A#: A distributed version of A* for factored planning

Factored planning consists in driving a modular or distributed system to a target state, in an optimal manner, assuming all actions are controllable. Such problems take the form of path search in a product of graphs. The state space of each component is a graph, in which one must find a path to the local goal of this component. But when all components are considered jointly, the problem amounts to finding a path in each of these state graphs, while ensuring their compatibility on the actions that must be performed jointly by some components of the system. This paper proposes a solution under the form of a multi-agent version of A*, assembling several A*, each one performing a biased depth-first search in the graph of each component.