ROUGH-TERRAIN ROBOT MOTION PLANNING BASED ON TOPOLOGY AND TERRAIN CONSTITUTION

摘要

Abstract. We present a randomised potential-field based motion planning approach for rough-terrain mobile robot navigation. The motion planning component is closely integrated with a real-time capable rigid-body kinematics and dynamics simulation module. The terrain is approximated as a polygonal surface generated from height data. The robot interacts with the terrain, modelled as a series of physically accurate collisions between its geometry and the polygons of the terrain. This model includes physical material properties, such as coefficients of restitution, static and dynamic friction of the underground as well as the robot parts. For each polygon of the terrain, a different set of material properties can be specified to represent variations in the underground constitution. In real rough-terrain settings it is typically not possible to accurately define a binary obstacle area, therefore we have included a continuous obstacle model. Each point of the terrain has a degree of “obstacleness ” associated with it. This coefficient depends on the local terrain geometry and material properties. When navigating on the terrain, some areas of the workspace might be unreachable for the robot. Importantly, these areas need not a priori be specified as obstacles, but are a consequence of the physics-based robot-terrain interaction model. A motion plan is generated by performing a randomised search on a regular discreet grid potential computed over the workspace of the robot. Since no binary obstacle area exists, each point of the terrain contributes to the repulsive obstacle potential according to its obstacle coefficient. The computed motion plan allows the robot to navigate through the terrain taking into consideration not only the topology but also the physical underground