Psi0 Apple-to-Plate VR Teleoperation Dataset provides 79 human-demonstrated loco-manipulation trajectories for fine-tuning the Psi0 Vision-Language-Action model. The data was collected using a Unitree G1 29-DOF humanoid robot within the Isaac Lab simulation environment. It was published by cloudwalk-research and last updated on April 13, 2026.
Use Cases
- Fine-tuning a VLA model for humanoid robot manipulation based on the described teleoperation trajectories.
- Benchmarking robot learning algorithms for loco-manipulation tasks based on the 'pick, walk, and place' scenario.
- Studying human teleoperation strategies for bimanual robots based on data from the Unitree G1's dexterous hands.
- Developing simulation-to-reality transfer methods for robot control based on data generated in Isaac Lab.
Strengths
- Contains 79 episodes of human-demonstrated robot trajectories.
- Focuses on a specific, complex loco-manipulation task involving picking, walking, and placing.
- Uses a high-degree-of-freedom platform (Unitree G1 with 29 DOF and dexterous hands).
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Data is limited to a single, specific task performed in simulation.
Provenance
- Source
- cloudwalk-research
- Collection Method
- Human teleoperation in VR within the Isaac Lab simulation.
- Freshness
- Last updated 2026-04-13 14:39:49; freshness should be verified.