Robot joint friction compensation learning enhanced by 6D virtual sensor

Abstract

High-performance robot control is one of the most investigated topics in both research and industry. Being able to compensate for robot dynamics is indeed one major challenge. Joint friction is commonly the main issue, especially in sensorless (i.e., no availability of torque/force sensors) compliance-controlled robots for interaction application purposes. The presented paper aims to propose a methodology for sensorless Cartesian impedance controlled robots to learn local friction compensation controllers. Exploiting a 6D virtual sensor to quantify the joint friction effects, a Bayesian optimization (BO)-based algorithm is proposed to minimize the estimated external interaction in free-motion tasks (related to friction effects). The BO algorithm enhances the impedance control performance by tuning the model-based friction compensator parameters. To validate the proposed approach, experimental tests have been executed on a Franka EMIKA panda robot, highlighting the suitability of the proposed 6D virtual sensor and BO-based algorithm to minimize the estimated external interaction for joint friction compensation purposes.