Vision-based Safe Local Motion on a Humanoid Robot

Humanoid soccer robots are increasingly becoming more autonomous as sophisticated approaches are being developed for challenges in vision, motion and team coordination. The stated goal of the RoboCup initiative is to beat the human soccer champion team by 2050 [1]. In order to achieve this goal, it is essential to enable the robot to fully utilize the information extracted from the available sensors. In the standard platform league of RoboCup [2], one major challenge is the ability to detect and avoid the mobile obstacles i.e. the other robots on the field. This paper presents an image gradient-based scheme to efficiently and reliably characterize the obstacles in the environment. In addition, information extracted from color images and range sensors is incorporated to build a robust obstacle model. Furthermore, a potential fieldbased method is used to navigate safely in the presence of obstacles. All algorithms are implemented and tested on the Aldebaran Nao [3] robot platform.