A physics-based optimization approach for path planning on rough terrains

The following paper addresses the problem of applying existing path planning methods targeting rough terrains. Most path planning methods for mobile robots divide the environment in two areas-free and occupied-and restrict the path to lie within the free space. The presented solution addresses the problem of path planning on rough terrains, where the local shape of the environment are used to both constrain and optimize the resulting path. Finding both the feasibility and the cost of the robot crossing the terrain at a given point is cast as an optimization problem. Intuitively, this problem models dropping the robot at a given location (x,y) and determining the minimal potential energy pose (attitude angles and the distance of the centre of mass to the ground). We then applied two path planning methods for computing a feasible path to a given goal: Fast Marching Method (FMM) and Rapidly exploring Random Tree (RRT). Processing the whole mapped area, determining the cost of every cell in the map, we apply a FMM in order to obtain a potential field free of local minima. This field can then be used to either pre-compute a complete trajectory to the goal point or to control, in real time, the locomotion of the robot. Solving the previously stated problem using RRT we need not to process the entire area, but only the coordinates of the nodes generated. This last approach does not require as much computational power or time as the FMM but the resulting path might not be optimal. In the end, the results obtained from the FMM may be used in controlling the vehicle and show optimal paths. The output from the RRT method is a feasible path to the goal position. Finally, we validate the proposed approach on four example environments.