Towards optical flow-based robotic homing

Abstract

This paper presents a novel biologically-inspired approach for tackling the problem of robot homing. In our method the only information employed is optical flow. Optical flow, which is not a property of landmarks like colour, shape, and size but a property of the camera motion, is used for localising an autonomous robot in a priori unknown environment. Our method exploits the optical flow `fingerprint' of landmarks caused by the motion of the robot in the environment. For this purpose, we have developed a training algorithm that estimates the probability of observing the same landmark from varying distances and velocities. Our method promises to be computationally efficient and inexpensive. The simulation results we present show the validity of our methods.