Asynchronous design environments: architecture and behavior

Computer-based design environments are becoming increasingly important in architecture and engineering. They offer a common interface for multiple tools, as well as a shareable repository for design information. A design environment's main advantage, however, lies in the potential for its agents to exchange information during the course of solving design problems. Agents communicate to resolve data conflicts, predict the consequences of design decisions, and provide critique and feedback that can be used to improve design solutions. The degree to which a design environment fulfills this potential depends as much on its organization as on the sophistication of its individual components. Unfortunately, existing design environments are weak organizationally: their communication and control strategies are often inflexible, limiting the environment's effectiveness. In addition, their organizational structure limits their ability to change and grow. Furthermore, architectural design presents some additional challenges that exceed the capabilities of most existing environments, including: incomplete information, multicriteria evaluation, lack of a pre-defined design methodology, complex systems integration and distributed task execution. In this thesis, I propose that these weaknesses can be corrected by using an asynchronous team (A-team) organization. A-teams are characterized by autonomous, self-activating agents, broadcast communications, and a control strategy that plans tasks dynamically and executes them concurrently. A-teams have no "managers" or "controllers": their components activate opportunistically and asynchronously. The behavior of such an organization is one of "contracting search"--it emphasizes breadth early in the design process, then "contracts" around the most promising solutions while abandoning other less-successful explorations. To demonstrate the benefits of an asynchronous design environment, I have constructed a prototype system called Anarchy. Its agents are capable of generating, analyzing and modifying designs for medium and high-rise office buildings. Anarchy uses a modified simulated annealing algorithm to implement contracting search. Experiments with Anarchy show that asynchronous design environments are not only feasible, but that they demonstrate both structural and operational flexibility.