WM4VLA

OSCAR companion data: robot processed releases. Paper: TBA. Authors: anonymous (under review).

Per-subset license

Subset	Upstream license	Notes
agibot	CC-BY-NC-SA-4.0	⚠️ NC + ShareAlike + upstream gated
airoa_moma	CC-BY-NC-SA-4.0	⚠️ NC + ShareAlike
droid	CC-BY-4.0
interndata	CC-BY-NC-SA-4.0	⚠️ NC + ShareAlike
rh20t_cfg5	CC-BY-NC-4.0	⚠️ NC
rh20t_cfg7	CC-BY-NC-4.0	⚠️ NC

What's inside

Total episodes: 132154

• agibot: 93981 episodes — AgiBot World Beta
• airoa_moma: 10612 episodes — AIRoA MoMa (Mobile Manipulation)
• droid: 23887 episodes — DROID: Large-Scale In-the-Wild Robot Manipulation
• interndata: 2321 episodes — InternData-A1 (InternRobotics)
• rh20t_cfg5: 796 episodes — RH20T cfg5 (Franka Panda)
• rh20t_cfg7: 557 episodes — RH20T cfg7 (KUKA iiwa)

Why no rgb.mp4?

Upstream RGB videos are subject to each upstream dataset's license (see the per-subset table above). To respect each upstream's distribution terms, we redistribute only our derived artifacts: the rendered skeleton overlay, the captions we generated, and the numeric episode metadata.

To recover RGB frames, see Mapping back to upstream.

Dataset structure

<root>/
├── agibot/
│   └── ...episode dirs, each containing:
│       ├── caption.pickle            # OSCAR-generated
│       ├── episode_meta.npz          # numeric meta (see schema below)
│       └── gripper_scenario.mp4            # OSCAR-rendered skeleton overlay
├── airoa_moma/
│   └── ...episode dirs, each containing:
│       ├── caption.pickle            # OSCAR-generated
│       ├── episode_meta.npz          # numeric meta (see schema below)
│       └── gripper_scenario.mp4            # OSCAR-rendered skeleton overlay
├── droid/
│   └── ...episode dirs, each containing:
│       ├── caption.pickle            # OSCAR-generated
│       ├── episode_meta.npz          # numeric meta (see schema below)
│       └── gripper_scenario.mp4            # OSCAR-rendered skeleton overlay
├── interndata/
│   └── ...episode dirs, each containing:
│       ├── caption.pickle            # OSCAR-generated
│       ├── episode_meta.npz          # numeric meta (see schema below)
│       └── gripper_scenario.mp4            # OSCAR-rendered skeleton overlay
├── rh20t_cfg5/
│   └── ...episode dirs, each containing:
│       ├── caption.pickle            # OSCAR-generated
│       ├── episode_meta.npz          # numeric meta (see schema below)
│       └── gripper_scenario.mp4            # OSCAR-rendered skeleton overlay
├── rh20t_cfg7/
│   └── ...episode dirs, each containing:
│       ├── caption.pickle            # OSCAR-generated
│       ├── episode_meta.npz          # numeric meta (see schema below)
│       └── gripper_scenario.mp4            # OSCAR-rendered skeleton overlay

Per-file schema

caption.pickle

Pickle dict with keys: text (str), model (str, our caption model id), version (str).

episode_meta.npz

Field	Shape	Dtype	Origin	Semantics
joint_angles	(T, n_dof)	float64	upstream passthrough	per-frame joint angles (rad)
ee_pose	(T, 4, 4)	float64	upstream passthrough	end-effector SE(3) — for DROID, reconstructed from upstream cartesian_position [x,y,z,rx,ry,rz]
gripper_openness	(T,)	float64	upstream passthrough	normalized open/close signal; sign convention per subset README
camera_intrinsic	(3, 3) or (T, 3, 3)	float64	upstream passthrough	pinhole K; per-frame for interndata, static otherwise
camera_extrinsic	(4, 4) or (T, 4, 4)	float64	upstream passthrough	world→camera SE(3)
camera_is_per_frame	()	bool	OSCAR bookkeeping	True if extrinsic is time-varying

gripper_scenario.mp4

H.264 (libx264), frame-count matches the (deleted) rgb.mp4. Contains the OSCAR-rendered skeleton overlay with gripper. Per-link / per-finger color conventions follow OSCAR's internal skeleton_def.py.

Filters applied

robot filters: arm visibility >= 50%, motion threshold >= 0.1 rad

Mapping back to upstream

Episodes here preserve the upstream's relative paths and episode identifiers (with one caveat per subset).

1. Identify the subset (top-level directory).
2. Match the relative path against the upstream release.
   • See each subset README for the exact naming convention and the upstream download link.
   • AgiBot caveat: episode names encode (start, end) × 2 of the original video; multiply both endpoints by 2 to recover raw frame indices.
3. Download the upstream raw to fetch RGB and align with our skeleton render.

We plan to release a helper script that automates upstream-raw download + RGB alignment in a follow-up release. Until then, please follow the per-subset instructions linked below.

License

Each subset inherits its upstream license. See the subset README for canonical license text. The agibot subset specifically is CC-BY-NC-SA-4.0, meaning any derivative you build must also be CC-BY-NC-SA-4.0 and must not be used commercially.