Finding the shortest path by use of neural networks

The authors present a method for finding the shortest trajectory in 2D space by neural networks. To solve effectively the trajectory planning problem with obstacles of arbitrary shape, they propose a neural network to transform the free space into a structured path network characterizing its topological property. The representative of each topological class is then optimized by a cellule network simulating a retraction minimizing the energy of the system. And the shortest one from different classes gives therefore the final solution. This method works well for obstacles of arbitrary shape; it is simulated and tested for 2D trajectory planning tasks, and the experimental results are satisfactory.<>