In existing heterogeneous multi-agent distributed reinforcement learning methods, existing heterogeneous multi-agent distributed reinforcement learning methods cannot fully depict the different characteristics between heterogeneous agents through unified state and action space definitions, which restricts the realization of adaptive competition and complementarity within the task environment. However, differentiated state and action space definitions bring new challenges to consensus among heterogeneous agents. To achieve collaborative work among heterogeneous multi-agent systems in a vehicle-UAV (Unmanned Aerial Vehicle) heterogeneous cooperative scenario, we propose a distributed reinforcement learning method based on game consensus for heterogeneous multi-agent systems. In this method, to achieve adaptive energy complementarity between vehicles and UAVs, we define a hierarchical action policy network for UAVs, distinct from that of vehicles. Its upper-layer action enables adaptively switching between task execution and charging behaviors of UAVs, and its lower-layer action selects the vehicles that can assist the charging of UAVs. Further, to leverage the differentiated competitive advantages of vehicles and UAVs under traffic congestion, we design a game consensus mechanism based on the GS (Gale-Shapley) algorithm. In this mechanism, we define incentive factors related to congestion parameters to guide the differentiated task competition between vehicles and UAVs. Based on these incentive factors, we compute the estimated cost on tasks to optimize both cost and efficiency. The experimental results show that, compared to existing methods, the proposed method can reduce the average time to complete tasks by 7.58% and decrease the average energy consumption of agents by 10.05%, thus having advantages in efficiency and energy consumption.