针对Q学习状态空间非常大,导致收敛速度非常慢的问题,给出一种基于边界样本协调的多智能体在线合作学习方法,使得智能体在特定的子空间上进行特化并通过边界状态上的开关函数相互协调,从而能够较快地学习到局部最优.仿真实验表明该方法能够取得比全局学习更好的在线学习性能. Aiming at the problem that the learning space of Q is very large and the convergence speed is very slow, a collaborative multi-agent collaborative learning method based on the coordination of boundary samples is proposed, which enables the agent to specialize in a particular subspace and pass the boundary state Which can be used to learn local optimum quickly.Experimental results show that the proposed method can achieve better online learning performance than global learning.