Entropy-based visual homing

Abstract

Visual homing allows a robot to navigate towards a previously visited goal relying on visual input. In this study, the robot homes by optimising the difference surface produced from the mutual image information between a snapshot image taken at the goal and the current image. This study proposes a new homing method with automatic gain tuning inspired by entropy distance. We also present experimental results using omni-directional image sets, which demonstrate not only the feasibility of the proposed idea for visual homing but also adaptive and robust homing performance for both static and dynamic environments.