With the rapid development of intelligent transportation systems, the demand of the Internet of vehicles (IoV) for real-time computation and low-latency services has surged. Vehicular edge computing (VEC) significantly reduces the transmission delay by offloading tasks to edge nodes. However, traditional algorithms are not sufficiently adaptive to task offloading in complex dynamic traffic environments. Deep reinforcement learning (DRL) is capable of handling complex tasks and learning optimal offloading strategies for vehicles in complex dynamic environments. Firstly, this study sorts out the IoV architecture, communication technology, and core offloading technology of VEC. Secondly, it introduces the basic theory of DRL, classification of methods, and the mechanism of multi-intelligence body collaboration. Then, a comprehensive overview of the current research status at home and abroad is given from the vehicle-vehicle, vehicle-edge layer, and cloud-edge-end resource cooperative computing offloading dimensions. Finally, the possible future research directions of deep reinforcement learning-based VEC and task offloading are pointed out.