Partially centralized approach for dynamic hierarchical-plans merging

摘要

The design of multi-agent systems evolving in complex and dynamic environments requires using a flexible scheme for coordinating and executing agents plans. This paper aims to provide an approach for dynamic coordination taking advantage of the hierarchical plans representation and the ability to reasoning on abstract tasks. The concept of abstraction here are central, by ignoring some less important details, a solution can be found more simply. Our contribution is hence including two complementary aspects. The first contribution focuses on the definition of a representation model supporting the decentralized and flexible execution of multi-agent plans. In a second contribution, we use the obtained model to provide a hierarchical scheme for agents coordinating. Viewed as a hierarchical-like structure of coordination cells (CC), this scheme allows the merge, in partially centralized way, the agents' plans. Based on the ability to localize the interference between tasks supported by our model, the global plan is dynamically decomposed to set of partial-global-plan that may be analyzed in a decentralized manner. This hybridization, between centralized and distributed coordination, can favor the interleaving of planning and execution, some part of global plan can be repaired when others are in execution.