libero-r6-pointmap

Depth videos, per-frame camera parameters and robot-base poses for the four LIBERO suites, aligned frame-for-frame with DAVIAN-Robotics/libero-r6. Together the two datasets are the fine-tuning data for the PointmapVLA LIBERO and LIBERO-Plus results. PointmapVLA does not store pointmaps: its data loader back-projects these depth maps with the camera parameters and expresses the result in the robot-base frame at training time.

• 1,693 episodes over 40 tasks, 20 fps, 256×256
• 2 cameras: agentview and wrist
• Rendered by replaying the original LIBERO demonstrations in the simulator (same procedure that produced the RGB), recording depth and camera parameters every frame

Format

Each suite is its own LeRobot v3.0 dataset directory (libero_<suite>_no_noops_1.0.0_lerobot/). RGB videos are not included here; use libero-r6 for RGB, state and actions. Episodes match by episode_index.

Feature	Shape / type	Meaning
depth.<camera>	video, FFV1 gray16le in .mkv	uint16 = depth[m] × 10000; 0 = invalid. Valid range used by PointmapVLA: 0.05 m < z < 6.5 m. Decode with PyAV in gray16le.
<camera>_camera_intrinsics	float32 (4,)	fx, fy, cx, cy in pixels. fx is negative: LIBERO images are stored rotated by 180°, and the sign encodes that mirror so the same back-projection code works unchanged.
<camera>_camera_extrinsics	float32 (7,)	camera→world pose: position (3) + unit quaternion w, x, y, z (4), OpenCV camera axes, per frame
state.base_position	float32 (3,)	robot base position in the world frame
state.base_rotation	float32 (4,)	robot base quaternion x, y, z, w
observation.state, action	(11,), (10,)	same as libero-r6

To obtain a robot-base-frame pointmap: back-project each valid pixel with the intrinsics, transform by the camera→world extrinsics, then by the inverse of the base pose. The reference implementation is starVLA/dataloader/gr00t_lerobot/transform/pointmap_math.py in the PointmapVLA repository.

Verification

MANIFEST.json lists the size and SHA-256 of every file; check a download with python scripts/verify_data.py <local_dir> from the PointmapVLA repository.

License and attribution

Derived from LIBERO (MIT License). The depth renders and camera annotations are released under the MIT License.