It is a great challenging but solvable task in an actual robot confrontation: how to effectively pursue the target robot. In this paper, a maximized entropy-based deep reinforcement learning mobile robot active pursuit strategy (APS) approach is proposed for target tracking challenges, then our strategy approach will be implemented on a physical robot. The use of maximizing entropy reinforcement learning allows the agent to have diverse strategies to adapt to dynamic environment changes. We combine learning-based algorithms with rule-based algorithm that allow agent to avoid being targeted by human players to the greatest extent possible in confrontations. We conducted experiments by transferring the learned pursuit strategies to a physical world mobile robot and analyzed the feasibility of application in ICRA RoboMaster AI Challenge (RMUA) environment. The experimental results demonstrate the feasibility and potential of the algorithm in learning multiple tracking strategies.