yxma commited on
Commit
34e7ca3
·
verified ·
1 Parent(s): 15cb33e

README: update episode count to 30; note dropped 2026-05-11 episodes

Browse files
Files changed (1) hide show
  1. README.md +8 -5
README.md CHANGED
@@ -45,7 +45,7 @@ The dataset is organized **task-first** so new tasks can be added without renami
45
 
46
  | Task | Description | Dates | Episodes |
47
  |---|---|---|---|
48
- | `motherboard` | Bimanual manipulation of components on a computer motherboard | 2026-03-23, 2026-05-10, 2026-05-11 | 32 |
49
 
50
  `tasks.json` at the repo root is the source of truth for the task registry.
51
 
@@ -115,11 +115,11 @@ from huggingface_hub import hf_hub_download
115
  path = hf_hub_download(
116
  repo_id="yxma/React",
117
  repo_type="dataset",
118
- filename="processed/mode1_v1/motherboard/2026-05-11/episode_000.pt",
119
  )
120
  ep = torch.load(path, weights_only=False)
121
  print(ep["view"].shape, ep["tactile_left_intensity"].shape)
122
- # torch.Size([263, 3, 128, 128]) torch.Size([263])
123
  ```
124
 
125
  Load contact metadata only (much smaller — useful for filtering):
@@ -129,7 +129,7 @@ import json
129
  path = hf_hub_download(
130
  repo_id="yxma/React",
131
  repo_type="dataset",
132
- filename="processed/mode1_v1/motherboard/2026-05-11/episode_000.contact.json",
133
  )
134
  meta = json.load(open(path))
135
  print(meta["drift_left"], meta["drift_right"], meta["drift_warning"])
@@ -137,7 +137,10 @@ print(meta["drift_left"], meta["drift_right"], meta["drift_warning"])
137
 
138
  ## Known caveats
139
 
140
- - **Missing episode**: `motherboard/2026-05-11/episode_001` was lost at recording time (HDF5 superblock never finalized when the writer was killed mid-write). It is intentionally absent.
 
 
 
141
  - **Lossy resize**: the 128×128 `view` and tactile fields are downsampled from native 480×640. Native resolution is **not** preserved in this release.
142
  - **Single camera**: only `realsense/cam0/color` is included. The other two RealSense views and all depth streams are not in `processed/mode1_v1/`.
143
  - **OptiTrack alignment**: the per-step poses are nearest-neighbor matched to camera ticks. The full ~120 Hz pose streams are not preserved here.
 
45
 
46
  | Task | Description | Dates | Episodes |
47
  |---|---|---|---|
48
+ | `motherboard` | Bimanual manipulation of components on a computer motherboard | 2026-03-23, 2026-05-10, 2026-05-11 | 30 |
49
 
50
  `tasks.json` at the repo root is the source of truth for the task registry.
51
 
 
115
  path = hf_hub_download(
116
  repo_id="yxma/React",
117
  repo_type="dataset",
118
+ filename="processed/mode1_v1/motherboard/2026-05-11/episode_003.pt",
119
  )
120
  ep = torch.load(path, weights_only=False)
121
  print(ep["view"].shape, ep["tactile_left_intensity"].shape)
122
+ # torch.Size([10032, 3, 128, 128]) torch.Size([10032])
123
  ```
124
 
125
  Load contact metadata only (much smaller — useful for filtering):
 
129
  path = hf_hub_download(
130
  repo_id="yxma/React",
131
  repo_type="dataset",
132
+ filename="processed/mode1_v1/motherboard/2026-05-11/episode_003.contact.json",
133
  )
134
  meta = json.load(open(path))
135
  print(meta["drift_left"], meta["drift_right"], meta["drift_warning"])
 
137
 
138
  ## Known caveats
139
 
140
+ - **Missing / dropped episodes** on `motherboard/2026-05-11`:
141
+ - `episode_000` and `episode_002` — short test recordings (8.8s and 10.4s) with no tactile contact on either sensor; intentionally excluded.
142
+ - `episode_001` — lost at recording time (HDF5 superblock never finalized when the writer was killed mid-write); intentionally absent.
143
+ - The remaining episode IDs are non-contiguous as a result. Don't infer ordering from filename gaps.
144
  - **Lossy resize**: the 128×128 `view` and tactile fields are downsampled from native 480×640. Native resolution is **not** preserved in this release.
145
  - **Single camera**: only `realsense/cam0/color` is included. The other two RealSense views and all depth streams are not in `processed/mode1_v1/`.
146
  - **OptiTrack alignment**: the per-step poses are nearest-neighbor matched to camera ticks. The full ~120 Hz pose streams are not preserved here.