SAT-Based Cooperative Planning: A Proposal

We present a work-in-progress on distributed planning, which relies on the "planning as satisfiability" paradigm. It allows for multi- agent cooperative planning by joining SAT-based planning and a par- ticular approach to distributed propositional satisfiability. Each agent is thus enabled to plan on its own and communicate with other agents during the planning process, in such a way that synchronized and pos- sibly cooperative plans come out as a result. We discuss in some details both piers of our construction: SAT-based planning techniques and dis- tributed approaches to satisfiability. Then, we propose how to join them by presenting a working example. Planning is a research area in artificial intelligence aiming at the construction of algorithms — called planners — that enable an agent (a robot or a "softbot") to synthesize a course of actions that will achieve its goals. Planning has been studied since the early days of AI; recently the interest has been renewed and results abound, with proposals of systems capable to deal with real life applica- tions. We cannot go into many details here, neither illustrate the vast literature in the field; a comprehensive presentation of the state of the art can be found - for instance - in the survey paper by Weld (23). Much research has recently been done in multi-agent systems and in robot teams, i.e., scenarios where several software or embodied systems cooperate at the solution of a given goal. In this case, planning becomes "cooperative plan- ning". Cooperative planning has various meanings in the current AI literature. All of them are geared around the idea that each agent - participating in the cooper- ative achievement of a common goal - has to take into account, in its planning activity, that other agents are present. Both the presence and the actions of the various agents affect the scenario. In other words, things not only change be- cause of the effects of the actions of an individual agent, but also because of the effects of the actions of its companions. In some cases, planners take also into account the possibility of exogenous actions; of course, in a multi-agent scenario, actions of the companion agents are not to be considered exogenous. This view is exemplified - for instance - in (12).