Learning hybrid position force control of a quadruped walking machine using a CMAC neural network

Learning control algorithms based on the cerebellar model articulation controller (CMAC) have been successfully applied to control non-linear robotic systems in the past. Most of these previous works are focused on the position controls of manipulators. In this article, a CMAC-based learning control method for the hybrid force/position control of a quadruped walking machine on soft terrains is presented. The relationship between the foot force and the control variables is derived for various force control methods. By using the CMAC to approximate the dynamics of one leg, we are able to demonstrate the improved control accuracy without the exact leg model. The same concept is extended to the control of a quadruped walking machine. (C) 1997 John Wiley & Sons, Inc.