MADLA Planner

The MADLA (Multi-Agent Distributed/Local Asynchronous) Planner is a domain-independent planner for solving deterministic cooperative multi-agent (aka MA-STRIPS) planning problems.

Cite as:

Štolba, Michal, and Antonın Komenda. "MADLA: Planning with Distributed and Local Search." Competition of Distributed and Multi-Agent Planners (CoDMAP-15) (2015): 21. paper as PDF

The planner has participated in the CoDMAP competition

The planner takes as input

• a) PDDL (description of the domain+problem) + ADDL (descritpion of the agents), in which case it first performs multi-agent partitioning according to MA-STRIPS rules
• b) unfactored MA-PDDL

The main principle used is multi-agent heuristic state-space search aka MAFS/MAD-A* (not optimal in our case)

The planner can run in various configurations consisting of

• configuration name - see below

• recursion limit - limits recursion computation for various heuristics, thus swithching between distributed, projected and in-between heuristics

• add 0 ... projected additive heuristic

• add-sync -1 ... distributed additive heuristic

• max 0 ... projected heuristic

• max-sync -1 ... distributed heuristic

• FF 0 ... projected FF heuristic

• saFF-sync -1 ... distributed FF heuristic

• PPsaFF-sync -1 ... privacy-preserving distributed heuristic

• glcl -1 ... Distributed/Local Asynchronous search using both projected and distributed FF heuristic

Other relevant papers:

1. Relaxation heuristics for multiagent planning M Štolba, A Komenda 24th International Conference on Automated Planning and Scheduling (ICAPS)

2. Fast-Forward Heuristic for Multiagent Planning M Štolba, A Komenda Distributed and Multi-agent Planning (DMAP) Workshop of 23rd International ...