LEARNING-ADAPTIVE CONTROL OF INTELLIGENT ROBOTS BASED ON DIVIDING WORKING SPACE

Most of the adaptive control approaches of robots are now based on the linearized dynamical models of the robots. They use some identification algorithms in order to estimate the parameters of the models explicitly or implicitly and correct the control laws on line. In the paper, a learning-adaptive control method that is different from the approaches above is presented in accordance with the inherent attributes of the movement of robots themselves and the variation of their dynamics parameters.a two-levels structure in the hierarchical control system of intelligent robots is the basis of the method presented and the parameters variation of the robots is treated in the spacial domain not in the time one. The robot working space is divided several sub-spaces that contain the effect of gravity load and to every sub-space there correspond a set of controls. The robot path planed in the high level of the two-levels structure mapping onto the working space, a sequence of sub-spaces is formed into and then a correlative sequence of the optimal controls is picked out by means of a learning algorithm.