FreeTacMan: Visuo-Tactile Robot Manipulation Demonstrations

FreeTacMan (FiftyOne) is a grouped video dataset built from the OpenDriveLab/FreeTacMan visuo-tactile manipulation dataset. Each group is a single demonstration trajectory, with one video per camera/sensor view and the full tool-center-point and gripper trajectory stored as per-frame fields. The dataset was uploaded by Voxel51 and was last updated on July 13, 2026.

Use Cases

• Train robot manipulation policies based on multi-view video demonstrations.
• Analyze tool-center-point and gripper trajectories for robotic arm motion planning.
• Develop models for visuo-tactile sensor fusion using synchronized video and trajectory data.
• Benchmark imitation learning algorithms on grouped demonstration trajectories.

Strengths

• Contains synchronized multi-view video data for each demonstration trajectory.
• Includes per-frame tool-center-point and gripper trajectory data, providing detailed motion information.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

OpenDriveLab/FreeTacMan visuo-tactile manipulation dataset.

Collection Method

Built into a grouped FiftyOne video dataset.

Freshness

Last updated 2026-07-13 19:47:10; freshness should be verified.