7e_phase1_Sort_RGBblock_10fps.canonical.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a3a60175a2e1462ea85b75f66e5e4ba599a249b81d084a0309821866ea1770f
3
+ size 1361616
skill_dct/UR7e_phase1_Sort_RGBblock_10fps.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d3fb466912f11a750c736c7b8cc29fb751775e9c15945b2db3dd10e1ec26b77a
3
+ size 1358948
vector_db/server_skill_wise_vector_db.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9748b78a48c19e5bd87f865f220d7fa9d6141a7777b53855752d522926cce0ab
3
+ size 7102690
vector_db/skill_wise_vector_db.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9748b78a48c19e5bd87f865f220d7fa9d6141a7777b53855752d522926cce0ab
3
+ size 7102690