Prejudicial searches and the pole balancer

The authors describe a modified random search method called 'prejudicial searching'. This method always finds the global minimum of any continuous function on a compact domain. The authors consider the method in the context of minimizing an objective function for the pole-balancer control problem. They begin by reviewing the derivation of cooling schedules for simulated annealing. Using similar methods they prove the convergence of the 'prejudicial search' algorithm. Preliminary simulation results indicate that the random search and simulated annealing have roughly the same efficiency. The authors expect the prejudicial search to provide substantial improvements over the random search.<>