Field Experiments in Rover Navigation via Model-Based Trajectory Generation and Nonholonomic Motion Planning in State Lattices

This paper presents field experiments of two novel approaches to local and regional motion planning applied to planetary rover navigation. The first approach solves the two-point boundary value problem using a model-based trajectory optimization technique that inverts an arbitrary dynamics model to generate a feasible motion plan. The second approach utilizes this result to build a special discretization of the state space that allows employing standard search algorithms for solving the motion planning problem. These approaches enable robot autonomy by considering the robot’s dynamics, efficiently searching a finely discretized state space, and allowing the reuse of previous planning computation to improve runtime. We present results from the experiments on the Rocky 8 and FIDO planetary rover prototypes in the NASA/JPL Mars Yard.