An entropy-based reliability assessment technique for intelligent machines

A new method for measuring the performance of intelligent robot systems is presented. The method utilizes entropy, a concept borrowed from informaton theory, to provide a unified technique for measuring the performance of various combinations of control and sensing algorithms available in an intelligent machine in response to a given task specification. It can be shown that the entropy of a system can be decomposed into two independent terms, i.e., a term associated with the system state description, and a term associated with the task specification. It can be shown that the total system entropy is directly analogous to the reliability of the system. A review of entropy methods in reliability analysis is presented, and the derivation of the proposed reliability assessment technique is shown. The method is demonstrated in a case study.<
>