This paper addresses the autonomous decentralized cooperation conveyance algorithm to improve the task efficiency by selecting a leader from other robots autonomously. This algorithm aims at coexistence of two exclusive features. One is the effectiveness that is a typical advantage of centralized control systems. The other is the adaptability and robustness to the environment which are advantages of autonomous-decentralized systems. From the several experimental results, the algorithm can improve the task efficiency in the cooperative conveyance including clever agent by human operation or known optimal trajectory. Further, to selects the larders on real-time, evaluation indexes by interaction of agents are discussed.