FreeTacman: 3 Million Visuo-Tactile Image Pairs for Robotics

OpenDriveLab released FreeTacman in 2026, providing 3 million visuo-tactile image pairs and 10,000 trajectories. The data covers 50 manipulation tasks collected using a robot-free system designed for tactile-intensive interaction research.

Use Cases

• Training visuo-tactile fusion models using the 3 million image pairs
• Developing trajectory prediction algorithms across 50 manipulation tasks
• Learning manipulation policies using tactile feedback for contact-based interactions

Strengths

• 3 million visuo-tactile image pairs
• 10,000+ trajectories
• 50 distinct manipulation tasks

Limitations

• Robot-free collection method may require domain adaptation for specific robotic hardware dynamics
• Column-level schema and specific file formats are not detailed in the initial release summary

Provenance

Source

OpenDriveLab, ICRA 2026 publication

Collection Method

Robot-free visuo-tactile data collection system

Freshness

Last updated February 2026.

Geography

United States