A robot control architecture called SPOTT is proposed and implemented as a soft real-time system of concurrently executing and co-operating modules. What distinguishes SPOTT from other architectures with a behavioral control component is that (1) it is able to guarantee task completion in certain scenarios; and (2) it is scalable to different tasks, control laws, computational resources, and robot platforms. The dynamic path planning module's optimality properties give SPOTT its task completion guarantees. Scalability claims are made possible because of the (1) modularity of the behavioral control programs; and (2) SPOTT's software architecture. SPOTT consists of various components with both deliberative and reactive properties: a behavioral controller, a local dynamic path planner, and a global path planner. In addition, other components include a map database and a graphical user interface. The behavioral control formalism is called TR$+$ and is based on an adaptation and extension of the Teleo-Reactive (TR) formalism. TR$+$ control rules (i.e., condition-action pairs) are structured as modular software components. TR$+$ conditions are based on sensed and internal model states and TR$+$ actions either affect actuator control or perform map database maintenance. The local dynamic path planner is based on a potential field method using harmonic functions, which are guaranteed to have no spurious local minima. The global planning module advises the local planning module on the local effect of the global goal. A real-time and parallel implementation of SPOTT using a message passing software package called PVM has been developed and tested across a collection of interchangeable heterogeneous workstations. Navigational experiments have consisted of moving two different robot platforms in a structured office and an unstructured laboratory environment to known spatial locations with no or a partial a priori map.