GM-100: Multi-Embodiment Robotics Benchmark with 100 Tasks

GM-100 contains 100 robotic tasks across three hardware platforms, featuring video demonstrations and metadata. Created by robbyant and updated in February 2026, the data is organized in the LeRobot 2.1 format for multi-embodiment benchmarking.

Use Cases

• Training imitation learning models using the video demonstrations and task metadata
• Benchmarking multi-embodiment generalization across the three robotic platforms
• Developing robotic control policies using the info.json configuration files

Strengths

• Includes 100 distinct robotic tasks
• Covers three different robotic platforms
• Uses standardized LeRobot 2.1 format

Limitations

• The specific robotic platforms beyond AgiBotG1 are not explicitly named in the provided documentation
• Geographic bias toward the US as indicated by metadata tags

Provenance

Source

robbyant

Collection Method

sensor

Freshness

Last updated February 2026

Geography

United States