A Concurrent Learning Approach to Monocular, Vision-Based Regulation of Leader/Follower Systems

Abstract

This paper describes a concurrent learning approach to relative range estimation by a single vehicle following a lead vehicle and using only passive, monocular vision for feedback. The standard extended Kalman filter approach is augmented by a concurrently executed parametric estimator, which modifies the execution of the filter. The primary difficulty with this monocular vision range estimation and regulation scenario arises from the loss of observability during target acceleration, which is presumed unknown, as well as the need for persistent excitation (PE). The concurrent learning inspired approach relaxes the PE constraint by learning the target size during feedback-induced moments of PE, then using it to provide a range pseudo-measurement. Simulated scenarios demonstrate improved range regulation compared to existing methods while minimizing the a priori knowledge needed.