Dexterous Hand-Object Interaction Simulation Data from RoboCasa

Featuring RoboCasa simulation data for training and evaluating the DexWM world model. It includes two data regimes: RoboCasa Random with exploratory_movement and gripper_open_and_close sequences, and another unspecified regime.

Use Cases

• Train world models on exploratory_movement sequences to learn dexterous hand dynamics.
• Evaluate gripper control policies using gripper_open_and_close interaction sequences.
• Benchmark reinforcement learning agents in simulated dexterous manipulation tasks.

Strengths

• Data is specifically designed for training and evaluating the DexWM world model.
• Includes distinct data regimes for different training and evaluation phases.
• Associated with a published research paper and project.

Limitations

• Data is purely from simulation (RoboCasa), lacking real-world validation.
• Specific details on data volume, file formats, and column structure are unavailable.
• The scope of objects and interaction scenarios in the simulation is unknown.

Provenance

Source

facebook

Collection Method

Generated from the RoboCasa simulation environment.

Freshness

Last updated 2026-03-20.