In this thesis, we use a deep reinforcement learning algorithm called Deep Deterministic Policy Gradient (DDPG) to achieve vehicle longitudinal control in the CoMoVe framework. And we properly design the reward function, considering some key performance indicators like safety, comfort and stability to guide the model's learning direction. In the meanwhile, we apply multiple methods to reduce overfitting: fine-tuning hyperparameters, diversifying training scenarios and regularization, which improve the performance and robustness of DDPG. In addition, we also test its improved version called Twin Delayed DDPG (TD3) algorithm for comparison as well as providing reference for future research.