The question of how to navigate is critically important to the success of any lunar rover mission. While humans typically have good judgement about which routes to take, they often get fatigued and disoriented when teleoperating rovers with time delay. On the other hand, while autonomous systems can produce safe and reliable navigation, they tend to be myopic. We are investigating mixed-mode methods of control that combine the strengths of humans and rovers. The rover uses range maps produced by stereo vision and a detailed model of the vehicle to evaluate the traversability of various paths. The evaluations are combined with recommendations from a human operator to produce a commanded steering angle and speed that is both safe and responsive to the operator’s objectives. We have implemented and are testing such a system, using a prototype lunar rover that operates in outdoor, natural terrain.