NDAS: the nondeterministic agent system for engineering drawing analysis

We describe the use of nondeterministic agent systems (NDAS) to achieve a more flexible system for technical drawing analysis. They are called nondeterministic because the agents explore alternative parts of the solution space simultaneously, and every agent works to produce some result which may or may not contribute to the final result. (This is a form of speculative parallelism [N. Carriero et al., (1990)]; this can also be viewed as a distributed blackboard system [P. Winston (1984)].) The final result derives from only a subset of the work put in by all the agents. We explore nondeterminism in this problem domain since deterministic systems usually make irrevocable decisions (e.g., threshold selection) that eliminate possible solutions. The technical drawing problem domain contains many factors that vary with the drawing: thresholds, text fonts and size, noise levels, etc., and this variation makes it interesting to explore the possible solution space dynamically and in a breadth-first way.