next up previous

Comparison with Other Agent Architectures

The most similar agent architectures are FLiPSiDE [[17]] and that of Genesereth and Singh [[9]]. Like FLiPSiDE (Framework for Logic Programming Systems with Distributed Execution), our Open Agent Architecture uses Prolog as the interagent communication language, and introduces a uniform meta-layer between the blackboard Server and the individual agents. Some aspects of FLiPSiDE's blackboard architecture are more complex than in our system. It uses a multi-level locking scheme to try to reduce deadlock and minimize conflicts in blackboard access during moments of high concurrency. The system also uses separate knowledge sources for controlling triggers, ranking priorities and scheduling the executing of knowledge sources, whereas we incorporate these sorts of actions directly into the blackboard server. Some features important to our system that are not addressed by FLiPSiDE are the ability to handle temporal contraints over variables, and the possibility for an agent to explicitly request AND and OR-parallel solvingof a list of distributed goals.

Genesereth and Singh's architecture is more ambitious than ours in its employing a full first-order logic as the interagent communication language. As yet, we have not needed to expand our language beyond Horn clauses with temporal constraints, but this step may well be necessary. Genesereth and Singh use KIF (Knowledge Interchange Format) [[13]] as their basic language of predicates and as a knowledge integration strategy. Because of our user interface considerations, which in turn are heavily influenced by the form-factor constraints of future handheld devices, we will need to be able to merge contributions by different agents of their natural language vocabulary, related pronunciations, and semantic mappings of those vocabulary items to underlying predicates.

Adam Cheyer
Mon Aug 12 15:12:15 PDT 1996