Entropy searches for robotic reliability assessment

A technique for selecting a reliable algorithm for accomplishing some robotic task from m possible algorithms containing zero mean Gaussian errors is developed. The method is especially suitable for determining sensor placement and sensing parameters because alternate sensing strategies can be analytically compared. The reliability algorithm will analyze any robotic plan whose performance in noise can be measured by the ability to meet a quadratic constraint on a zero mean Gaussian random vector. The joint entropy of that random vector is used to direct the search for a reliable algorithm. Very simple identities are found for calculating the entropy of position measurements made by parallel stereo vision systems, and the entropy is found to depend solely on the distance of the measured point from the image planes (or on the disparity)