Vehicle Planning in Unstructured Environments

Abstract

A new physics-based method for autonomous vehicle guidance and planning in unstructured outdoor environments. The method guides a vehicle autonomously over dicult terrain, including areas that approach the maximum rough-terrain ability of the vehicle. Current algorithms focus on static capabilities of the vehicle, such as the largest obstacle that can be negotiated. The new planning system considers the vehicle dynamics, kinematics and geometry to optimise performance. The method utilises a detailed forward simulation of a vehicle interacting with terrain. The forward model has three main parts; Firstly vehicle dynamics are simulated including material properties and engine constraints. Second the three-dimensional shape of the vehicle and environment are modelled, giving points of interaction. Third the planning algorithm searches a space that inherently includes the vehicle characteristics relevant to negotiating unstructured terrain. Terrain properties other than geometry are associated with the various objects in the environment and are active during forward simulation. To plan a path, the forward simulation is guided ahead from the current vehicle position to explore the surrounding environment. The planner both controls and gains information from this exploration. A series of measures are collated at each time step summarising the interaction for the planner. The measures include; average and worst slope, derivative of the direction of the vehicles acceleration, energy expended and vehicle environment interpenetration. Factors such as the velocity to clear a particular section and the eect of the angle of approach to an obstacle are inherently taken into account by the planning system. The new method has been evaluated in field trials of a small o-road vehicle, negotiating unprepared o-road terrain.