In distributed planning, plans are generally generated by "classical" planners which resolve the goal assigned to a single autonomous system, that is to say an agent in our terminology, coordination and conflict resolution being after that delegated to appropriate mechanisms. This twofold approach has been tailored so as to deal with combinatorial complexity due to one shot centralized planning or mandatory knowledge scattering.
The purpose of the CoRe project is to handle distribution issues as soon as possible within the planning process. To that end, a specifical planner, CoRe-planner, has been implemented from scratch. This Plan-Space planner is based on the generation of conjectures and hypothesis about the goals that the involved agents can achieve. Hence, the CoRe-planner generates plans that can be executed if some hypothesis are dispelled. Unlike "classical" planners, the planning process does not fail if some conditions are not verified ("the door is not open") but rather an agent proposes an assumption-based plan or conjecture as reasonable as possible and counts on its teammates' competences to make those assumptions become true ("if that door could be opened, I could..."). In other words, the assumptions made by an agent become additional goals to be satisfied. The assumption-based planning is justified by the possible collaborative reasoning in a multi-agent context.
This assumption-based planning is the primary level on which the agents elaborate a contradictory or dialectical reasoning: in order to take into account the partial knowledge and the heterogeneous skills of the agents, we propose to consider the planning problem as a defeasible reasoning where agents exchange proposals and counter-proposals and are able to conjecture, that is to say formulate plan steps based on hypothetical states of the world. The dialogue between the agents is a joint investigation protocol enabling them to progressively prune objections, compose their competences, solve causal and temporal constraints, and progressively converge towards a solution: the production of a global shared plan is obtained by conjecture - refutation cycles (CoRe stands for Conjectures/Refutations). This is the second level of this approach.
The main difficulty of this approach lies in the elaboration of efficient heuristics to generate assumptions and conjectures. Our heuristics are based on the notion of resilience: a conjecture is all the more resilient since it is able to resist to the agents' refutations; the generated assumptions enable to swiftly converge to a solution; and the committed resources and competences are the less critical.
This project is made of three sub-projects: the first one is the assumption-based planner, CoRe-planner which enables agents to devise conjectures; the second one is a middleware used by the agents to interact, CoRe Distributed Message Service (CoRe-DMS) and finally, the third one is an integrated CoRe-IDE that enables to develop distributed agent planners. These sub-projets have been implemented with Java 1.5.