Radial basis function networks for mobile robot localisation
Townsend NW., Brownlow MJ., Tarassenko L.
The problem of estimating a robot's x - y position in its environment is considered. A neural network solution is proposed, using an RBF network whose hidden layer is a Kohonen feature map which relies on (optical) range data for its input The optimisation of such a network is considered and results from such a network are given. It is shown that the network learns to compensate for the discontinuities in input space which are associated with obstacles within the environment The network gives an average localisation error which is less than the diameter of the robot and is therefore acceptable for robot navigation tasks.