Towards Robust Multiagent Plans

Abstract : This project, titled Towards Robust Multiagent Plans with its extension Domain-Independent Multiagent Planning: Models, Stability, and Complexity focused on theoretical and applied research in field of multiagent planning in dynamic environments. The objective of the project was to connect, both formally and empirically, the developments in domain-specific multiagent planning and the concepts of generic, domain-independent problem solving and propose solutions and techniques robust to uncertainty, dynamism and non-determinism of environments modeled closely to real world. This final report summarizes general overview of the project, achievements of the project mostly in the form of published research work, and describes the demonstrator. The research work in the project comprised of four accepted and one submitted journal publication aiming at advanced techniques for multiagent plan repair, simple regret optimization in online planning, oversubscription planning, and a submitted journal article on novel type of multi-heuristic search for multiagent planning together with overview of the Multiagent Distributed and Local Asynchronous (MADLA) planner. Furthermore, eight accepted papers at the top artificial intelligence and planning conferences focused on fault tolerant planning and interruptible exploration technique usable in Monte-Carlo tree search algorithms. The submitted and accepted workshop papers provided good ground for valuable discussion at the specific research forums and propagated the work.