The invention discloses a multi-underwater-robot formation control method based on reinforcement learning, and relates to the field of control of underwater robots. In the invention, after each robotnode in an underwater robot formation acquires the position of the robot node, a control center gives track information of a virtual leader and sends the track information to neighbor nodes of the virtual leader; a topological communication network is established among underwater robot nodes, and each underwater robot node only communicates with the neighbor nodes to keep the formation stable; theunderwater robot formation uses the current control strategy to track, each node calculates a one-step cost function by interacting with the environment and the neighboring nodes, the current controlstrategy is improved by minimizing a value function, when two steps including value iteration and strategy improvement both reach convergence, the control strategy of the underwater robots tracking an expected trajectory is optimal, and the optimal control strategy is used to achieve the goal of accurate tracking.