Cooperation of CPU and GPU Programs for Real-time 3D Map Building

This paper presents how the CPU and GPU programs coordinate i th context of 3D map modeling for a mobile home service robot. In this study, the representatio n of the environment is given as point clouds, and each scan of point clouds is quite efficiently processed u ing the parallel processing capability of GPU. Then, the result is read back to CPU for incrementally constr ucting the map. Due to the coordination between the CPU and GPU, a 3D map can be built at real time. This paper pr esents the software architecture of the CPU-GPU coordination, the GPU algorithm, and its performan ce gain.