Robust Appearance Based Visual Route Following for Navigation in Large-scale Outdoor Environments

Abstract

In this paper we present a navigation algorithm that enables mobile robots to retrace routes previously taught under the control of human operators in outdoor environments. Possible applications include robot couriers, autonomous vehicles, tour guides and robotic patrols. The appearance-based approach presented in the paper is provably convergent, computationally inexpensive compared with map-based approaches and requires only odometry and a monocular omnidirectional vision sensor. A sequence of reference images is recorded during the human-guided route-teaching phase. Before starting the autonomous phase, the robot needs to be positioned at the beginning of the route. During the autonomous phase, the measurement image is compared with reference images using image cross-correlation performed in the Fourier domain to recover the difference in relative orientation. Route following is achieved by compensating for this orientation difference. Over 18 km of experiments performed under varying conditions demonstrate the algorithm's robustness to lighting variations and partial occlusion. Obstacle avoidance is not included in the current system.