Linear Time-Varying MPC for Quadruped Locomotion
Published:
Summary
This project formulates the heavily used MIT cheetah convex model predictive controller (MPC) for quadruped locomotion. We linearize across the reference trajectory and use a single-shooting method to formulate the optimal control problem under dynamic constraints as a convex model predictive controller. For more details, see our arxiv paper.
Results
Our results can be seen in our github repository. We showcase here the quadruped tracking a reference trajectory to move diagonally across the plane. The corresponding MPC optimizes the reference trajectory, giving corresponding ground reaction force (GRFs) as the control input.
