high performance computing on graphics processing units: hgpu.org

The release of the CUDA architecture made massively parallel computing possible on ordinary desktops and opened a door to enormous computing power of graphics adapters. The trajectory planning for aerial vehicles is one of the tasks that can benefit from it. The sought path must respect all physical limitations of the airplane and avoid all no-flight zones. The thesis presents two algorithms for trajectory planning on the CUDA architecture: a parallel version of A* algorithm and Accelerated A* algorithm that uses varying planning steps to speed up the planning. The parallelization relies on a distribution of states between individual threads. To implement the proposed algorithms, a block synchronization that is not officially supported in CUDA is required. Two solutions to this problem are given in the thesis. Both algorithms are experimentally evaluated and compared to their sequential version.