Performance Optimization of Human Control Strategy
Performance evaluation is only a part of the solution for effectively applying models of human control strategy. Once the skills of a human control strategy (HCS) and corresponding models have been evaluated, we would like to optimize performance of the HCS models based on specific criteria. This chapter proposes an iterative optimization algorithm, based on simultaneous perturbed stochastic approximation for improving the performance of learned HCS models. The initial HCS model serves as a good starting point for the algorithm, since it already generates stable control commands. Dick’s optimized HCS model not only improves tight turning performance, but obstacle-avoidance performance as well. The algorithm keeps the overall structure of the learned models intact, but tunes the parameters in the model to achieve better performance. While performance improvements vary between HCS models, the optimization algorithm always settles to stable, improved performance after only a few iterations.