Real-time navigation of a mobile robot using Kohonen's topology conserving neural network

This paper proposes a real-time sensor based navigation method using Kohonen's topology conserving network for navigation of a mobile robot in any uncertain environment. The sensory information including target location with respect to current location of the mobile robot, have been discretely conserved using a two dimensional Kohonen lattice. Reinforcement learning based on a stochastic real valued technique have been implemented to compute the action space for this Kohonen lattice. The proposed scheme learns the input and output weight space of the Kohonen lattice which is generalized to any workspace. The effectiveness of the proposed scheme has been established by simulation where the complete domain of the input-space is quantized based on experience on sensory data encountered in real-time. The input-output mapping conserved by the Kohonen lattice during simulation was used to guide a mobile robot in a real-time environment. Successful navigation of the mobile robot without further training confirms the robustness of the proposed scheme