Integrated L1 adaptive cartesian PD control for robotic loco-manipulation

Abstract

This paper presents an integrated framework of L1 adaptive control and Cartesian PD control that overcomes the inherent limitations of the latter controller in tracking end effector position for loco-manipulation tasks. In a simplified system, this work represented the loco-manipulation problem with the Cartesian space dynamics of the end effector object, capturing the dominating parts of the dynamics in linear terms. The unmodeled nonlinear dynamics, along with other model uncertainties and external disturbances, are accounted for by the L1 adaptive controller, taking the Cartesian PD input as a component of the control law to resemble its behavior for the desired system dynamics modeled in the state predictor of the L1 adaptive controller. Through simulation results of various combinations of model perturbations and external disturbances, we show that this integrated framework can achieve similar tracking performances of the end effector and object without the need to re-tune controller parameters. This work demonstrated the mitigation of effects from system uncertainties with model perturbation of up to 100% higher than expected mass and up to 15N of sinusoidal external disturbances, with higher disturbance rejection possible through designing for more conservative estimates of unknown model parameters. Combined with the low computational cost of this framework, the improvement in state command tracking gives rise to potential implementation on robotic systems controlled with linear optimization formulations.