|Office||A303 Hamerschlag Hall|
Increasing numbers of design and manufacturing tasks will be tackled by large, widely distributed teams of people, computer-based agents and manufacturing facilities. This project is developing ways by which the members of such teams can collaborate over the Internet, benefiting from its coverage without being compromised by its failings.
Organizations in which agents are allowed to act autonomously can react faster and are more robust than hierarchic, centrally controlled organizations. This project has been developing techniques and rules by which autonomous agents can cooperate (help one another) and learn (automatically transform experience into improved competence). A number of off-line optimization problems have been successfully tackled, including integer programming, robot design and process scheduling. The current focus is on real-time control, particularly the control of large, distributed networks, such as electric grids and traffic networks. We are in the process of developing a generic, autonomous, context-dependent agent that, when installed in a network, will automatically learn its job and how to do it better.
The key to running an electric power system is in controlling its dynamics. Today, these dynamics are determined by natural laws. In the future, they will be determined by policy and market forces. This project is developing the methods and tools by which to understand, analyze and control these future dynamics.
Carnegie Mellon, 1974
Signals and Systems
Autonomous agents, distributed control, power systems, cascading failures