Applying Stochastic Control Theory to Robot Sensing, Teaching, and Long Term Control

Abstract

Robot applications in industry are essentially repetitious in nature and have a strong stochastic component. Random errors occur because of inaccuracies or wear in jigs, manufacturing tolerances in parts, and imperfect robot behavior. Sensors which provide feedback to the robot also are error-prone. In addition there are systematic errors due to misaligned coordinate frames between the robot and its environment. This paper shows mathematically how stochastic control theory can be used to help robots monitor themselves, check for drift, learn the correct spacing between holes or the pitch of palletized parts, and other long term matters of interest to practical intelligent robot behavior. Examples are worked out and computer simulations are presented.