Planning for Loosely Coupled Agents Using Partial Order Forward-Chaining
Linköping University, Department of Computer and Information Science, KPLAB - Knowledge Processing Lab
Linköping University, The Institute of Technology
Conference paper (Refereed)
Proceedings of the 21st International Conference on Automated Planning and Scheduling (ICAPS)
Fahiem Bacchus, Carmel Domshlak, Stefan Edelkamp, Malte Helmert
21st International Conference on Automated Planning and Scheduling (ICAPS 2011), 11-16 June 2011, Freiburg, Germany
We investigate a hybrid between temporal partial-order and forward-chaining planning where each action in a partially ordered plan is associated with a partially defined state. The focus is on centralized planning for multi-agent domains and on loose commitment to the precedence between actions belonging to distinct agents, leading to execution schedules that are flexible where it matters the most. Each agent, on the other hand, has a sequential thread of execution reminiscent of forward-chaining. This results in strong and informative agent-specific partial states that can be used for partial evaluation of preconditions as well as precondition control formulas used as guidance. Empirical evaluation shows the resulting planner to be competitive with TLplan and TALplanner, two other planners based on control formulas, while using a considerably more expressive and flexible plan structure.