Fuzzy reinforcement compliance control for robotic assembly

Abstract

Compliance inherently involves modification of the robot trajectory based on the contact forces occurring during the motion and enables the robot to perform a variety of manipulation tasks which require fine motion skills. Learning of active compliance behavior can endow a robot with some form of autonomous intelligence which can be very useful for the control of manipulators working in a partially known environment and for manufacturing automation. This paper reports on the acquisition of robot fine motion skills by means of learning a compliance control strategy using fuzzy reinforcement learning. The fuzzy reinforcement compliance controller is applied to a typical robotic assembly task and its performance is compared with other learning controllers.