Many applications of wireless ad hoc sensor and actuator networks (WSANs) rely on the knowledge of node locations. These are challenging to obtain when nodes are mobile and are not equipped with any specific positioning hardware. In this paper, we are interested in scenarios where there are constraints on the movement of nodes, such as with cars on a road network. We develop and analyse a tracking algorithm called MOONwalk that explicitly takes such constraints into account in order to improve the tracking precision. Furthermore, MOONwalk does not require global knowledge of the network, and therefore lends itself well to large-scale and high-mobility applications. We evaluate the accuracy of MOONwalk by comparing it to the optimal maximum likelihood estimator, under different radio conditions and deployment scenarios. We find that MOONwalk performs well despite its localized operation.