OPTIMAL LEARNING CONTROL VIA CELL MAPPING

In this paper, a new learning method bused on the concept of cell-to-c ell mapping is developed to generate automatically a controller from e xperimental data without using a system model. The continuous state sp ace is divided into many discrete cells initially, and then the dynami cs of the system are approximated by the cell mapping. A series of lea rning signals is applied to excite the system, the experimental data a re collected in a data buffer, and are then processed into a control t able which is used to store the experienced dynamics of the system. Th e control table is optimized and updated through learning to improve t he controller performance. Some generalized controls called pseudo-exp erience production rules are used to fill some empty entries in the co ntrol table. Finally, the control table is used as a controller via a table look-up scheme for the system.