Originality: Uses evolutionary algorithms to synthesize minimal dynamical networks for path-integration behaviour. Introduces new kinds of plastic continuous-time recurrent neural networks; analyses resulting models and links them to improved versions of mathematical models demonstrating their implementability at the neuronal level. Rigour: uses improved fitness function criteria for incremental evolution; deploys dynamical systems analytical tools and compares results with real data on desert ant path integration. Significance: demonstration of how an embodied system may afford quite compact and simple home vector navigation. Shows the significance of compass sensor profiles in facilitating navigation, shows how the effects of neural activation decay can be compensated for. First publication of an evolved model in this experimental journal.