Temporal Multiagent Planning with Concurrent Action Constraints

This paper investigates how centralised, cooperative, multiagent planning problems with concurrent action constraints and heterogeneous agents can be encoded with some minor additions to PDDL, and how such encoded domains can be solved via a translation to temporal planning. Concurrency constraints are encoded on affordances (object-action tuples) and determine the conditions under which a particular object can (or must) be utilised concurrently. The effectiveness of the approach is evaluated on the Vehicles testing domain and on a new Warehouse domain, which is inspired by a real-world warehouse problem in which a centralised mission planner must find a concurrent plan for a fleet of robots in a manufacturing plant. The approach is shown to be promising, with the potential to support future work in the area.