USING PROBABILITY ESTIMATES TO IDENTIFY ENVIRONMENTAL FEATURES FOR A NONHOLONOMIC CONTROL-SYSTEM

A third-order set of nonlinear, ordinary differential equations models the relationship between internally measurable wheel rotations and th e position and orientation of an automatically guided vehicle, but the se relationships are imprecise, growing increasingly inadequate as the ir integrals, and the vehicle, proceed from point of departure, An ext ended Kalman filter (EKF) is used to combine video observations of fea tures on that portion of the environment that does not move, together with the sensed wheel rotations, to produce the ongoing estimates need ed for navigation, The experimental usefulness is examined of a byprod uct of the filter, the estimate error covariance matrix, to an integra lly related process: the process of identifying video observations wit h features of known location within the environment; these identities are required for application of new vision observations to the state e stimates, The goodness of the EKF's probability density functions is e xperimentally examined by comparing them against actual, accumulated d ata; experimental results are presented from the use of an extensive t heoretical development that assesses, based on relative probabilities inferred from these distributions, the identities of densely occurring , nondistinct cues.