Bayesian Theory of Decision Making , Implemented on Simulated and Real Robots

It has been shown that human problem solving obeys Bayesian theory, where this theory makes strong assumptions about prior knowledge. Distributed adaptive control (DAC) is a modeling series approaching the biological processes of acquisition, retention and expression of knowledge. In this project the question is investigated, whether the mechanisms of DAC to express its knowledge of the world follow Bayesian inference principles. A Bayesian analysis of the behavioral tasks of nding targets and avoiding collisions is presented and a new version of the learning model, DAC5, is de ned. It is shown that DAC5 is making exactly the decisions, that are optimal according to Bayesian theory. By implementing this model on simulated and real robots detailed analysis of its di erent components has been made and a high degree of performance could be observed.