Learning the terrain and planning a collision-free trajectory for indoor post-disaster environments

Mobile robots dedicated in post-disaster missions should be capable of moving arbitrarily in unknown cluttered environments so as to accomplish their assigned security task. The paper in hand describes such an agent equipped with collision risk assessment capabilities, while it is able to trace an obstacle-free path in the scene as well. The robot exploits machine learning techniques for the traversability evaluation of the environment by making use of geometrical features, which derive from a postprocessing step of the depth map, obtained by an RGBD sensor. Then, the traversable scenes, are assessed by the likelihood the robot to collide on any arbitrary direction in front of it. Besides, the collision risk likelihood is combined with a path tracing algorithm based on Cellular Automata so that an obstacle-free route is then detected. The proposed method has been examined for several indoor scenarios revealing remarkable efficiency.