Abstract: The adaptability of Open Multi-Agent System (OMAS) is embodied in the adaptation of agents to the changes of environment at runtime by co-adjusting the System behavior．Aiming at the problem that the predefined collaboration mechanism of existing multi-agent design cannot satisfy the runtime adaptive collaboration, this paper proposes an adaptive multi-agent Capability Commitment collaboration Graph-planning Protocol (CCGP) algorithm based on the Goal-Capability-Commitment (GCC) metamodel. Firstly, the GCC model supporting the semantic similarity calculation of context environment is proposed. Secondly, the semantic matching degree between goals and capabilities (or commitments) is introduced into the graph-planning protocol based on capability collaboration to optimize the algorithm. Finally, the medical waste Automated Guided Vehicle transportation system is taken as the experimental scenario to conduct two sets of comparative experiments. The experimental results show that the execution time of CCGP algorithm for graph planning protocol generation algorithm is significantly improved.

Key words: goal-capability-commitment model, capability commitment, adaptive collective, semantic similarity, graph planning