Fixed-wing unmanned aerial vehicle autonomous control cooperation strategy training method

Abstract

The invention discloses a fixed-wing unmanned aerial vehicle autonomous control cooperation strategy training method, which comprises the following steps of (1) constructing a fixed-wing unmanned aerial vehicle control simulation environment Es based on dynamics, collecting real trajectory data of a pilot controlling an unmanned aerial vehicle, and learning in a supervised learning mode to obtainan unmanned aerial vehicle flight control strategy, (2) constructing a simplified abstract environment Ea from which flight control is stripped, creating two groups of unmanned aerial vehicle groups for grouping confrontation, and learning by using an APEX_QMIX algorithm to obtain a cooperation strategy, (3) combining the flight control strategy and the cooperation strategy in a layered reinforcement learning mode, and learning in a simulation environment Es to obtain a fusion strategy, and (3) migrating to a real environment. The method is of great significance in a real scene and has the advantages of being good in generalization, low in cost, high in robustness and the like.

Description

technical field [0001] The invention relates to a fixed-wing unmanned aerial vehicle autonomous control cooperation strategy training method based on layered reinforcement learning and multi-agent reinforcement learning, and the technical field of unmanned aerial vehicle autonomous control cooperation strategy. Background technique [0002] For the traditional autonomous control and cooperation strategy of fixed-wing UAV, it mainly adopts the method of automatic control, artificial modeling, and formulation of strategy. Rely on the development of flight rules by experts in the relevant field. The cost is high and due to the frequent scene changes in the complex and changing environment, there are a large number of situations that are not considered in the flight rules. Therefore, flight rules are usually unable to deal with complex and changing environments, and their capabilities are low. [0003] Recently, with the vigorous development of machine learning technology, rei...

AI Technical Summary

Problems solved by technology

However, simple reinforcement learning also has its limitations. The space for exploration and learning is too large, and the effect depends heavily on parameter tuning tricks, making training difficult.

Images