The Center for Optimization and Semantic Control (COSC)

The Center for Optimization and Semantic Control (COSC), established in 1987, consists of graduate and undergraduate students, members of faculty from the Schools of Engineering, Business, and Arts and Sciences at Washington University, affiliate members from industry, from the US Air Force and from the US Transportation Command. The Center provides opportunities for inter-disciplinary mathematical and computer modeling efforts, optimization and application of advanced technologies to decision-aiding, scientific and engineering problems from conceptual design through implementation. It fosters teamwork with diverse groups, including industrial, academic, medical and military.

The main focus of the Center has been the development and application of multi-disciplinary designs for modeling, simulation, optimization and control of several problems in aerospace and transportation systems.

Such an inter-disciplinary approach has become a necessity in order to cope with the large, time-dependent, complex, nonlinear and uncertain nature of such systems, for which the existing and classical methodologies of solution are not available or are not sufficiently powerful.

Since the mathematical models of such systems are typically vague, Center researchers use a judicious combination of classical mathematical methodologies (modeling, optimization, control theory, differential equations, stochastics, etc.) together with artificial intelligence paradigms, such as rule-based systems, logic programming and artificial neural networks. The semantic control paradigm, (introduced by E. Y. Rodin in 1985) a three-layered hierarchical system, provides a suitable framework for the realization of this synergistic approach. It consists of the following modules:

an Identifier which interfaces the exosystem via sensors and develops an abstraction of the observations; i.e., provides a representation, or performs systems identification;

a Goal Selector which plans the course of action in order to find an optimal or near-optimal choice among those available; and

an Adapter which selects and executes appropriate control laws.

In order to optimize performance, a machine learning loop (both symbolic and connectionist approaches) has been incorporated in the above design, through a reinforcement learning loop.

This multi-disciplinary approach to the design and simulation of control and decision support modules is effective in coping with many large-scale, dynamic, complex, hybrid, nonlinear and uncertain systems.

Background

Beginning in 1985 we considered the optimal control of an aircraft in uncertain environments using aspects of differential games, decision theory, AI, and computational geometry; we developed algorithms for optimal path planning in a dynamic, hostile environment using Delaunay triangulation, multi-objective A* search, and maximum utility theory; we improved on all existing methods for the recognition of aircraft maneuvers through a neural network paradigm; we used flexible Delaunay triangulations and Voronoi diagrams in navigational planning for collision avoidance. We introduced new time-dependent mathematical programming paradigms for real-time resource allocation using time-sweeps and rule-based technology. This last area also involved theoretical and computational advances for dealing with very large scale optimization problems through new aspects of polyhedral theories.

As an example, we applied Semantic Control Theory to several aerospace-related problems in air combat games and for the optimal collision-free motion of mobile robots in a time-varying environment. The methodology culminated in the development of a Tactical Decision-Aiding Expert System (TDAES).

The Center enjoys a close working relationship with both industrial corporations and the military. Some of our recent collaborative efforts include:

Transportation, Optimization and Scheduling Theory

Joint work with USAF's Air Mobility Command (AMC) in modeling, simulation and optimization of DoD's large-scale air transport operations;
Joint work with USTRANSCOM on modeling, simulation and optimization of various aspects of USTRANSCOM's air, ground and water operations;
Joint work with Systems & Electronics Inc. in Intelligent Transportation Systems: urban traffic prediction and management; vehicle routing and automatic control.

Semantic Control for Game Theory

Development of a Flight & Fire Control System with Rockwell International (which was tested in Germany by Messerschmidt and adapted as a "pilot's assistant" for the Advanced Euro Fighter).
Development of a tactical decision support system with ESCO for situation assessment, navigation and control of a vehicle engaged in evasive maneuvers against multiple pursuers.
Joint work with Electronics & Space Corp.-St. Louis on embedded control and decision-aiding system for situation assessment, guidance and control of a vehicle engaged in evasive maneuvers against primary and secondary pursuers.

System Identification and Intelligent Control

Development of a Flight & Fire Control System with Rockwell International (which was tested in Germany by Messerschmidt and adapted as a "pilot's assistant" for the Advanced Euro Fighter).
Development of a tactical decision support system with ESCO for situation assessment, navigation and control of a vehicle engaged in evasive maneuvers against multiple pursuers.
Joint work with Boeing Corporation to develop a neural network augmented anti-skid control system for MD-90 and similar transport aircraft.

Medical Informatics

Diagnosis Prediction via an Artificial Neural Network Knowledge Base