Terrain-adaptive Generation of Optimal Continuous Trajectories for Mobile Robots

The problem of generating continuous trajectories for motion over general 3D terrain is important. The more naïve and common approach of compensating, in the execution phase, via feedback control, for an incorrect flat terrain assumption, is not always viable. The flat terrain assumption is also almost never necessary since the terrain shape must already be known for autonomous vehicles to operate competently in 3D terrain. We propose a fairly general constrained optimization approach to trajectory generation over arbitrary terrain, for arbitrary vehicles, which optimizes arbitrary utility/cost functionals while satisfying arbitrary constraints. The approach achieves its generality, in part, by numerically linearizing and inverting forward kinematic and dynamic models of propulsion, suspension, and motion prediction. It achieves efficiency by adopting a parametric optimal control approach from earlier related work. An implementation of this algorithm is exhibited using a model based on the Rocky 7 Mars rover platform. Several utility functions minimizing time and/or slope dwell are illustrated, while demonstrating convergence in a variety of terrain shapes.