RLBench V3: Robot End-Effector Demonstrations for 18 Tasks

RLBench V3 EEF Test contains robot demonstration data in the PerAct format for 18 manipulation tasks. The data includes dense end-effector quaternion poses (xyzw) and is provided as a test split. It was created by GT-111 and uses data from the PerAct paper and the RLBench simulator.

Use Cases

• Benchmarking imitation learning models based on the 18 RLBench tasks.
• Training robot control policies based on dense end-effector pose trajectories.
• Evaluating generalization of robot manipulation models using the defined test split.
• Developing and testing robot perception-action models based on the PerAct format.

Strengths

• Data is structured for 18 distinct robot manipulation tasks.
• Uses a standardized format (LeRobot v3.0) with raw values.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Data originates from the PerAct research and the RLBench simulation environment.

Collection Method

Likely generated through simulated robot task execution.

Freshness

Last updated 2026-05-11 11:18:34; freshness should be verified.