An exploration-based approach to terrain traversability assessment for a walking robot

In this paper we present a traversability assessment method for motion planning in autonomous walking robots. The aim is to plan the motion of the robot in a real scenario on a rough terrain, where the level of details in the obtained terrain maps is not sufficient for motion planning. A guided RRT (Rapidly-exploring Random Trees) algorithm is used to plan the motion of the robot on rough terrain. We are looking for a method that can learn the terrain traversability cost function to the benefit of the guiding function of the planning algorithm. A probabilistic regression technique is used to solve the traversability assessment problem. Computing the predictions of the traversability values we use the RRT planner to explore the space of possible solutions. We demonstrate efficiency of the prediction method and we show results of experiments on the real walking robot.