A DISTRIBUTED APPROACH TO N-PUZZLE SOLVING Alexis Drogoul and Christophe Dubreuil Equipe MIRIAD - LAFORIA - UNIVERSITE PARIS VI Boîte 169 4, Place Jussieu 75252 PARIS CEDEX 05 FRANCE e-mails: drogoul@laforia.ibp.fr, dubreuil@laforia.ibp.fr Content areas Distributed Problem Solving, Multi-Agent Planning, Reactivity Abstract We present in this paper a distributed approach for solving the N-puzzle. This approach is based on the decomposition of a problem into the set of the smallest independent subgoals that can be described, be they serializable or not. These subgoals are at their turn decomposed into agents whose task is to satisfy the subgoal. We have chosen the Eco-Problem-Solving model for describing the internal functioning of these agents. Each of them is then characterized by a state, a goal, a satisfaction behavior and a flight behavior. Those kernel behaviors invoke domain- dependent actions that must fit with the domain in which the agent is involved. The description of these actions is made by presenting their basic algorithm and the heuristics used in it, namely the MGB and VMGB distance computations. A simple example of solving is presented, followed by an example of what can be called an “emergent solution” to the problem of the corners. We prove, then, that the method is complete and decidable for any size of N-puzzle. Its theoretical complexity is calculated, with respect to the size of the puzzle, for three values: the room used in memory, the solution length and the execution time. Following this theoretical part, the experimental results allow us to show the first published results for puzzles whose size is over 100.