This paper describes the software architecture and the initial algorithms that have proved to be effective for a real time robot planning system. The architecture is designed to incorporate planning technology from research on artificial intelligence while at the same time supporting the high performance decision making needed to control a fast-moving autonomous vehicle. The symbolic representation of the vehicle's plan is a key element in this architecture. Our initial algorithms use an especially efficient version of dynamic programming to find the best routes. The route is then translated into a symbolic plan. Replanning happens at several levels with the cost of replanning proportionate to the scope of the changes. This software is currently running in an environment which simulates the vehicle and perception systems, but it will be transferred to the DARPA Autonomous Land Vehicle built by Martin Marietta Denver Aerospace [Lowrie 86].