This paper addresses the problem of recovering metric consistency in a global gridmap for mobile robot navigation. Gridmaps can only be updated consistently using exact estimates of the robot position, a requirement which is very hard to fulfil in real world environments because the same sensor data must be used for both map building and self-localisation. To overcome this problem, we use a hierarchy of robot maps which integrates topological and grid-based representations. The consistency problem is solved at the topological level, by applying a relaxation technique to generate coordinates for the places in the robot's map. Consequently, the robot is able to recover a globally consistent gridmap without requiring accurate sensors or high computational costs. Experiments on a Nomad 200 robot are presented which demonstrate the efficacy of our approach.