Hopfield Neural Network for Dynamic Path Planning and Obstacle Avoidance

Navigation is a central issue in the area of autonomous agents and the dynamic nature of the real world environments requires fast path planning methods for successful operation of the agent in them. We present a dynamic path planning and obstacle avoidance scheme based on a Hopfield neural network. The external input (sensory input) provides information to the system about the position of the obstacles, the current position of the agent and the desired target position in the agent’s configuration space. The dynamics of the network create an equilibrium surface homomorphic to the configuration space from where a path can be easily constructed by applying the steepest ascent technique. In the dynamic case, the network is updated continuously and the system suggests the most appropriate next move at each time. The provided examples demonstrate the approach in both static and dynamic cases. At the end, a detailed discussion about the capabilities of the system, possible extensions and future work, points out a promising direction of research.