This work presents a neural model of self-localisation implemented on a simulated mobile robot with a realistic visual input. A population of modelled place cells with overlapping receptive fields is constructed online during exploration. In contrast to similar models of place cells, parameters of neurons in the sensory pathway adapt online to the environments statistics in order to maximise information transmission. The robot's position can be decoded from the population activity with high accuracy. The information transmission rate of the cells is comparable to the information rate of biological place cells.