Abstract
In the potential field approach to path planning, the development of the artificial potential field (APF) is a computationally intensive operation. The realization that the APFs can be developed by parallel distributed techniques has prompted interest in using neural networks for developing APFs. This paper describes a neural network called the
wave expansion neural network (WENN) and shows that it is capable of developing a variety of APFs that are useful for path planning. The discretized environment including information about the target configuration (position and orientations) and the obstacles are applied to the WENN as input. Activity is then propagated in the form of waves throughout the WENN neural field and at equilibrium, the resulting neural activity distribution forms the desired APF. We analyze the computational complexity of the WENN based APF generation and compare it with conventional development of APFs. We also describe different path planners which use these APFs to plan paths for moving objects with two and three degrees of freedom.