DOI: 10.5176/978-981-08-7656-2 A-14
Authors: Ralf Seidler, Michael Schmidt, Andreas Sch¨afer, Dietmar Fey
Abstract: In robot systems several computationally intensive tasks can be found, with path planning being one of them.
Especially in dynamically changing environments, it is difficult to meet real-time constraints with a serial processing approach. For those systems employing standard computers, a viable option is to use a GPGPU as a coprocessor in order to offload those tasks which can be efficiently parallelized. We present implementation results for selected parallel path planning algorithms, taking representatives of every class (graphs, potential fields and iterative), on NVIDIA’s CUDA platform compared to implementations on a standard multicore system using OpenMP. There is no best algorithm for every use case, but one of the iterative methods, the marching pixels solution, seems to be a good choice.
Keywords: parallel algorithms; path planning; GPGPU; robotics; CUDA; OpenMP
Updating...