Fast reactive path planning by 2D and 3D multi-layer spatial grids for mobile robot navigation

Abstract

Describes a new and fast path planning algorithm for sensor-based robot navigation. The goal of path planning is the computation of the fastest path from a start point to a destination point in a workspace. It should be able to react to unexpected obstacles by means of fast reactive replanning. The authors use a diffusion algorithm for multi-dimensional workspaces (e.g. 2D, 3D, etc.). Application of a multi-layer map which is permanently updated by new sensor information to intelligently select one of the possible paths distinguishes this approach from prior work in this field. The proposed diffusion algorithm is formulated in the data-parallel language C*; in the case of a 3D workspace there is a comparison between response time duration on a SUN Sparcstation 20 and a Connection machine CM-5 with 32 nodes.