Multi-Contact Feedback MPC

Real time model predictive control that responds to multiple contacts at unknown locations.

← Back to Control & Safety

Real time model predictive control that responds to multiple contacts at unknown locations.

Overview

This project introduces an explicit contact feedback loop into model predictive control for interactive robotic tasks. A multi-contact particle filter estimates contact locations and forces in real time, and a spring contact model incorporates that information into the MPC formulation.

Using differential dynamic programming, the controller operates a seven degree of freedom robot without simplifying its dynamics. The resulting update rates allow the robot to respond to unexpected contacts during real world tasks.

Videos

Publication

S. W. Han, M. Iskandar, J. Lee, and M. J. Kim, “Online Multi-Contact Feedback Model Predictive Control for Interactive Robotic Tasks,” IEEE ICRA, 2024. arXiv · IEEE