Probabilistic algorithm and training rule for a new identification and control kernel application to robotics systems

Abstract

A stochastic program is developed for adaptive control and identification of industrial design applications. Our program is executed at two levels: a stochastic trajectory planner and an on-line trajectory follower based on the complete stochastic dynamic model of the process. The modeling is first done in the deterministic case based on the Lagrangian formalism. This gives the stochastic model of the process. This study is applied to a case study of mobile robots agents. The mobility of the robot is also considered; first static mobility is given. Then we consider dynamic mobility. After that the mobility is randomized and taken as an output of our dynamic system. Our program is one of identification of the doubly stochastic process of hidden Markov chains minimizing the function of information of Kullback-Leibler convergence and the consistence of functions of parameters evaluation. Simulations for the case of the SARAH robot are given to demonstrate the efficiency of our algorithms.