Robot joint friction compensation learning enhanced by 6D virtual sensor

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