Application of Learning Automata to Controller Design in Slow-Active Automobile Suspensions

This study considers a new design methodology in the context of active vehicle suspension control. The approach combines concepts from Stochastic Optimal Control with those of Learning Automata. A learning automaton effectively learns optimal control on-line in the vehicle, in an appropriate stochastic “test-track” environment. For practical application, the overwhelming advantage of this approach is that no explicit modelling is required, and considerable time savings may be expected in system development. This simulation study considers the on-line learning of optimal control in a low-bandwidth active suspension system, where control feedback is confined to a body-mounted accelerometer at each corner of the vehicle. It is shown that learning can successfully take place under a range of conditions, including the case when there is substantial transducer noise. The performance of the resulting control system is shown to depend heavily on the nature of the learning environment.