An empirical approach to path planning in unstructured outdoor environments

In the outdoor ground contact environment, the surface over which planning takes place is not flat and the cost of traversing a region of space becomes non-isotropic: the cost of traversing it from east to west can be very different from the cost of traversing it from south to north. Scoring one potential path through the terrain over another involves developing a cost function that describes the cost associated with traversing a portion of the terrain along a specific curve and then integrating this function along potential paths to identify the path with minimum cost. Rather than assuming a specific form of the traversability cost function this work (i) develops a methodology for empirically measuring the motion costs associated with traversability, and (ii) builds an empirically validated model of local motion costs that can be integrated within standard field robotics path planning algorithms. A computer controlled variable tilt platform coupled with simultaneous measurements of the power consumption of the motion actuators of the robot, is used to construct the traversability cost model. This is augmented with empirically determined angles of slip and vehicle overturn to arrive at a concise and comprehensive encapsulation of vehicle behavior over the entire spectrum of terrain the vehicle can be expected to navigate. The empirical model is evaluated in simulation where it is compared against an existing cost function. The paths computed using the empirical model are decidedly different for situations where perturbations in the terrain plane are small (less than 10 degrees roll or pitch) and show deviations in other situations as well. Overall, costs associated with constructing an empirical terrain cost function and the more complex form of the resulting function is more than offset by the quality of the paths generated using the approach.