Terrain Aware Inversion of Predictive Models for High Performance UGVs

Abstract

The capacity to predict motion adequately over the time scale of a few seconds is fundamental to autonomous mobility. Model predictive optimal control is a general formalism within which most historical approaches can be cast as special cases. Applications continue to grow in ambition to seek higher precision of motion and/or higher vehicle speeds. Predictions must therefore improve in fidelity and speed simultaneously. We favor an approach to precision motion control that we call parametric optimal control. It formulates the optimal control problem as one of nonlinear programming - optimizing over a space of parameterized controls encoding all feasible motions. It enables efficient inversion of the solutions to the equations of motion for a ground vehicle. Such an inversion enables a computation of precisely the command signals necessary to drive the vehicle to goal position, heading, and curvature while following the contours of the terrain under arbitrary wheel terrain interactions. Dynamics inversion is so fundamental that many other mobility behaviors can be constructed from it. Fielded applications include pallet acquisition controls for factory AGVs, high speed adaptive path following for military UGVs, compensation for wheel slip on the Mars Exploration Rovers and full configuration space planning in dense obstacle fields.