Learning impedance parameters for robot control using an associative search network

An evaluation of the associative search network (ASN) learning scheme when used for learning control parameters for robot motion is presented. The control method used is impedance control in which the controlled variables are the dynamic relations between the motion variables of the robot manipulator´s tip and the forces exerted by the tip. The main task used is that of wiping a surface whose geometry is not precisely known. The learning scheme does not use a model of the robot and its environment. It is a stochastic scheme that uses a single scalar value as a measure of the system performance. The scheme is found to perform quite well. A few variants of the main scheme are discussed. Modifying the virtual trajectory, externally to the ASN scheme, shows an improved performance