With the rapid development of Internet of Vehicle (IoV) technology,the role of edge computing in IoV information processing applications is increasingly significant.Vehicle devices in IoV have limited computational resources and storage capacities,yet they need to process massive amount of data and perform complex computational tasks.A single offloading method cannot satisfy the needs of various scenarios,and the existing single communication mode restricts the flexibility and efficiency of task offloading,therefore it is necessary to consider the integration of multiple offloading approaches in a dynamic network environment to improve the overall performance and flexibility of the system.To address the above challenges,a multi-mode collaborative edge computing task offloading scheme for IoV based on deep reinforcement learning is proposed.By adopting communication modes such as Vehicle to Vehicle (V2V),Vehicle to Vehicle Alliance (V2A),Vehicle to Roadside Unit (V2R),and Roadside Unit to Base Station (R2B),the scheme achieves collaborative processing of offloading tasks with high computational density and stringent latency requirements.To solve the high-dimensional optimization problem,a modeling approach based on Markov Decision Process (MDP) and deep reinforcement learning is adopted to cope with the complexity of dynamic state space.In particular,this article designs a Multi-mode Collaborative Task Offloading (MCTO) optimization algorithm to solve the task offloading problem using multiple communication modes under constraints of transmission rate,computational resource,and dynamic network environment.Simulation results show that the proposed MCTO algorithm exhibits good convergence and significantly improves the system latency compared to other reinforcement learning algorithms,with an overall performance increase of 28.67%.