Quadratic program based nonlinear embedded control of series elastic actuators

Abstract

This paper presents a method for embedded motor control based upon rapidly exponentially stabilizing control Lyapunov functions (RES-CLFs) implemented through Quadratic Programs (QPs). This will give guaranteed exponential convergence via an optimal nonsmooth nonlinear embedded level controller that provides the minimal control effort necessary to achieve the desired convergence in torque. Utilizing this novel control methodology, we are able to formally establish that the dynamics of series elastic systems can be approximated by rigid system models. Importantly, the RES-CLF based QP is presented in a way that will allow for its real-time implementation at the embedded level via a closed form solution to a QP; the end result is a nonlinear optimal controller able to run at over 5 kHz. To demonstrate this, simulation and experimental results are presented showing the performance of the embedded controller.