A Feedforward Quadrotor Disturbance Rejection Method for Visually Identified Gust Sources Based on Transfer Reinforcement Learning

Fulin Song,Zhan Li,Xinghu Yu

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS(2023)

Cited 0|Views6
No score
Abstract
With the rapid development of urbanization, quadrotors are flying more and more frequently between high-density buildings. In urban areas, there are many gust sources with obvious visual characteristics, such as ventilation outlet of buildings or subways, outdoor unit of heating ventilating and air conditioning, or other artificial gust sources, which can cause sudden gust disturbances to the passing quadrotors. Aiming at reducing the effect of visually identified gust disturbances, this article presents a quadrotor feedforward compensator framework based on transfer reinforcement learning (TRL). In this structure, the discriminative model prediction (DiMP) algorithm is used to extract the visual features of the gust source and achieve tracking of such sources. Then, the TRL algorithm is used to transfer the obtained tracking network to a compensation network, which completes a DiMP-TRL structure. The output of the trained DiMP-TRL compensation network is added to the quadrotor controller as a feedforward part, which will help generate preventive action when encountering the identified gust disturbance. Only a front-facing camera is needed to collect gust information; no additional anemometer or accurate wind model is required. The ultimately uniformly bounding of the quadrotor system can be guaranteed using the Lyapunov stability criteria. The proposed method is compared with a cascade proportional-integral-differential controller and a novel feedback-compensation-based method called dual command filter in a realistic simulation environment built by a physics engine. Two sets of results demonstrate that the proposed method can reject gust disturbance more efficiently.
More
Translated text
Key words
Quadrotors,Cameras,Feedforward systems,Reinforcement learning,Fluid flow measurement,Atmospheric modeling,Deep learning,Feedforward compensator,gust source perception,quadrotor disturbance rejection,transfer reinforcement learning (TRL)
AI Read Science
Must-Reading Tree
Example
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined