Reference: McIlraith, S.; Biswas, G.; Fromherz, M.; Howe, J.; Fikes, R.; Bobrow, D.; Cutkosky, M.; Engelmore, R.; & Neller, T. Model-Enabled Control of Hybrid Systems. Knowledge Systems Laboratory, July, 1998.
Abstract: Software is used to control complex systems as varied as spacecraft, land vehicles, life support systems, and factories. Traditional methods for developing control software do not support the flexibility, autonomy, and reliability required of these systems, and often do not support the hierarchical composition of such systems from independently developed components. For many of these inherently hybrid (discrete and continuous) systems, combining traditional feedback PID control with discrete computer-based controllers is a significant challenge in the design and implementation of the control systems. This project proposes a new paradigm, model-enabled control, that uses declarative models augmented with automatic reasoning systems to support the design, development, implementation, and validation of controllers for hybrid, hierarchical systems.
This project will develop techniques for computational modeling and analysis, and a computational infrastructure to support the autonomous model-enabled control of multiple hybrid systems. Model-enabled control is realized by embedding rich computational device models and associated reasoning and analysis machinery directly into on-line control systems. Such model-enabled controllers can autonomously adapt to new high-level task requirements and unanticipated changes in their environment. They also can detect deterioration or failure of devices and compensate for such contingencies through reconfiguration of the remaining components. A multi-level model-enabled control architecture can provide the machinery for coordination of tasks among multiple autonomous systems.
The autonomous control of multiple airborne vehicles for space and earth science, surveillance, and weather mapping is an example of a complex problem for which traditional simulation and control techniques are inadequate. Each vehicle in the group or fleet is, of itself, a complex hybrid dynamic system. Traditional approaches to the fleet control problem require human operators who determine mission-level tasks, convert the tasks to detailed commands, and continually upload the commands to individual vehicles. The result is inflexible, custom-built, real-time operations with very little ability to adapt and react to unexpected situations caused by failures in subsystems, changes to the vehicle environment, and the introduction of new goals and tasks. Model-enabled control is particularly suited to the task of autonomous control of multiple airborne vehicles. We will, therefore, use that problem domain to test and demonstrate the technology developed in the project.
Developing software in support of model-enabled control presents diverse computational challenges. The success of our research thus requires a coupling of expertise from multiple disciplines of science and engineering including control theory, artificial intelligence, model-based reasoning, and hybrid systems modeling and control, as well as disciplinary expertise in the design, modeling, and control of airborne vehicles. This project is proposed by a multi-disciplinary team of researchers from the Stanford Computer Science Departmentís Knowledge Systems Laboratory (KSL), the Stanford Mechanical Engineering Departmentís Center for Design Research (CDR), Stanfordís Aeronautics & Astronautics Department (Aero/Astro), Vanderbiltís Computer Science Department, and an industrial collaboration with the Systems and Practices Laboratory (SPL) at the Xerox Palo Alto Research Center.
The proposed approach is to build on the previous work of the team members. Specifically:
Our research will bring together these independent strands of research to produce systems with dramatically enhanced capabilities in autonomous coordinated control. The results will have impact on those who design and develop controllers, especially in control of airborne vehicles, and on those who simulate, diagnose, and verify hybrid systems. The research will result in new computational capabilities and tools for both computer scientists, control engineers, and design engineers.
Full paper available as ps.