With the development of technologies such as communication, computing, and semiconductors, autonomous driving in vehicles has garnered wide-spread attention, becoming a current research hotspot. Autonomous driving places high demands on the real-time aspects of vehicle communication, sensing, and decision-making, necessitating more efficient resource management approaches. However, the rapid changes in channel conditions due to high-speed vehicle movement make centralized resource management less suitable. Therefore, this paper focuses on the power control problem in vehicular networks and investigates the adaptive energy allocation issue in the context of the Dual-Function Radar and Communication (DFRC) system. By modeling the resource-sharing problem as a Markov decision process and leveraging multiagent deep reinforcement learning, a distributed Multi-Agent Double Deep Q-Learning Network (MADDQN) algorithm is proposed. Each V2V link autonomously engages the communication environment, monitors conditions independently, and receives a unified reward. Through experience-based updates to the Q-networks, the algorithm enhances spectrum access and power allocation. Experimental findings demonstrate that, through appropriate design of reward structure and training method, multiple agents can achieve decentralized collaboration, thereby enhancing both the overall capacity of V2I connections and the likelihood of successful payload transmission in V2V links simultaneously.