Evaluation of Variable Impedance- and Hybrid Force/MotionControllers for Learning Force Tracking Skills

For robots to perform real-world force interaction tasks with human level dexterity, it is crucial to develop adaptable and compliant force controllers. Learning techniques, especially reinforcement learning, provide a platform to develop adaptable controllers for complex robotic tasks. This paper presents an evaluation of two prominent force control methods, variable impedance control and hybrid force-motion control in a robot learning framework. The controllers are evaluated on a Franka Emika Panda robotic manipulator for a robotic interaction task demanding force and motion tracking using a model-based reinforcement learning algorithm, PILCO. Utilizing the learning framework to find the optimal controller parameters has significantly improved the performance of the controllers. The implementation of the controllers integrated with the robot learning framework is available on https://github.com/martihmy/Compliant_control.