RoboMIND introduces a comprehensive dataset of 107,000 demonstration trajectories across 479 diverse manipulation tasks, enabling more adaptable robotic systems.

• Covers 96 object classes with multi-view observations and proprioceptive data
• Supports multiple robot platforms including Franka, UR5e, and humanoid robots
• Enables consistent and reliable imitation learning for robot manipulation
• Creates a foundation for cross-embodiment skill transfer in industrial applications

This research significantly advances robot manipulation capabilities for engineering and manufacturing environments, providing a standardized benchmark for developing more versatile and capable robotic systems.

RoboMIND: Benchmark on Multi-embodiment Intelligence Normative Data for Robot Manipulation