TacPAC
LogosRoboticsGroup
Tactile prediction and execution-time action correction for contact-rich robot manipulation.
What you can explore
Inspect the two-stage learning pipeline and asynchronous tactile correction mechanism.
Before you use it
Correction depends on sensor timing, action chunking and the documented robot setup. Check third-party models and data independently of the MIT code.
Original contributors
TacPAC authors and Logos Robotics Group contributors
License / access: Code: MIT · check external data, models and hardware terms separately
Keep exploring
T-Rex
Tactile-reactive dexterous manipulation that uses touch to respond during task execution.
DeCAL
A dexterous vision-language-action model combining contact-aware multimodal prediction and action generation.
ManipForce
Force-guided policy learning with frequency-aware representations for contact-rich manipulation.
Editorial reference · checked 2026-09-22