H. Meng and S. Dost (UK)
Neural networks, Robotics, Pathplanning.
This paper describes an improvement over Park’s original
neural path planning algorithm. The Park’s network needs
to set the weights by hand and have serious local minimaproblems. In our network the weights are found using
back-propagation. To improve Park’s method the fast
simulated annealing schedules were employed in the path
finding procedure to avoid local minima. The improved
algorithm provides massive parallelism in computation
and good performance in finding path.