Unmanned aerial vehicle autonomous flight training method based on reinforcement learning and transfer learning

Abstract

The invention discloses an unmanned aerial vehicle autonomous flight training method based on reinforcement learning and transfer learning. The method comprises the following steps of: (1) creating an unmanned aerial vehicle simulator environment; (2) constructing an environment transfer model based on deep learning, and randomly initializing mapping therein; (3) constructing an A3C algorithm of reinforcement learning, and randomly initializing a flight strategy of the A3C algorithm; (4) constructing an environment inverse transfer model based on deep learning; (5) collecting flight data obtained through operating an unmanned aerial vehicle to fly in a real environment by an unmanned aerial vehicle operator and a strategy; (6) updating the environment transfer model based on the real flight data; (7) using and carrying out transfer learning based on action correction, correcting a flight strategy, and executing the corrected strategy in a simulator to obtain simulated flight data; and (8) based on the simulated flight data, updating the flight strategy by using the A3C algorithm, and updating the environment inverse transfer model at the same time, until the strategy converges, and finally obtaining a strategy as an initial flight strategy of the real unmanned aerial vehicle.

Description

technical field [0001] The invention relates to a UAV autonomous flight training method based on reinforcement learning and transfer learning, and belongs to the technical field of UAV autonomous flight control. Background technique [0002] Autonomous flight control of UAVs in diverse, complex, and rapidly changing environments has always been a difficult point in the field of UAV flight control. Traditional flight control manually writes flight control rules, that is, considers all possible situations encountered during the flight of the UAV in advance, and then combines the professional knowledge and experience of experts in the UAV field, through feedback control, writing rules, etc. - Handle the enumerated cases. However, first of all, the writing of rules requires a lot of labor costs; secondly, the interaction between various situations, if some situations or the interaction between situations are not taken into account, it will lead to the failure of flight control;...

AI Technical Summary

Problems solved by technology

[0005] Purpose of the invention: For the problem of autonomous flight control of UAVs, the situation that manual rule control cannot handle complex and changeable environments, and the unavoidable differences between the simulated environment and the real environment for UAV flight strategy training by reinforcement learning algorithms, resulting in The problem that the flight strategy cannot be applied to the real environment, the present invention provides a UAV autonomous flight training method based on reinforcement learning and transfer learning

Images