A quadtree based neural network approach to real-time path planning

The paper proposed combining traditional quadtrees and framed-quadtrees with the shunting equation based neural network model to improve the efficiency of path planning. The introduction of quadtrees is used for improving the efficiency of the trajectory generation and enlarging the representation capability of maps, especially in sparse environments. And the introduction of framed-quadtree is used for the generation of Euclidean shortest paths. The introduction of quadtrees and framed-quadtrees does not change the structure of the neural network model based on the shunting model; so the stability and the convergence of the neural network were reserved. And the feature that a map can be represented by quadtrees with multi-resolution was betaken to simplify the selection of parameters in the neural network model. Theoretical analyses and Simulation studies of the proposed method were done to demonstrate following conclusions: the Euclidean shortest paths can be generated without collision and without much computational complexity; the improved neural network method does not suffer from undesired local minima; the proposed method can generate shorter collision free trajectory and has bigger representation capabilities of maps.