To prolong the lifetime of sensor networks, various scheduling schemes have been designed to reduce the number of active sensors. However, some scheduling strategies, such as partial coverage scheduling and target coverage scheduling, may result in disconnected network topologies, due to the low density of the active nodes. In such cases, traditional routing algorithms cannot be applied, and the shortest path discovered by these algorithms may not have the minimum packet delivery latency. In this paper, we address the problem of finding minimum latency routes in intermittently connected sensor networks by proposing an on-demand minimum latency (ODML) routing algorithm. Since on-demand routing algorithm does not work well when the source and destination frequently communicate with each other, we propose two proactive minimum latency routing algorithms: optimal-PML and quick-PML. Theoretical analysis and simulation results show that (1) ODML can effectively identify minimum latency routes which have much smaller latency than the shortest path, and (2) optimal-PML can minimize the routing message overhead and quick-PML can significantly reduce the route acquisition delay.