Stereo vision based autonomous navigation for 3-DOF systems in unstructured environments

Abstract

A stereo vision based autonomous navigation method for 3-DOF systems is presented in this paper. It is able to tackle the learning and recognition problem of generic scenes in an unstructured environment, providing motion-planning capability to control all the 3 DOFs of a robotic system. In this method, 3 spatial constraints are generated from a single visual recognition to estimate the robot pose. A feedback strategy is utilised for robot motion control, without the necessity of knowing any explicit distance information of the environment. The performance of the proposed method is evaluated in a novel wire-frame simulation environment, under the perturbation of multiple uncertainty sources. Autonomous navigation is achieved with good accuracy in the simulation environment, while preserving high robustness to all the uncertainty sources.