Reframe dataset as dynamics / world-model interaction data (not policy demos); document 2026-03-23 as right-sensor-only pilot; add active_sensors per-date in tasks.json; lead with hours and contact frames instead of episode count

README.md

@@ -10,9 +10,11 @@ tags:
   - gelsight
   - realsense
   - motion-capture
+  - dynamics
+  - world-model
 pretty_name: React (Tactile-Visual Manipulation)
 size_categories:
-  - 10K<n<100K
+  - 100K<n<1M
 configs:
   - config_name: motherboard
     data_files:
@@ -26,37 +28,46 @@ configs:
 
 # React
 
-Bimanual manipulation recordings with vision-based tactile sensors and motion capture.
+Dense, contact-rich, synchronized multimodal interaction recordings for **tactile-visual dynamics / world-model learning** — *not* a policy / demonstration dataset.
 
 ![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**
+> **138 min of synchronized multimodal interaction · 88 min (66 %) of confirmed bimanual tactile contact · 4,136 distinct contact events · 240 k frames @ 30 Hz across 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 |
+| Intended use | Dynamics / world-model learning over short multimodal windows. Sample short trajectories (1 s – 10 s); episode boundaries are not action boundaries. |
+| Total synchronized duration | **138.4 min** at 30 Hz (239,759 multimodal frames) |
+| Bimanual tactile-contact time | **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 |
+| Tasks | `motherboard` (more coming) |
 | License | CC-BY-4.0 |
 
 ![Comparison with other manipulation datasets](figures/dataset_figures/F7_comparison_table.png)
 
-## Quick start
+## Recording sessions
+
+| Date | Kind | Active sensors | Notes |
+|---|---|---|---|
+| 2026-03-23 | pilot | **right only** | Early trials. The left GelSight and left rigid body were not in use; ignore `tactile_left*` and `sensor_left_pose` for these recordings. |
+| 2026-05-10 | session | left + right | First full bimanual session. |
+| 2026-05-11 | session | left + right | Largest session. A handful of GelSight LED-flicker frames + one mocap teleport; see [`bad_frames.json`](bad_frames.json). |
 
-Load by task with `datasets`:
+See [`tasks.json`](tasks.json) for the machine-readable registry (per-date `active_sensors`, etc.).
+
+## Quick start
 
 ```python
+# Load by task with `datasets`
 from datasets import load_dataset
 ds = load_dataset("yxma/React", "motherboard", split="train")
 ```
 
-Or grab a single episode:
+Or grab a single episode file directly:
 
 ```python
 import torch
@@ -75,21 +86,27 @@ ep = torch.load(path, weights_only=False)
 # 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.
+Sampling short windows for dynamics learning:
 
-See [`tasks.json`](tasks.json) for the task registry.
+```python
+import json
+with open("bad_frames.json") as f:
+    bad = json.load(f)["episodes"]
+# Drop ~0.085 % of frames flagged in bad_frames.json — see docs/quality.md
+# For 2026-03-23 episodes, also ignore the left-sensor fields (right-only pilot).
+```
 
-## Episode previews
+## Episode-file 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.
+Per-file 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. (The on-disk recording unit is called an "episode" purely for file naming — episode boundaries don't carry semantic meaning for this dataset.)
 
 ## Repository layout
 
 ```
 README.md                                        # this file
-tasks.json                                       # task registry
+tasks.json                                       # task / session registry
 bad_frames.json                                  # data-quality skip-list
-processed/mode1_v1/<task>/<date>/episode_*.pt    # per-episode tensors
+processed/mode1_v1/<task>/<date>/episode_*.pt    # per-file tensors
 figures/                                         # previews + analysis figures
 docs/                                            # extended documentation
 ```

docs/caveats.md

@@ -2,9 +2,28 @@
 
 For known sensor-level anomalies and the `bad_frames.json` skip-list, see [`quality.md`](quality.md).
 
-## Caveats
+## What this dataset is and isn't
 
-- **Missing / dropped episodes** on `motherboard/2026-05-11`:
+React is a **dense multimodal interaction stream** intended for learning **dynamics / world models** over short multimodal windows. It is **not**:
+
+- A demonstration / policy-learning dataset. "Episodes" here are just file boundaries — they don't carry semantic / action structure.
+- Comparable apples-to-apples with BridgeData V2, DROID, RT-1, ALOHA, etc. on "number of demos." The relevant comparison is *hours of synchronized multimodal contact-rich interaction*.
+
+## Per-session caveats
+
+### 2026-03-23 — right-sensor-only pilot
+
+The first 3 recording files (2026-03-23/episode_{000,001,002}) are **early pilot trials with only the right GelSight sensor in use**. For these files:
+
+- `tactile_left` is essentially flat (peak intensity ≤ 1.4 vs ≥ 12 when actively used). Treat as empty.
+- `sensor_left_pose` is not informative — the left rigid body was either not mounted or stationary off-workspace; the few "pose teleport" events flagged in `bad_frames.json` for these files are mocap noise on an unused tracker.
+- `tactile_right` and `sensor_right_pose` are valid and contribute ~7.5 min of useful single-sensor contact data.
+
+Dataloaders consuming these files should either mask the left modalities or skip the date entirely if they require bimanual data. The `active_sensors` field in [`../tasks.json`](../tasks.json) records this explicitly.
+
+## Other caveats
+
+- **Missing / dropped files** 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.**

docs/quality.md

@@ -1,35 +1,39 @@
 # 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.
+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 this dataset is intended for **short-window dynamics / world-model learning**, where a glitch landing inside a training window can dominate the loss for that step.
 
 ## Headline numbers
 
 | | |
 |---|---:|
-| Total frames | 239,759 |
-| Episodes | 30 |
+| Total synchronized frames | 239,759 (138.4 min @ 30 Hz) |
+| Recording files | 30 |
 | Frames flagged in `bad_frames.json` | **202 (0.084 %)** |
-| Episodes with ≥1 flagged frame | 10 |
+| Recording files with ≥1 flagged frame | 10 |
 
 ## Failure modes — quantitative breakdown
 
-| # | Mode | Frames | % of dataset | Episodes | Symptom | Example |
+| # | Mode | Frames | % of dataset | Files | 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) |
+| 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} (these are on the unused **left** tracker — see Notes), 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} (latter two are on the unused left tracker). | [`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`.
+**Modes 1+2 together: 202 / 239,759 = 0.084 % of frames are corrupted in any modality.** Mode 3 is tracked separately because it is healthy data, just non-informative for dynamics. `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`.
+
+### Notes on the 2026-03-23 entries
+
+`2026-03-23/episode_{001,002}` were **right-sensor-only pilot trials** (see [`caveats.md`](caveats.md)). The "pose teleport" flags on those files are on the **left** tracker, which was not in use, so the position values were essentially random mocap noise. If you are masking out the left modalities for those files (which you should), these teleport intervals can also be ignored.
 
 ## Inspection figures
 
-**Mode 1 — GelSight LED flicker** (overview across all 5 affected episodes):
+**Mode 1 — GelSight LED flicker** (overview across all 5 affected files):
 
 ![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/).
+Per-file 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):
+**Mode 2 — OptiTrack pose teleport** (2 examples for 2026-03-23, where only the processed `.pt` is on the publishing machine; the other 4 affected files 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)
@@ -76,9 +80,12 @@ A per-mode aggregate is also available: [`figures/dataset_figures/data_quality_b
 
 ```python
 import json
+from pathlib import Path
 
 with open("bad_frames.json") as f:
     bad = json.load(f)["episodes"]
+with open("tasks.json") as f:
+    sessions = json.load(f)["tasks"]["motherboard"]["per_date_notes"]
 
 def is_clean_window(ep_name, t_start, t_end):
     """Return True if [t_start, t_end] does not overlap any flagged interval."""
@@ -90,9 +97,16 @@ def is_clean_window(ep_name, t_start, t_end):
             return False
     return True
 
-# In your sampler: if not is_clean_window(...), resample.
+def active_sensors(ep_name):
+    """Returns e.g. ['right'] or ['left', 'right']."""
+    date = ep_name.split("/")[0]
+    return sessions[date]["active_sensors"]
+
+# In your sampler:
+#   if not is_clean_window(...): resample
+#   sides = active_sensors(...)   # mask out inactive tactile + pose modalities
 ```
 
-## Full report (per-episode CSV)
+## Full report (per-file 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).
+Every file'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).

tasks.json

@@ -2,8 +2,26 @@
   "tasks": {
     "motherboard": {
       "language": "Bimanual manipulation of components on a computer motherboard.",
+      "purpose": "Dense, contact-rich, synchronized multimodal interaction data for tactile-visual dynamics / world-model learning. Not a policy-learning / demonstration dataset.",
       "dates": ["2026-03-23", "2026-05-10", "2026-05-11"],
-      "n_episodes": 30,
+      "per_date_notes": {
+        "2026-03-23": {
+          "kind": "pilot",
+          "active_sensors": ["right"],
+          "note": "Early trials. ONLY the right GelSight + right rigid body were used. The left tactile stream is essentially flat (peak intensity ≤ 1.4 vs ≥ 12 when active) and the left OptiTrack pose is not informative — treat as right-sensor-only data."
+        },
+        "2026-05-10": {
+          "kind": "session",
+          "active_sensors": ["left", "right"],
+          "note": "First full bimanual recording session."
+        },
+        "2026-05-11": {
+          "kind": "session",
+          "active_sensors": ["left", "right"],
+          "note": "Largest session. A few GelSight LED-flicker frames and one mocap teleport event — see bad_frames.json."
+        }
+      },
+      "n_episode_files": 30,
       "notes": "2026-05-11/episode_001 was lost at recording time (HDF5 superblock not finalized). 2026-05-11/episodes 000 and 002 were short test recordings with no tactile contact and were dropped."
     }
   },