Abstract: To achieve high-precision and robust motion control for bipedal wheel-legged robots on complex terrains, we propose a TD3-PID-VMC dynamic hybrid control method. First, based on a symmetric four-link closed-loop structure, the kinematic model of the bipedal wheel-legged robot is designed and constructed to achieve coordinated control of wheel rolling and leg support. Then, a setpoint weighting strategy is introduced to improve the cascade Proportional-Integral-Derivative (PID) controller, enhancing the fast response and steady-state speed regulation capability of the hub motors, while Virtual Model Control (VMC) is utilized to optimize system model parameters, thereby improving motion performance and environmental adaptability. Finally, a hybrid weight dynamic allocation module driven by the twin delayed deep deterministic policy gradient (TD3) algorithm is designed. The optimal fusion strategy is obtained through offline reinforcement learning training and deployed online in real-world control, enabling dynamic and adaptive adjustment of the PID and VMC weights to accommodate diverse terrains and motion states. Simulation and hardware experimental results demonstrate that in multi-mode trajectory tracking tasks, the proposed method reduces the root mean square error (RMSE) by approximately 32.9% and 28.7% compared to the fixed-weight PID-VMC method and the LQR-VMC method, respectively, with a recovery time of less than 3.2 s under sudden impact disturbances. The proposed method effectively enhances the stability and precision of the bipedal wheel-legged robot in static balance, disturbance recovery, and trajectory tracking tasks.