Mapless Urban Robot Navigation by Following Pedestrians

Abstract

Navigating effectively and safely in unknown urban environments is a crucial ability for service robot applications such as last-mile package delivery. To reach the entrance of its target destination, the robot must make informed local and global path planning decisions. We present a mapless global planning strategy based on pedestrian following. Our method allows the robot to exploit natural routes taken by surrounding pedestrians to make informed and efficient path planning decisions for reaching its goal. The algorithm also includes a recovery system to assist the robot when insufficient progress is made (i.e. robot stuck in dead end). Once the robot is within the vicinity of the target building, a wall following behaviour is used to reach the entrance of the target building. Simulated experiments and a proof-of-concept demonstration on a real robot were shown to validate the approach.