Unmanned aerial vehicle attitude training method and device

Abstract

The embodiment of the invention relates to an unmanned aerial vehicle attitude training method and device. The unmanned aerial vehicle attitude training method comprises the following steps: obtainingsensor information, panel contact information of an unmanned aerial vehicle in a set scene, and obtaining actual attitude information of the unmanned aerial vehicle in the set scene based on controlinformation; determining reward information based on the panel contact information, the actual attitude information, and the target attitude information; training a reinforced learning deep neural network according to the principle of minimizing loss based on the reward information, so that the actual attitude responds quickly to the target attitude, and a trained deep neural network model is obtained to control the flight attitude of the unmanned aerial vehicle through environmental state information. The power unit of the unmanned aerial vehicle is directly controlled through enhancing the learning model, and end-to-end direct learning and control is realized; through the reinforced learning, the unmanned aerial vehicle has the ability of intelligent decision control, which can realizesautomatic attitude control, and can be applied to more complex real scenes.

Description

technical field [0001] The embodiments of the present invention relate to the field of artificial intelligence, and in particular to a method and device for training an attitude of an unmanned aerial vehicle. Background technique [0002] Automatic control systems usually consist of several control loops. The "inner loop" is used for low-level control, such as stability control, and the "outer loop" is used for task-level control, such as path control. For general UAV control, it is mainly composed of multiple control loops combined between layers, and each control loop is controlled by a PID controller. But generally speaking, multi-loop control requires manual parameter setting. When faced with changes in environmental parameters, the multi-loop controller may become unstable. At the same time, parameter setting requires extensive knowledge in related fields, making adjustment difficult. Integrated and intelligent UAV control is becoming more and more important, and the e...

AI Technical Summary

Problems solved by technology

But generally speaking, multi-loop control requires manual parameter setting. When faced with changes in environmental parameters, the multi-loop controller may become unstable. At the same time, parameter setting requires extensive knowledge in related fields, making adjustment difficult.

Integrated and intelligent UAV control is becoming more and more important, and the emerging reinforcement learning (RL) method has a good performance in the field of robot control, especially in task decision-making, but there is no real Most of the relevant physical training devices and methods for the "inner loop" control of UAVs are also in the simulation stage, and there is a model migration problem between the simulation and the real environment, that is, the "reality gap" problem

Images