Urban multi-vehicle and multi-unmanned aerial vehicle dynamic distribution cooperative control method based on double-agent layered architecture

The invention discloses a multi-vehicle and multi-unmanned aerial vehicle cooperative distribution path optimization method based on double-agent hierarchical reinforcement learning, relates to path optimization, and aims to solve the problems of difficult vehicle and unmanned aerial vehicle cooperative scheduling, complex multi-agent joint decision and the like in urban end distribution. According to the invention, a double-agent layered architecture is designed and is responsible for collaborative decision-making of a vehicle group and an unmanned aerial vehicle group. The method comprises the following steps: firstly, modeling a distribution system into a collaborative decision problem of two heterogeneous agents; secondly, defining a Markov decision process, and establishing a collaborative distribution path optimization model by taking minimization of total distribution time as a target; then, designing a reward function, and guiding learning of a collaborative distribution strategy; and finally, designing a network structure and a training algorithm, realizing a training framework, and solving a multi-vehicle and multi-unmanned aerial vehicle collaborative distribution path optimization model. According to the method, the heterogeneous characteristics of the vehicle and the unmanned aerial vehicle can be processed through the double-agent layered architecture, and the path planning efficiency and the distribution timeliness of urban end distribution are improved.