Efficient vehicle motion planning using GPU
Sammanfattning: This master thesis concerns path planning for autonomous vehicles. The focus of this thesis is to evaluate whether we can create a path planner suitable for parallelization, and thus if it is advantageous to implement it to a GPU rather than using it on a CPU. The path planner is built by constructing several Bézier curves originating from an initial point and connecting them to parallel paths following the road. Three different collision checks are evaluated, one grid-based approach and two using vehicle approximation and Euclidean distance. Finally, a cost function containing dynamic and static cost terms is used to evaluate the most suitable path choice. Implementing the planner both on a GPU and a CPU, it is shown that some parts are more suitable for parallelization. To evaluate the planner, examples and simulations in MATLAB are done.
HÄR KAN DU HÄMTA UPPSATSEN I FULLTEXT. (följ länken till nästa sida)