Use of a real-time perception program supervisor in a driving scenario

This paper describes the use of PLANETE, a real-time knowledge-based system for program supervision in the framework of an intelligent on-board guidance system (or co-pilot) for a motor vehicle in a realistic situation. The vehicle is equipped with several sensors, producing data, that can be processed by several programs. The role of PLANETE is to convert the requirements of the high-level situation analysis module (expressed in terms of symbolic requests), into the correct low-level processing commands, under computational and time constraints. We discuss the underlying concepts of program supervision, we detail the system architecture (knowledge base and inference engine) and its implementation, and eventually we analyze a realistic example of a partial scenario, describing the path from the events received by the co-pilot, to commands ([de]activations of sensors and programs).