Abstract
This paper deals with new method of generating an optimum approach velocity of a manipulator undergoing collision. A contact motion with collision phenomena has strong nonlinearity and high speed phenomena. Furthermore, there are many unknown parameters, e.g. environmental parameters, tolerance of measurement. It is difficult to control a contact motion of a robotic manipulator with linear control theory. In this paper, we will present a new method of generating an optimum approach velocity of a manipulator. Our approach is based on the reinforcement learning method, because it is not necessary to know all valuses of parameters. After repeating trials the robot can successfully generate approach velocity to modify its nominal velocity. Simulation results show the effectiveness of this method.
