Sensible Planning for Vehicles Operating Over Difficult Unstructured Terrains

Emerging autonomous vehicle applications require vehicles that can operate capably and safely over difficult unstructured terrains, including the harsh environments found in mining, rescue and extraterrestrial exploration. A planning system has been developed for vehicles operating in 'difficult' unstructured terrains: terrains that in many cases require areas to be negotiated rather than avoided. Under such circumstances a planner cannot simply plan over smooth sections between obstacles. Aspects of vehicle physics such as shape and chassis articulation, dynamics, available power and terrain surface properties must be factored into the planning process. Three local planning algorithms are compared: a terrain-based algorithm commonly used in unstructured environments, an algorithm that incorporates various aspects of vehicle dynamics, and a third algorithm that extends the second by manipulating future control input selection. Due to the computationally expensive nature of such algorithms, planning horizons must be kept small and this can lead to the planners becoming trapped in local minima. The second result compares planning horizon strategies in this problem context. Long and short planning horizons are compared with a combined long and short planning strategy to quantify the extent to which the advantages of both approaches can be secured by one hierarchical planning algorithm.