Dept. of Computer Science and Electrical Engineering
Oregon Health & Science University
This dissertation demonstrates the feasibility of a logic-based declarative language for programming teams of autonomous agents that exhibit correct team and communicative behavior without having to program that behavior explicitly. Teams tend to outperform any loose collection of individuals and are more robust to failures because team members coordinate as required and they communicate with each other appropriately for the success of the team as a whole. As such, the metaphor of team- work is increasingly being employed to build intelligent systems consisting of distributed software entities (agents) that co-operate, coordinate, and communicate effectively as a team. However, teams of software agents are currently constructed by implementing predictions of teamwork theories in a very limited way due to the lack of a sound, comprehensive, and easily programmable approach for building such systems. Therefore, an important problem in multi-agent systems is the creation of a programming framework that enables teamwork and communication in a manner that bridges the gap between the theory and practice of these concepts. This dissertation extends an existing formal theory of teamwork (Joint Intention Theory) by providing a comprehensive formal semantics of multi-agent communication based on that theory along with support for a wider variety of teams. Thereafter, it presents a domain independent agent programming language called STAPLE with built-in support for teamwork and multi-agent conversations based on these theoretical contributions. STAPLE agents are programmed using a subset of modal logic, dynamic logic of actions, and temporal logic along with teamwork constructs and communication primitives that have a well-founded formal semantics. The usefulness of STAPLE for programming teams of autonomous agents is demonstrated by showing that correct team and communicative behaviors follow from agent specifications in two different domains without having to program those behaviors explicitly in every possible situation. Firstly, the fault-tolerance specification of an agent architecture that is robust to sudden broker unavailability is provided to brokers written in STAPLE and the resulting STAPLE-based multi-agent system is shown to duplicate that fault-tolerant behavior. Secondly, STAPLE agents are shown to exhibit correct collaborative behavior in a simulated game that involves human-agent collaboration.
OGI School of Science and Engineering
Kumar, Sanjeev, "A formal Semantics of Teamwork and Multi-agent Conversations as the Basis of a Language for Programming Teams of Autonomous Agents" (2006). Scholar Archive. 16.