RLBench 18 Tasks: Robot Learning Benchmark for Simulated Manipulation

A mirrored dataset of 18 robot manipulation tasks originally hosted by the PerAct project. The data is organized into train, validation, and test splits. It was uploaded to Hugging Face by hqfang on 2025-02-07 to provide easier access than the original Google Drive source.

Use Cases

• Training robot manipulation policies based on the 18 defined tasks.
• Benchmarking reinforcement learning algorithms on a standardized set of simulated tasks.
• Developing visual perception models for robotic control based on the likely visual observations.
• Studying task generalization in robotics based on the multi-task structure.

Strengths

• Provides a structured benchmark with 18 distinct tasks for robot learning.
• Dataset is organized into three standard splits (train, validation, test).
• Mirrored on Hugging Face for reliable access, addressing issues with the original Google Drive source.

Limitations

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

Provenance

Source

Originally hosted by the PerAct project on Google Drive.

Collection Method

Mirrored from the original source for easier distribution.

Freshness

Last updated 2025-02-07 12:29:21; freshness should be verified.