Autonomous urban navigation and its application to patrol

In urban environment, the Global Positioning System (GPS), including differential variant of it, does not have sufficient accuracy for unmanned autonomous robot to drive along a lane because of tall buildings and trees. It is required to keep the variance of the distance between the robot and the boundary of the road less than 10-20cm to safely follow the road, as human beings drive. To comply the requirement, our autonomous navigation system detects local features of the road, e.g., curb and lines on the roadside using laser range finder and vision camera. In this paper, we develop and implement probabilistic approach for the robustness of the detection system; a model-based Bayesian method to detect the curb and line and a weighted sum-based sensor fusion method to increase the reliability. Finally, we are applying the techniques proposed to a mobile robot for unmanned patrol in urban environment.