Reliability-based complexity in intelligent machines

This paper introduces a novel methodology, reliability-based complexity (RBC), that uses system models and a priori statistics in order to regulate the sensor measurements and algorithms (static and dynamic information) utilized by intelligent machines during the execution of some task. The objective is to produce tasks with the greatest level of performance and the least amount of cost. Task performance is evaluated through the development of reliability estimates that measure the individual types of uncertainties present in the task. These reliability estimates, in conjunction with cost measures, are incorporated into the mathematical framework developed in RBC. Within this framework, the optimal information selected for each task ensures that the task will be executed with maximum reliability and minimum cost. A case study involving robotic assembly is presented in order to illustrate these results.