Humanoid robot navigation and obstacle avoidance in unknown environments

Abstract

In this paper, the problem of humanoid robot navigation in an unknown environment is considered. A path planning and obstacle avoidance system based on the GODZILA algorithm is proposed. The proposed approach is computationally very light-weight and does not require building of an environment map. The algorithm follows a purely local approach based on measurements from a pair of ultrasonic sensors (sonars) mounted on the robot. A primary challenge in the implementation of the path planning and obstacle avoidance system is the limited spatial information that is available due to the wide beam angle of the sensors. However, it is seen in simulation and experimental studies in this paper that a light-weight path planning and obstacle avoidance can be implemented with these ultrasonic sensors. Also, it is shown that the spatial uncertainty inherent in these sensors can be addressed through introduction of virtual sensors based on the beam angle; this concept of virtual sensors fits nicely within the GODZILA framework that is formulated based on range measurements from an arbitrary set of sensing directions, which in this implementation, is comprised of actual sensor directions and the virtual sensor directions. The performance of the proposed algorithm is demonstrated through simulations and experimental studies utilizing a NAO humanoid robot.