Autonomous navigation unmanned aerial vehicle power optimization method based on deep reinforcement learning

Abstract

The invention discloses a power optimization method based on deep reinforcement learning in an unmanned aerial vehicle driving application. The method achieves the low-delay and high-energy-efficiency autonomous navigation task execution through the dynamic configuration of the calculation scale of a convolutional neural network in combination with the state features of an environment where an unmanned aerial vehicle is located. The method comprises the following steps of firstly, designing and training a deep neural network capable of receiving input layers of different sizes, and calculating the control direction and speed of the unmanned aerial vehicle according to the image input of a front camera, and then, by utilizing reinforcement learning, according to the environment complexity, the obstacle hybrid factor and the historical action vector of the current time block, deducing the optimal neural network configuration of the calculation power consumption suitable for the current environment, so that the utilization rate of the calculation energy consumption of the unmanned aerial vehicle equipment is improved, and the endurance time of the autonomous navigation unmanned aerial vehicle is prolonged.

Description

technical field [0001] The present invention relates to the technical fields of edge computing, deep learning, reinforcement learning and automatic driving, and more specifically, relates to a power optimization method for autonomous navigation UAV based on deep reinforcement learning. Background technique [0002] In recent years, the autonomous navigation capability of drones has attracted widespread attention from the robotics community. The advantages of autonomous navigation drones, such as easy deployment, agility and mobility, have been widely used in many fields, such as fire detection , precision agriculture, express delivery and security inspections, etc. The traditional way to realize self-navigation is to use SLAM algorithm, which includes two processes of perception of a given map and calculation of control commands. However, separating the perception process from the control process not only hinders the positive feedback between the perception process and the ...

AI Technical Summary

Problems solved by technology

However, a large amount of data (such as image and video data) is transmitted to a remote cloud server through a long wide area network, resulting in a large end-to-end delay and energy consumption on the drone, and due to the mobility of the drone, its Performance is greatly affected by bandwidth fluctuations and cannot provide a stable performance

The second is to deploy the neural network model directly on the UAV's local computing device to achieve high reliability and low-latency reasoning. However, because the deep learning model usually requires large computing and storage overhead, it cannot provide a good power consumption performance.

Images