Walking in unknown environments — A step towards more autonomy

We present a new approach to autonomous navigation for humanoid robots in previously unknown environments. Our method uses only on-board cameras and does not rely on color coding, textured surfaces or predefined object models. We propose using a reactive approach towards navigation that does not require building an explicit model of the environment. The basic idea of the navigation system is to check a set of precalculated feasible trajectories (called tentacles) for passability instead of searching for a path within a map. The computer vision system was combined with our real-time trajectory planning and control, enabling our robot LOLA to safely navigate in previously unknown environments.