A Holistic Framework for Planning, Real-time Control and Model Learning for High-speed Ground Vehicle Navigation Over Rough 3D Terrain

Nima Keivan, Steven Lovegrove, and Gabe Sibley

IROS 2012 Workshop on Robot Motion Planning: Online, Reactive, and in Real-time, 2012

Abstract

This paper describes a local planning, control and learning framework enabling high-speed autonomous ground-vehicle traversal of rough 3D terrain replete with bumps, berms, banked-turns and even jumps. We propose an approach based on fast physical simulation and prediction, which we find offers numerous benefits: first, it takes advantage of the full expressiveness of the inherently non-linear, highly dynamic systems involved; second, it allows for the fusion of local planning and model-based feedback control all within a single framework; third, it allows vehicle model learning. The final and most important reason to use physical simulation as a unifying framework is that it works well in practice. The system is experimentally validated on a high speed nonholonomic remotely controlled vehicle on undulating terrain using a scanned 3D ground model and motion capture ground-truth data. Parameter reduction is achieved with the use of cubic curvature control primitives and a fast precomputed lookup table.