Evolutionary adaptation of control processes in robots operating in nonstationary environments

Abstract

The paper outlines a new approach to the adaptation problem is nonstationary conditions, the method of evolutionary adaptation, which is characterized by a lack of formalized a priori information about the evolution of the characteristics (parameters) under estimation. The basic philosophy of the approach is in tracking the time varying characteristics on the basis of current information. The two levels (performing and tactical) of the hierarchically organized robots control system are considered and the evolutionary adaptation technique is formulated and developed for each level. The parameters estimation algorithms show the following properties: minimum variance unbiased estimates, recurrent form, on-line time varying parameters identification (ensuring the parameters tracking) and convergence independent of the choice of the estimator initial conditions. The proposed adaptive control scheme may be applied when designing robots operating in an environment essential to the occurrence of nonstationary phase coordinates.