Byrnes, Ronald B., The Rational Behavior Model: a Multi-Paradigm, Tri-Level Software Architecture for the Control of Autonomous Vehicles, Ph.D. Dissertation, Naval Postgraduate School, Monterey California, March 1993.
Abstract. There is currently a very strong interest among researchers in the fields of artificial intelligence and robotics in finding more effective means of linking high level symbolic computations relating to mission planning and control for autonomous vehicles to low level vehicle control software. The diversity exhibitied by the many processes involved in such control has resulted in a number of proposals for a general software architecture intended to provide an efficient yet flexible framework for the organization and interaction of relevant software components. The Rational Behavior Model (RBM) has been developed with these requirements in mind and consists of three levels, called the Strategic, the Tactical, and the Execution levels, respectively. Each level reflects computations supporting the solution to the global control problem based on different abstraction mechanisms. The unique contribution of the RBM architecture is the idea of specifying different programming paradigms to realize each software level. Specifically, RBM uses rule-based programming for the Strategic level, thereby permitting field reconfiguration of missions by a mission specialist without reprogramming at lower levels. The Tactical level realizes vehicle behaviors as the methods of software objects programmed in an object-based language such as Ada. These behaviors are initiated by rule satisfaction at the Strategic level, thereby rationalizing their interaction. The Execution level is programmed in any imperative language capable of supporting efficient execution of real-time control of the underlying vehicle hardware. The viability of this architecture has been established through computer simulation studies of control of an autonomous submarine, the NPS Autonomous Underwater Vehicle. These experiments have confirmed that the RBM architecture provides important advantages in terms of program conciseness, maintainability, reliability, and modifiability. In addition, by constraining the interfaces between the levels and limiting the accessibility of state variables, the team development of autonomous vehicle control software is significantly enhanced.