Stochastic Optimization Methods in Robust Adaptive Control of Robots

In the optimal control of industrial or service robots, the standard procedure is to determine first off-line a feedforward control and a reference trajectory based on some nominal values of the model parameters, and to correct then the resulting inevitable deviation of the trajectory or performance of the system from the prescribed values by on-line (local) measurement and control actions. Due to stochastic parameter variations, increasing correction actions are then needed during the process. By adaptive optimal stochastic trajectory planning and control (AOSPTC), the a priori and sample information available about the robot is incorporated into the control process by using stochastic optimization techniques. Moreover, the feedforward control and the reference trajectory are updated somewhat later in order to maintain a high quality of the reference functions. As a consequence, the deviation between the actual and prescribed trajectory or performance of the robot is reduced. Hence, the on-line correction expenses can be reduced, and more reliable, robust controls are obtained. Analytical estimates for the reduction of the online correction expenses are given.