Invited Speaker: Professor Robert F. Stengel
             Princeton University

Abstract

Computational neural networks provide a powerful and flexible basis for controlling dynamic systems. They can accept a large number of command and measurement inputs and can produce a large number of control outputs, they implement nonlinear control logic as readily as linear control logic, and they can be continuously trained on-line to adapt to a changing environment. As digital computation becomes more ingrained in vehicles and vehicular systems, neural-adaptive controllers should find increasing use.

We examine biological and cognitive paradigms for control system design, leading to three applications that demonstrate the versatility and promise of neural-adaptive control. Such control systems can be viewed as intelligent agents, that is, systems that are driven by objectives and current knowledge to achieve goals, with logical structures that mimic natural systems. These systems embrace feedback, adaptation, decision-making, and task planning. The first example uses a cerebellar model articulation controller to regulate a fuel-cell-driven automotive powertrain. The controller maximizes hydrogen conversion from methane or gasoline while minimizing carbon-monoxide production and adapting to driving conditions. The adaptive critic controller, our second example, minimizes a control cost function using neural networks to characterize nonlinearities in the dynamic system. This model-based approach is remarkably robust in the face of system variations and failures, as demonstrated by aircraft flight control. The final example addresses rule-based guidance for autonomous vehicles driving in traffic. The main functions of the guidance system are braking, accelerating, and changing lanes in the presence of multiple neighboring vehicles traveling at different speeds. The guidance law demonstrates the utility of logic based on certainty-equivalent control and stochastic estimation of worst-case outcomes.

About the Speaker:

Robert F. Stengel is Professor and former Associate Dean of Engineering and Applied Science at Princeton University, where he directs the undergraduate Program in Robotics and Intelligent Systems. Prior to his Princeton appointment, he was with The Analytic Sciences Corporation, Charles Stark Draper Lab­o­ratory, U.S. Air Force, and National Aeronautics and Space Administration. A principal designer of the Project Apollo Lunar Module manual attitude control logic, he also con­tributed to the design of the Space Shuttle guidance and control system. Dr. Stengel received degrees from M.I.T. and Princeton University. He is a Fellow of the IEEE and of the AIAA. He received the AACC John R. Ragazzini Control Education Award (2002), the AIAA Mechanics and Control of Flight Award (2000), and the FAA's first annual Excellence in Aviation Award (1997). Dr. Stengel wrote the books, OPTIMAL CONTROL AND ESTIMATION (Dover Publications, 1994) and FLIGHT DYNAMICS (Princeton University Press, 2004).

Location: CEE Conference Room, Phillips Hall 6^th Floor
          The George Washington University, 801 22nd St. NW, Wash. DC 20052
Time: Friday, February 12, 2007, 11:00 AM-12:00 PM