This dissertation presents a methodology and automated algorithms for collaborative design. The methodology calls for individuals to independently create designs achieving their own goals, and then collectively derive a single unified design using automated negotiation tech­niques. From a software engineering perspective, the methodology provides parallelism, simplic­ity, rationale, and reuse. From a negotiation perspective, the methodology provides multiple agent preference maximization and novel resolution synthesis. From an artificial intelligence perspective, the algorithms provide automation for the complex processes of conflict detection, resolution synthesis, and resolution selection. This dissertation describes how the selfish interests of individuals or subgroups can productively aid the derivation of robust collaborative designs through the automated negotiation of their conflicts.

This dissertation describes formal representations for modeling individual perspectives, design conflicts, and subtasks involved in negotiation. Specifically described are representations for: (1) goals and preferences over domain operators, objects, and relations, (2} categories of design and goal conflicts, and (3) categories of conflict resolutions. Automated processes can manipulate these representations to aid group negotiation.

This dissertation describes formal algorithms for detecting conflicts and synthesizing resolutions. Specifically described are algorithms for: (1) distinguishing between simple design differences and design interference, (2) mapping between goals and their supporting design components, (3) detecting goal conflicts, (4) synthesizing analytic and heuristic resolutions, and (5) reintegrating resolved goals into a design. Analytic resolution consists of compromise generation using a multiple criteria linear programming method. Heuristic resolution consists of search through domain hierarchies to synthesize dissolutions and compensations. These methods have been implemented and applied.

This dissertation describes the implementation of our negotiation algorithms and their application to library design problems. The design of library systems is a complex, multiple agent, negotiation enterprise. We have represented portions of documented library designs in our implemented collaborative design tool, Oz. Oz has been used to detect conflicts and derive negotiated resolutions similar to those published by expert librarians. The implementation and its application to the library domain support the central tenet of this dissertation: processes of negotiated design can be automated through the representation of a generic domain model and specific representations of individual peispectives.