Datasets:

yxma
/

React

@@ -26,11 +26,11 @@ configs:
 # React
-Multimodal manipulation recordings from a bimanual setup with vision-based tactile sensors and motion capture.
 ![Tactile intensity timeline](figures/contact_intensity_full.png)
-> **30 episodes · 138.4 min total · 87.9 min (66%) of confirmed bimanual tactile contact · 3× RGB-D cameras + 2× GelSight + 3-body OptiTrack.**
 ## At a glance
@@ -38,293 +38,76 @@ Multimodal manipulation recordings from a bimanual setup with vision-based tacti
 |---|---|
 | Tasks | `motherboard` (more coming) |
 | Episodes | 30 |
-| Total duration | **138.4 min** (median 4 min/episode, longest 19 min) |
-| Tactile contact | **87.9 min / 66% of frames** (4,136 contact events, median 0.73 s each) |
-| Cameras | 3× Intel RealSense D415 (color + depth, 480×640, 30 FPS) |
-| Tactile | 2× GelSight Mini (left / right, 480×640, ~25 FPS) |
 | Motion capture | OptiTrack VRPN, 3 rigid bodies, ~120 Hz |
 | License | CC-BY-4.0 |
-![Comparison table](figures/dataset_figures/F7_comparison_table.png)
-## Recording setup
-| Stream | Hardware | Native shape | Rate |
-|---|---|---|---|
-| 3× RealSense color | Intel D415 (serials `143322063538`, `104122062574`, `217222066989`) | 480×640×3 uint8 (BGR) | 30 FPS |
-| 3× RealSense depth | same | 480×640 uint16 (mm) | 30 FPS |
-| 2× GelSight tactile | GelSight Mini (left / right) | 480×640×3 uint8 | ~25 FPS, resampled to camera ticks |
-| 3× OptiTrack rigid bodies | `motherboard`, `sensor_left`, `sensor_right` | 7-vector (x, y, z, qx, qy, qz, qw) | ~120 Hz |
-All streams share a common monotonic timestamp axis recorded under `timestamps`. Per-tracker OptiTrack streams carry their own higher-rate timestamps under `optitrack/<body>/timestamps`.
-## Tasks
-The dataset is organized **task-first** so new tasks can be added without renaming or recompute.
-| Task | Description | Dates | Episodes |
-|---|---|---|---|
-| `motherboard` | Bimanual manipulation of components on a computer motherboard | 2026-03-23, 2026-05-10, 2026-05-11 | 30 |
-`tasks.json` at the repo root is the source of truth for the task registry.
-## Episode previews
-Each `figures/episode_previews/motherboard/<date>/episode_NNN.gif` shows the first 2 minutes of that episode at 15× speed (≈8 s loop, 3-panel layout: overhead camera | tactile left | tactile right). Browse them in the [`figures/episode_previews`](figures/episode_previews) folder of this repo.
-## Statistics & analysis
-### Episode length
-![Episode length](figures/dataset_figures/F1_episode_length_histogram.png)
-Most episodes run 1–10 min; the median is **4 min** — roughly 8× longer than BridgeData V2's typical 30 s demo. The longest episode is 19 min.
-### Contact event durations
-![Contact event durations](figures/dataset_figures/F2_contact_event_duration_histogram.png)
-4,136 contact events total. The typical contact event lasts ≈ **0.7 s** (median), with a long tail out to 33 s — useful for grasp/contact-classification downstream tasks.
-### Where on the gel does contact land?
-![Contact heatmap](figures/dataset_figures/F3_contact_heatmap.png)
-Both sensors show contact concentrated in the central ~50% of the gel surface, dropping off toward the edges. The left gel has discrete bright spots from repeated contacts with specific features.
-### Tactile signal is real and varied (not flat noise)
-![Tactile montage](figures/dataset_figures/F4_tactile_pattern_montage.png)
-16 random contact frames sampled across the dataset — discrete pins, edges, smooth surfaces, multi-object contact.
-### Bimanual workspace
-![Pose trajectory](figures/dataset_figures/F5_pose_trajectory_3d.png)
-Multi-view projection of the longest episode (2026-05-11 / ep_017, 19 min). Left (blue) and right (orange) sensors operate over a ~30 × 40 × 30 cm workspace.
-### Tactile is independent of motion
-![Cross-modal correlation](figures/dataset_figures/F6_cross_modal_correlation.png)
-Sensor velocity vs tactile intensity is **essentially uncorrelated** (r ≈ +0.04 / −0.05). Tactile carries information that is **not** explained by pose+velocity — a direct argument for the value of including tactile in policy / world-model training.
-### Per-episode summary
-![Per-episode summary](figures/dataset_figures/F8_per_episode_summary.png)
-Detailed table also exported as CSV: [`figures/dataset_figures/F8_per_episode_summary.csv`](figures/dataset_figures/F8_per_episode_summary.csv).
-## Repository layout
-```
-tasks.json                                          # registry
-processed/
-└─ mode1_v1/
-    ���─ <task>/
-        └─ <date>/
-            ├─ episode_000.pt
-            ├─ episode_000.contact.json
-            └─ ...
-figures/
-├─ contact_intensity_full.png          # tactile intensity over full dataset (waveform view)
-├─ contact_intensity_20min.png         # 20-min zoom of the same
-├─ episode_previews/<task>/<date>/episode_*.gif   # per-episode GIF previews
-└─ dataset_figures/                    # F1–F8 statistics and analysis figures
-```
-The `processed/mode1_v1/` view is a **task-specific slice** of the underlying raw recordings, not the full sensor suite. It was produced by `twm/preprocess.py` + `twm/contact_index.py` from a private raw HDF5 mirror.
-## `processed/mode1_v1/` schema
-Each `episode_*.pt` is a Python dict loadable with `torch.load(..., weights_only=False)`.
-| Key | Shape | dtype | Description |
-|---|---|---|---|
-| `view` | `(T, 3, 128, 128)` | uint8 | Overhead camera (`realsense/cam0/color`), center-cropped to square then bilinear-resized to 128×128 |
-| `tactile_left` | `(T, 3, 128, 128)` | uint8 | Left GelSight, same crop/resize |
-| `tactile_right` | `(T, 3, 128, 128)` | uint8 | Right GelSight, same crop/resize |
-| `timestamps` | `(T,)` | float64 | Camera timestamps (seconds, monotonic clock) |
-| `sensor_left_pose` | `(T, 7)` | float32 | Left GelSight rigid body OptiTrack pose, nearest-neighbor aligned to camera timestamps |
-| `sensor_right_pose` | `(T, 7)` | float32 | Right GelSight rigid body OptiTrack pose, same alignment |
-| `tactile_{left,right}_intensity` | `(T,)` | float32 | Per-frame mean per-pixel L2 distance from a contact-free reference frame |
-| `tactile_{left,right}_area` | `(T,)` | float32 | Per-frame fraction of pixels with L2 diff > `tau` |
-| `tactile_{left,right}_mixed` | `(T,)` | float32 | Mean of (diff × mask), captures intensity restricted to contact pixels |
-| `_contact_meta` | dict | — | Per-episode contact metadata: `tau`, drift between first/p01 reference frames, p01 reference indices, the chosen reference RGB frames, etc. |
-Each `.contact.json` is a small summary of the metric distributions plus drift diagnostics, intended for filtering / sanity checking without loading the full tensors.
-### Contact metric definition
-For each tactile sensor independently:
-1. Pick a contact-free **reference frame**: the ~0.1th-percentile-quietest frame by mean L2 distance to the temporal median (`reference_strategy = "p01"`).
-2. For each frame `t`, compute per-pixel `diff[t, x, y] = || frame[t, :, x, y] − ref[:, x, y] ||_2` (RGB L2 over channels).
-3. Then:
-   - `intensity[t]   = mean(diff[t])`
-   - `area[t]        = mean(diff[t] > tau)`        (default `tau = 8.0` on the uint8 scale)
-   - `mixed[t]       = mean(diff[t] * (diff[t] > tau))`
-`_contact_meta["drift_warning"]` is `True` if either sensor's drift (L2 distance between the first frame and the p01-reference frame) exceeds `2·tau`; in this release no episode triggers it.
 ## Quick start
-Load one task via `datasets`:
 ```python
 from datasets import load_dataset
 ds = load_dataset("yxma/React", "motherboard", split="train")
-# Each row is one .pt file path; the actual tensors live inside.
 ```
-Or load a single episode directly:
 ```python
 import torch
 from huggingface_hub import hf_hub_download
 path = hf_hub_download(
-    repo_id="yxma/React",
-    repo_type="dataset",
     filename="processed/mode1_v1/motherboard/2026-05-11/episode_003.pt",
 )
 ep = torch.load(path, weights_only=False)
-print(ep["view"].shape, ep["tactile_left_intensity"].shape)
-# torch.Size([10032, 3, 128, 128]) torch.Size([10032])
-```
-Load contact metadata only (much smaller — useful for filtering):
-```python
-import json
-path = hf_hub_download(
-    repo_id="yxma/React",
-    repo_type="dataset",
-    filename="processed/mode1_v1/motherboard/2026-05-11/episode_003.contact.json",
-)
-meta = json.load(open(path))
-print(meta["drift_left"], meta["drift_right"], meta["drift_warning"])
 ```
-## Data quality and recommended sampling
-A small fraction of frames contain known sensor artifacts. The repo ships a
-`bad_frames.json` index so downstream code can avoid sampling on top of these
-intervals — useful since most training pipelines sample short windows that
-should not straddle a glitch.
-### Failure modes (quantitative breakdown)
-| # | Mode | Frames affected | % of dataset | Episodes | Symptom | Example |
-|---|---|---:|---:|---:|---|---|
-| 1 | **GelSight LED flicker** | 66 | **0.028 %** | 5 / 30 | 1–2 frames of uniform pink/magenta wash across the gel surface; adjacent frames normal. 2026-05-11/{003, 007, 008, 011, 017}. | `figures/dataset_figures/intensity_spike_samples/intensity_spike_overview.png` |
-| 2 | **OptiTrack pose teleport (>5 m/s)** | 136 | **0.057 %** | 5 / 30 | Position jumps by >5 cm in 33 ms (physically impossible) — mocap lost lock and reacquired. 2026-03-23/{001, 002}, 2026-05-10/{001, 002}, 2026-05-11/015. | `figures/dataset_figures/pose_teleport_samples/` |
-| 3 | Sensor at rest (>30 s freeze) | 37,601 | **15.68 %** | 5 / 30 | Multi-minute periods with pose ≈ const and tactile ≈ const (sensor set down on the table). **Not corruption**, but degenerate training data. 2026-05-11/{012: 5.4 min, 017: 10.7 min}, 2026-05-11/005 (1.3 min), 2026-03-23/{001: 1.8 min, 002: 1.6 min}. | `figures/dataset_figures/data_quality_report.png` |
-**Modes 1+2 (real corruption) together: 202 / 239,759 frames = 0.084 % of the dataset.** Mode 3 is tracked separately because it's healthy data, just not informative for dynamics learning. `bad_frames.json` (below) covers modes 1 and 2 only; if you want to also skip rest periods, intersect with the velocity track from `sensor_*_pose`.
-### Inspection figures (rendered inline)
-**Mode 1 — GelSight LED flicker** (overview across all 5 affected episodes):
-![Intensity spike overview](figures/dataset_figures/intensity_spike_samples/intensity_spike_overview.png)
-Per-episode close-ups (reference frame, ±1 s neighbors, peak frame): `figures/dataset_figures/intensity_spike_samples/2026-05-11_episode_{003,007,008,011,017}_*.png`.
-**Mode 2 — OptiTrack pose teleport** (2 examples for the dates without raw camera streams; the other 4 affected episodes have GIFs in the same folder):
-![Teleport 2026-03-23 ep001](figures/dataset_figures/pose_teleport_samples/2026-03-23/episode_001_teleport.png)
-![Teleport 2026-03-23 ep002](figures/dataset_figures/pose_teleport_samples/2026-03-23/episode_002_teleport.png)
-GIFs (10 s playback at 2× speed) for the other 4: `figures/dataset_figures/pose_teleport_samples/2026-05-{10,11}/episode_*_teleport.gif`.
-**Modes 1+2+3 — top-6 worst-offenders chart** (tactile intensity + sensor velocity time-series):
-![Data quality report](figures/dataset_figures/data_quality_report.png)
-### `bad_frames.json` schema
-The file lives at the repo root. Frame indices are inclusive on both ends and
-pre-padded by `buffer_frames` (3) on each side so context windows don't bleed
-into the glitch:
-```json
-{
-  "tau_intensity": 30.0,
-  "tau_velocity_mps": 5.0,
-  "buffer_frames": 3,
-  "summary": {
-    "n_episodes": 30,
-    "total_frames": 239759,
-    "total_bad_frames": 202,
-    "bad_fraction_overall": 0.000843,
-    "n_episodes_with_bad_frames": 10
-  },
-  "episodes": {
-    "2026-05-11/episode_003": {
-      "n_frames": 10032,
-      "duration_s": 338.2,
-      "intensity_spikes": [[260, 268], [...]],
-      "pose_teleports_L": [],
-      "pose_teleports_R": [],
-      "total_bad_frames": 19,
-      "bad_fraction": 0.0019
-    }
-  }
-}
 ```
-### Recommended use in a dataloader
-```python
-import json
-with open("bad_frames.json") as f:
-    bad = json.load(f)["episodes"]
-def is_clean_window(ep_name, t_start, t_end):
-    """Return True if [t_start, t_end] does not overlap any flagged interval."""
-    intervals = (bad[ep_name]["intensity_spikes"]
-                 + bad[ep_name]["pose_teleports_L"]
-                 + bad[ep_name]["pose_teleports_R"])
-    for s, e in intervals:
-        if s <= t_end and e >= t_start:
-            return False
-    return True
-# In your sampler: if not is_clean_window(...), resample.
 ```
-### Inspection figures
-Detailed per-episode evidence is published alongside the data:
-- **Intensity spikes**: `figures/dataset_figures/intensity_spike_samples/` — one PNG per affected episode showing the spike frame and ±1 s of neighbors next to the reference frame.
-- **Pose teleports**: `figures/dataset_figures/pose_teleport_samples/`
-  - 4 focused GIFs (10 s playback, 2× speed) around the worst velocity event for episodes where the raw camera streams were available.
-  - 2 static PNGs (with velocity time-series, top-down + side trajectories, and tactile/view thumbnails) for the 2026-03-23 episodes where only the processed `.pt` is on the publishing machine.
-- **Full report**: `figures/dataset_figures/data_quality_report.{png,csv}` — top-6 worst offenders and a per-episode CSV with all 30 episodes' tactile / pose stats.
-## Known caveats
-- **Missing / dropped episodes** on `motherboard/2026-05-11`:
-  - `episode_000` and `episode_002` — short test recordings (8.8 s and 10.4 s) with no tactile contact on either sensor; intentionally excluded.
-  - `episode_001` — lost at recording time (HDF5 superblock never finalized when the writer was killed mid-write); intentionally absent.
-  - The remaining episode IDs are non-contiguous as a result. Don't infer ordering from filename gaps.
-- **Lossy resize**: the 128×128 `view` and tactile fields are downsampled from native 480×640. Native resolution is **not** preserved in this release.
-- **Single camera**: only `realsense/cam0/color` is included. The other two RealSense views and all depth streams are not in `processed/mode1_v1/`.
-- **OptiTrack alignment**: the per-step poses are nearest-neighbor matched to camera ticks. The full ~120 Hz pose streams are not preserved here.
-- **Mode is opinionated**: contact metrics depend on the chosen `tau` and the p01 reference strategy. If you want a different `tau`, re-deriving from raw is necessary.
-## Roadmap
-- **More tasks** — registry in `tasks.json` will grow.
-- **LeRobot-format full-fidelity** variant (`lerobot/v1.0/`) is planned. It will include all three RealSense color and depth streams, GelSight at native resolution (FFV1 lossless), full-rate OptiTrack pose tracks for all three rigid bodies, and HF-native browser previews. The current `processed/mode1_v1/` slice will remain as a stable training-task view.
 ## Citation
 If you use this dataset, please cite (TODO: add bibtex).

 # React
+Bimanual manipulation recordings with vision-based tactile sensors and motion capture.
 ![Tactile intensity timeline](figures/contact_intensity_full.png)
+> **30 episodes · 138.4 min · 87.9 min (66 %) bimanual tactile contact · 3 × RGB-D + 2 × GelSight + 3-body OptiTrack**
 ## At a glance
 |---|---|
 | Tasks | `motherboard` (more coming) |
 | Episodes | 30 |
+| Total duration | 138.4 min (median 4 min, longest 19 min) |
+| Tactile contact | 87.9 min — 66 % of frames; 4,136 contact events, median 0.73 s |
+| Cameras | 3× Intel RealSense D415 (color + depth), 480×640, 30 FPS |
+| Tactile | 2× GelSight Mini (left, right) |
 | Motion capture | OptiTrack VRPN, 3 rigid bodies, ~120 Hz |
 | License | CC-BY-4.0 |
+![Comparison with other manipulation datasets](figures/dataset_figures/F7_comparison_table.png)
 ## Quick start
+Load by task with `datasets`:
 ```python
 from datasets import load_dataset
 ds = load_dataset("yxma/React", "motherboard", split="train")
 ```
+Or grab a single episode:
 ```python
 import torch
 from huggingface_hub import hf_hub_download
 path = hf_hub_download(
+    repo_id="yxma/React", repo_type="dataset",
     filename="processed/mode1_v1/motherboard/2026-05-11/episode_003.pt",
 )
 ep = torch.load(path, weights_only=False)
+# ep["view"]                                 (T, 3, 128, 128) uint8 — overhead cam
+# ep["tactile_left"], ep["tactile_right"]    (T, 3, 128, 128) uint8
+# ep["sensor_left_pose"], ep["sensor_right_pose"]
+#                                            (T, 7) float32 — xyz + quaternion
+# ep["timestamps"]                           (T,)  float64
+# Plus per-frame contact metrics:            tactile_{side}_{intensity, area, mixed}
 ```
+**Skip the 0.085 %** of frames flagged in [`bad_frames.json`](bad_frames.json) — see [`docs/quality.md`](docs/quality.md) for a one-line dataloader snippet.
+See [`tasks.json`](tasks.json) for the task registry.
+## Episode previews
+Per-episode GIF previews live under [`figures/episode_previews/`](figures/episode_previews) — first 2 minutes at 10× speed, showing all 3 RealSense cameras with projected GelSight axes plus both tactile pads.
+## Repository layout
 ```
+README.md                                        # this file
+tasks.json                                       # task registry
+bad_frames.json                                  # data-quality skip-list
+processed/mode1_v1/<task>/<date>/episode_*.pt    # per-episode tensors
+figures/                                         # previews + analysis figures
+docs/                                            # extended documentation
 ```
+## More documentation
+| File | Contents |
+|---|---|
+| [`docs/recording.md`](docs/recording.md) | Hardware setup, camera serials, sensor + mocap layout |
+| [`docs/schema.md`](docs/schema.md) | Full `.pt` field reference and contact-metric definitions |
+| [`docs/quality.md`](docs/quality.md) | Data-quality breakdown (per-mode), `bad_frames.json` schema, dataloader recipe, inspection figures |
+| [`docs/figures.md`](docs/figures.md) | Dataset statistics + analysis gallery (F1–F8) |
+| [`docs/caveats.md`](docs/caveats.md) | Known caveats and roadmap |
+## License
+Released under [Creative Commons Attribution 4.0](https://creativecommons.org/licenses/by/4.0/) (CC-BY-4.0).
 ## Citation
 If you use this dataset, please cite (TODO: add bibtex).

docs/caveats.md ADDED Viewed

	@@ -0,0 +1,19 @@

+# Known caveats and roadmap
+For known sensor-level anomalies and the `bad_frames.json` skip-list, see [`quality.md`](quality.md).
+## Caveats
+- **Missing / dropped episodes** on `motherboard/2026-05-11`:
+  - `episode_000` and `episode_002` — short test recordings (8.8 s and 10.4 s) with no tactile contact on either sensor; intentionally excluded.
+  - `episode_001` — lost at recording time (HDF5 superblock never finalized when the writer was killed mid-write); intentionally absent.
+  - The remaining episode IDs are non-contiguous as a result. **Don't infer ordering from filename gaps.**
+- **Lossy resize**: the 128×128 `view` and tactile fields are downsampled from native 480×640. Native resolution is **not** preserved in this release.
+- **Single camera**: only `realsense/cam0/color` is included. The other two RealSense views and all depth streams are not in `processed/mode1_v1/`.
+- **OptiTrack alignment**: the per-step poses are nearest-neighbor matched to camera ticks. The full ~120 Hz pose streams are not preserved here.
+- **Mode is opinionated**: contact metrics depend on the chosen `tau` and the p01 reference strategy. If you want a different `tau`, re-deriving from raw is necessary.
+## Roadmap
+- **More tasks** — the registry in [`../tasks.json`](../tasks.json) will grow.
+- **LeRobot-format full-fidelity** variant (`lerobot/v1.0/`) is planned. It will include all three RealSense color and depth streams, GelSight at native resolution (FFV1 lossless), full-rate OptiTrack pose tracks for all three rigid bodies, and HF-native browser previews. The current `processed/mode1_v1/` slice will remain as a stable training-task view.

docs/figures.md ADDED Viewed

	@@ -0,0 +1,57 @@

+# Statistics and analysis
+A gallery of dataset-level figures. All sources live under [`figures/dataset_figures/`](../figures/dataset_figures/).
+## F1 — Episode length distribution
+![Episode length](../figures/dataset_figures/F1_episode_length_histogram.png)
+Most episodes run 1–10 minutes (median 4 min, longest 19 min) — about **8× longer than BridgeData V2's typical 30 s demo**.
+## F2 — Contact event durations
+![Contact event durations](../figures/dataset_figures/F2_contact_event_duration_histogram.png)
+4,136 contact events total. The typical event lasts ≈ **0.7 s** (median), tail out to 33 s.
+## F3 — Where on the gel does contact land?
+![Contact heatmap](../figures/dataset_figures/F3_contact_heatmap.png)
+Both sensors show contact concentrated in the central ~50 % of the gel surface, dropping toward the edges. The left gel has discrete bright spots from repeated contacts with specific features.
+## F4 — Tactile signal is real and varied
+![Tactile montage](../figures/dataset_figures/F4_tactile_pattern_montage.png)
+16 random contact frames sampled across the dataset — discrete pins, edges, smooth surfaces, multi-object contact.
+## F5 — Bimanual workspace
+![Pose trajectory](../figures/dataset_figures/F5_pose_trajectory_3d.png)
+Multi-view projection of the longest episode (`2026-05-11/episode_017`, 19 min). Left (blue) and right (orange) sensors operate over a ~30 × 40 × 30 cm workspace.
+## F6 — Tactile is independent of motion
+![Cross-modal correlation](../figures/dataset_figures/F6_cross_modal_correlation.png)
+Sensor velocity vs tactile intensity is **essentially uncorrelated** (r ≈ +0.04 / −0.05). Tactile carries information that is **not** explained by pose+velocity — a direct argument for the value of including tactile in policy / world-model training.
+## F7 — Comparison with other manipulation datasets
+![Comparison table](../figures/dataset_figures/F7_comparison_table.png)
+React is the only entry combining tactile + RGB-D + motion capture + bimanual.
+## F8 — Per-episode summary
+![Per-episode summary](../figures/dataset_figures/F8_per_episode_summary.png)
+Full table for all 30 episodes. CSV: [`figures/dataset_figures/F8_per_episode_summary.csv`](../figures/dataset_figures/F8_per_episode_summary.csv).
+## Tactile-intensity timelines
+![Full dataset intensity timeline](../figures/contact_intensity_full.png)
+138-minute audio-waveform-style view across the entire dataset; left intensity rendered above the x-axis (orange), right intensity mirrored below (blue). A 20-minute zoom is in [`figures/contact_intensity_20min.png`](../figures/contact_intensity_20min.png).

docs/quality.md ADDED Viewed

	@@ -0,0 +1,98 @@

+# Data quality
+A small fraction of frames contain known sensor artifacts. The repo ships a [`../bad_frames.json`](../bad_frames.json) index so downstream code can avoid sampling on top of these intervals — useful since most training pipelines sample short windows that should not straddle a glitch.
+## Headline numbers
+| | |
+|---|---:|
+| Total frames | 239,759 |
+| Episodes | 30 |
+| Frames flagged in `bad_frames.json` | **202 (0.084 %)** |
+| Episodes with ≥1 flagged frame | 10 |
+## Failure modes — quantitative breakdown
+| # | Mode | Frames | % of dataset | Episodes | Symptom | Example |
+|---|---|---:|---:|---:|---|---|
+| 1 | **GelSight LED flicker** | 66 | **0.028 %** | 5 / 30 | 1–2 frames of uniform pink/magenta wash across the gel surface; adjacent frames normal. 2026-05-11 / {003, 007, 008, 011, 017}. | [`intensity_spike_overview.png`](../figures/dataset_figures/intensity_spike_samples/intensity_spike_overview.png) |
+| 2 | **OptiTrack pose teleport (>5 m/s)** | 136 | **0.057 %** | 5 / 30 | Position jumps by >5 cm in 33 ms — mocap lost lock and reacquired. 2026-03-23/{001, 002}, 2026-05-10/{001, 002}, 2026-05-11/015. | [`pose_teleport_samples/`](../figures/dataset_figures/pose_teleport_samples/) |
+| 3 | Sensor at rest (>30 s freeze) | 37,601 | **15.68 %** | 5 / 30 | Multi-minute periods with pose ≈ const and tactile ≈ const (sensor set down on the table). **Not corruption**, but degenerate for dynamics learning. 2026-05-11/{012: 5.4 min, 017: 10.7 min, 005: 1.3 min}, 2026-03-23/{001: 1.8 min, 002: 1.6 min}. | [`data_quality_report.png`](../figures/dataset_figures/data_quality_report.png) |
+**Modes 1+2 (real corruption) together: 202 / 239,759 = 0.084 % of the dataset.** Mode 3 is tracked separately because it is healthy data, just non-informative. `bad_frames.json` covers modes 1 and 2 only; if you also want to skip rest periods, intersect with the velocity track from `sensor_*_pose`.
+## Inspection figures
+**Mode 1 — GelSight LED flicker** (overview across all 5 affected episodes):
+![Intensity spike overview](../figures/dataset_figures/intensity_spike_samples/intensity_spike_overview.png)
+Per-episode close-ups (reference frame, ±1 s neighbors, peak frame): [`figures/dataset_figures/intensity_spike_samples/`](../figures/dataset_figures/intensity_spike_samples/).
+**Mode 2 — OptiTrack pose teleport** (2 examples for the dates without raw camera streams; the other 4 episodes have GIFs in the same folder):
+![Teleport 2026-03-23 ep001](../figures/dataset_figures/pose_teleport_samples/2026-03-23/episode_001_teleport.png)
+![Teleport 2026-03-23 ep002](../figures/dataset_figures/pose_teleport_samples/2026-03-23/episode_002_teleport.png)
+GIFs (10 s playback at 2× speed) for the other 4: [`pose_teleport_samples/2026-05-{10,11}/`](../figures/dataset_figures/pose_teleport_samples/).
+**Modes 1+2+3 — top-6 worst-offenders chart** (tactile intensity + sensor velocity time-series):
+![Data quality report](../figures/dataset_figures/data_quality_report.png)
+## `bad_frames.json` schema
+Frame indices are inclusive on both ends and pre-padded by `buffer_frames` (3) on each side so context windows don't bleed into the glitch:
+```json
+{
+  "tau_intensity": 30.0,
+  "tau_velocity_mps": 5.0,
+  "buffer_frames": 3,
+  "summary": {
+    "n_episodes": 30,
+    "total_frames": 239759,
+    "total_bad_frames": 202,
+    "bad_fraction_overall": 0.000843,
+    "n_episodes_with_bad_frames": 10
+  },
+  "episodes": {
+    "2026-05-11/episode_003": {
+      "n_frames": 10032,
+      "duration_s": 338.2,
+      "intensity_spikes": [[260, 268], "..."],
+      "pose_teleports_L": [],
+      "pose_teleports_R": [],
+      "total_bad_frames": 19,
+      "bad_fraction": 0.0019
+    }
+  }
+}
+```
+A per-mode aggregate is also available: [`figures/dataset_figures/data_quality_breakdown.json`](../figures/dataset_figures/data_quality_breakdown.json).
+## Use in a dataloader
+```python
+import json
+with open("bad_frames.json") as f:
+    bad = json.load(f)["episodes"]
+def is_clean_window(ep_name, t_start, t_end):
+    """Return True if [t_start, t_end] does not overlap any flagged interval."""
+    intervals = (bad[ep_name]["intensity_spikes"]
+                 + bad[ep_name]["pose_teleports_L"]
+                 + bad[ep_name]["pose_teleports_R"])
+    for s, e in intervals:
+        if s <= t_end and e >= t_start:
+            return False
+    return True
+# In your sampler: if not is_clean_window(...), resample.
+```
+## Full report (per-episode CSV)
+Every episode's individual stats (frames, duration, contact %, max intensity left/right, drift, max pose velocity, etc.) are in [`figures/dataset_figures/data_quality_report.csv`](../figures/dataset_figures/data_quality_report.csv).

docs/recording.md ADDED Viewed

	@@ -0,0 +1,42 @@

+# Recording setup
+This document describes the hardware and sensor configuration used to record the React dataset. Schema for the resulting `.pt` files is in [`schema.md`](schema.md).
+## Sensors
+| Stream | Hardware | Native shape | Rate |
+|---|---|---|---|
+| 3 × RealSense color | Intel D415 (serials `143322063538`, `104122062574`, `217222066989`) | 480×640×3 uint8 (BGR) | 30 FPS |
+| 3 × RealSense depth | same | 480×640 uint16 (mm) | 30 FPS |
+| 2 × GelSight tactile | GelSight Mini (left / right) | 480×640×3 uint8 | ~25 FPS, resampled to camera ticks |
+| 3 × OptiTrack rigid bodies | `motherboard`, `sensor_left`, `sensor_right` | 7-vector (x, y, z, qx, qy, qz, qw) | ~120 Hz |
+Each RealSense camera's calibration is stored in the source repo under `twm/calibration/result/T_mocap_to_cam_{middle,left,right}.json`. The serial→position mapping:
+| Camera | Serial | Calibration file | Role |
+|---|---|---|---|
+| `cam0` | `143322063538` | `T_mocap_to_cam_right.json` | right-side view |
+| `cam1` | `104122062574` | `T_mocap_to_cam_left.json`  | left-side view |
+| `cam2` | `217222066989` | `T_mocap_to_cam_middle.json`| overhead / center view |
+The OptiTrack-to-GelSight rigid transforms are stored alongside in `T_gel_to_rigid_{left,right}.json`. These let you project the GelSight contact center into any of the three camera views; see the per-episode GIFs under `figures/episode_previews/` for examples.
+## Timing
+All streams share a common monotonic timestamp axis, recorded in the `timestamps` dataset (camera tick rate, 30 Hz). Per-tracker OptiTrack streams keep their own higher-rate timestamps under `optitrack/<body>/timestamps` in the raw HDF5 (not preserved in the processed `.pt` slice).
+The processed dataset (`processed/mode1_v1/...`) nearest-neighbor-aligns the OptiTrack pose stream to camera ticks. The full ~120 Hz pose track is only in the raw HDF5 mirror.
+## Coordinate frames
+- OptiTrack reports positions in meters relative to the mocap room origin.
+- Quaternions follow `(qx, qy, qz, qw)` ordering (scalar last).
+- The `T_mocap_to_cam_*` extrinsics map mocap-frame coordinates into camera-frame coordinates (right-handed, +Z forward).
+## Tasks recorded
+| Task | Description |
+|---|---|
+| `motherboard` | Bimanual manipulation of components on a computer motherboard (CPU placement, lever seat, screw tightening). |
+More tasks will be added as collection continues; see [`../tasks.json`](../tasks.json) for the machine-readable registry.

docs/schema.md ADDED Viewed

	@@ -0,0 +1,42 @@

+# Schema
+Layout of each per-episode `.pt` file in the published `processed/mode1_v1/` slice. Hardware/recording details are in [`recording.md`](recording.md).
+## `.pt` fields
+Each `episode_*.pt` is a Python dict loadable with `torch.load(..., weights_only=False)`.
+| Key | Shape | dtype | Description |
+|---|---|---|---|
+| `view` | `(T, 3, 128, 128)` | uint8 | Overhead camera (`realsense/cam0/color`), center-cropped to square then bilinear-resized to 128×128 |
+| `tactile_left` | `(T, 3, 128, 128)` | uint8 | Left GelSight, same crop/resize |
+| `tactile_right` | `(T, 3, 128, 128)` | uint8 | Right GelSight, same crop/resize |
+| `timestamps` | `(T,)` | float64 | Camera timestamps (seconds, monotonic clock) |
+| `sensor_left_pose` | `(T, 7)` | float32 | Left GelSight rigid body OptiTrack pose, nearest-neighbor aligned to camera timestamps |
+| `sensor_right_pose` | `(T, 7)` | float32 | Right GelSight rigid body OptiTrack pose, same alignment |
+| `tactile_{left,right}_intensity` | `(T,)` | float32 | Per-frame mean per-pixel L2 distance from a contact-free reference frame |
+| `tactile_{left,right}_area` | `(T,)` | float32 | Per-frame fraction of pixels with L2 diff > `tau` |
+| `tactile_{left,right}_mixed` | `(T,)` | float32 | Mean of (diff × mask) — intensity restricted to contact pixels |
+| `_contact_meta` | dict | — | Per-episode contact metadata: `tau`, drift, p01 reference indices, the chosen reference RGB frames |
+Each `.contact.json` next to the `.pt` is a small summary of the metric distributions plus drift diagnostics — useful for filtering / sanity checking without loading the full tensors.
+## Contact metric definition
+For each tactile sensor independently:
+1. Pick a **contact-free reference frame** — the ~0.1th-percentile-quietest frame by mean L2 distance to the temporal median (`reference_strategy = "p01"`).
+2. For each frame `t`, compute the per-pixel `diff[t, x, y] = || frame[t, :, x, y] − ref[:, x, y] ||_2` (RGB L2 over channels).
+3. Then:
+   - `intensity[t] = mean(diff[t])`
+   - `area[t]      = mean(diff[t] > tau)`        (default `tau = 8.0` on the uint8 scale)
+   - `mixed[t]     = mean(diff[t] · (diff[t] > tau))`
+`_contact_meta["drift_warning"]` is `True` when either sensor's drift (L2 distance between the first frame and the p01 reference) exceeds `2 · tau`; in this release no episode triggers it.
+## Pose conventions
+- Position is in **meters** in OptiTrack room coordinates.
+- Quaternion follows `(qx, qy, qz, qw)` ordering (scalar last).
+- To project sensor positions into a camera image, compose `sensor_pose → T_mocap_to_cam → camera intrinsics`. The recording-side calibration files (`T_mocap_to_cam_*.json` and `T_gel_to_rigid_*.json`) are referenced in [`recording.md`](recording.md).