Quadratic programming-based inverse dynamics control for legged robots with sticking and slipping frictional contacts

Abstract

Inverse dynamics control is an extremely effective nonlinear control strategy if the inverse dynamics computation can be performed with sufficient speed.We describe our method for inverse dynamics control of legged robots that can deal with both sticking and slipping frictional contacts and mitigates the problems introduced by indeterminate rigid body contact. We improve this work, which previously used quadratically constrained quadratic programs (QCQPs), to use faster-to-solve quadratic programs via linear algebraic simplifications and a nullspace. We also show that Lemke's Algorithm is significantly faster than alternatives and permits multi-stage optimizations within control loops running as fast as 125Hz on commodity hardware.