Self-adaptive speed adjusting method for quad-rotor unmanned aerial vehicle

Abstract

The invention relates to the technical field of unmanned aerial vehicle speed automatic control, in particular to a quad-rotor unmanned aerial vehicle speed self-adaptive adjustment method, which comprises the following steps: continuously performing environment perception and environment evaluation in the flying process of a quad-rotor unmanned aerial vehicle to obtain environment information and obstacle information; inputting the environment information and the obstacle information into a reinforcement learning model, and outputting speed change quantity and direction selection; and calculating the speed and the angular speed of the four-rotor unmanned aerial vehicle based on the speed change quantity and the direction selection, calculating an actual control quantity based on the speed and the angular speed, and sending the actual control quantity to the four-rotor unmanned aerial vehicle to complete the speed self-adaptive adjustment of the four-rotor unmanned aerial vehicle. According to the invention, the unmanned aerial vehicle can automatically adjust the flight speed and direction in different environments through adaptive control, and safe and efficient autonomous navigation is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic control of unmanned aerial vehicle (UAV) speed, and particularly to a method for adaptively adjusting the speed of a quadrotor UAV. Background Art

[0002] With the rapid development of UAV technology, quadrotor UAVs have been widely used in many fields such as military reconnaissance, aerial photography, logistics transportation, and agricultural spraying due to their simple structure, good stability, and strong maneuverability. The flight control system of a quadrotor UAV controls the flight attitude and speed by adjusting the rotational speeds of the four rotors. One of the core tasks of the flight control system is to achieve precise adjustment of the flight speed to ensure that the UAV can maintain stable and accurate flight performance under various flight conditions. However, in the actual application process of quadrotor UAVs, problems such as wind speed changes, air pressure fluctuations, the complexity and variability of flight missions, etc. are often encountered. These factors make the traditional speed control methods often unable to provide flexible and reliable speed adaptive adjustment.

[0003] Currently, the speed control methods of quadrotor UAVs mainly include strategies based on PID controllers, fuzzy control, or adaptive control. These methods can meet the requirements of conventional flight missions to a certain extent, but there are still the following deficiencies when dealing with complex environments and variable missions:

[0004] Insufficiently precise speed adjustment: Existing control methods usually rely on preset speed curves or fixed PID parameters and cannot make precise adjustments according to the environmental factors (such as wind speed, air pressure, flight attitude, etc.) that change in real time during flight. Therefore, in some special flight missions (such as high-speed flight, low-speed hovering, target tracking, etc.), the quadrotor UAV may not be able to maintain an ideal flight state, resulting in unstable flight or mission failure.

[0005] Lag in response speed: Traditional control methods usually have a certain response delay. Especially in complex environments, the speed adjustment cannot reflect the changes during flight in real time. For example, when the wind speed suddenly changes, the traditional control system may not be able to adjust the speed of the UAV in time, resulting in a decrease in flight stability and even a risk of out-of-control.

[0006] Low energy efficiency: In most existing methods, the flight speed is usually adjusted by increasing the rotational speed of the rotors, which will lead to excessive energy consumption. Especially when flying for a long time or performing power-consuming tasks, it may reduce the endurance of the UAV. And the existing methods do not adequately consider the optimization of energy consumption and flight efficiency, and it is often difficult to balance the relationship between flight speed and energy efficiency.

[0007] Poor environmental adaptability: When a quadrotor UAV performs tasks in different environments (such as cities, mountains, seas, etc.), it faces complex flight conditions. Existing speed adjustment methods often do not fully consider the impact of the external environment, resulting in instability of flight performance in different environments. For example, in a strong wind environment, traditional control methods may not be able to respond to wind changes in real time, leading to an increase in the error of flight speed and affecting the quality of task completion.

[0008] Therefore, how to achieve adaptive adjustment of the UAV speed in a complex flight environment, enabling the UAV to adjust its flight speed in real time and accurately, ensuring flight stability, improving energy efficiency, and adapting to changing task requirements has become an urgent problem to be solved in UAV control technology. Summary of the Invention

[0009] The purpose of the present invention is to provide a speed adaptive adjustment method for a quadrotor UAV in view of the above technical problems, enabling the UAV to automatically adjust its flight speed and direction in different environments through adaptive control, and achieving safe and efficient autonomous navigation.

[0010] To achieve the above purpose, the present invention provides the following solution:

[0011] A speed adaptive adjustment method for a quadrotor UAV, comprising:

[0012] Continuously performing environmental perception and environmental assessment during the flight of the quadrotor UAV to obtain environmental information and obstacle information;

[0013] Inputting the environmental information and obstacle information into a reinforcement learning model to output a speed change amount and a direction selection;

[0014] Calculating the speed and angular velocity of the quadrotor UAV based on the speed change amount and direction selection, and calculating the actual control amount based on the speed and angular velocity and sending it to the quadrotor UAV to complete the speed adaptive adjustment of the quadrotor UAV.

[0015] Optionally, performing the environmental perception includes:

[0016] Mounting a sensing module on the quadrotor UAV, the sensing module including a vision sensor and an image processing unit;

[0017] Collecting environmental images through the vision sensor, preprocessing the environmental images and inputting them into the image processing unit for image feature extraction to identify obstacles, wherein the sensing area of the sensing module is:

[0018]

[0019] where ω is the sensing area of the quadrotor UAV, is the forward field of view angle of the quadrotor UAV, and d is the forward field of view distance of the quadrotor UAV.

[0020] Optionally, performing the environmental assessment includes:

[0021] Based on the environmental image and the identified obstacles, perform an environmental assessment using a preset environmental assessment index, where the environmental assessment index includes a field of view brightness measurement index, a field of view clarity measurement index, a passable gap measurement index, and an obstacle density measurement index.

[0022] Optionally, the field of view brightness measurement index is used to quantify the field of view brightness values in different environments; the field of view clarity measurement index is used to calculate the image quality of the environmental image and quantify the clarity; the passable gap measurement index is used to calculate the distance between the quadrotor UAV and the obstacles, as well as the gap size between the obstacles; the obstacle density measurement index is used to segment and count the obstacles in the environmental image and evaluate the density of the obstacles.

[0023] Optionally, inputting the environmental information and the obstacle information into a reinforcement learning model, and the output speed change amount and direction selection include:

[0024] Take the environmental information and the obstacle information as the state input of the reinforcement learning model, perform behavior selection through the Q function combined with the reward function, and obtain the action output, that is, the speed change amount and direction selection of the quadrotor UAV.

[0025] Optionally, the reward function is:

[0026] J = F(a1, a2, a3, loss);

[0027] where J is the comprehensive reward, F(·) is the reward function, a1 is whether the quadrotor UAV reaches the end point, a2 is the energy consumption of the quadrotor UAV, a3 is the distance between the quadrotor UAV and the obstacles, and loss represents the environmental loss rate.

[0028] Optionally, calculating the speed of the quadrotor UAV is:

[0029]

[0030] where and respectively represent the speeds of the UAV in the north, east, and down directions in the world coordinate system, g represents the acceleration due to gravity, m represents the mass of the UAV, u1 represents the total thrust perpendicular to the fuselage direction of the UAV, θ represents the pitch angle, φ represents the roll angle, and ψ represents the yaw angle.

[0031] Optionally, calculating the angular velocity of the quadrotor UAV is:

[0032]

[0033] Among them, τ x is the rolling moment of the propeller about the O b x b axis, and τ y is the pitching moment of the propeller about the O b y b axis, and τ z is the yaw moment of the propeller about the O b z b axis; J RP represents the total moment of inertia of the entire motor rotor and the propeller about the body rotation axis, and p, q, r represent the three components of the angular velocity on the body axis, represents the three components of the updated angular velocity on the body axis, and I xx 、I yy 、I zz respectively represent the inertia tensors of the UAV in the north, east, and down directions in the world coordinate system, and Ω represents the angular velocity vector of the UAV in the world coordinate system.

[0034] The beneficial effects of the present invention are as follows:

[0035] The present invention uses a high-resolution camera as a visual sensor, combined with advanced image processing technology for precise environmental perception, covering a 120-degree field of view in the front, and can identify obstacles in the unobstructed area within 3 meters; a comprehensive environmental assessment quantification index system is proposed, including vision brightness, clarity, passable gap, and obstacle density measurement indicators; a reinforcement learning algorithm is adopted, and a carefully designed reward function is used to make decisions by integrating multiple factors, and the ε-greedy strategy is used to balance exploration and exploitation; a flight control equation is established based on the motion model, and the actual control amount is calculated by combining the model predictive control (MPC) algorithm, which can automatically adjust the flight speed and direction of the UAV in different environments through adaptive control, realizing safe and efficient autonomous navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a schematic diagram of the sensing area of an embodiment of the present invention;

[0038] Figure 2 It is a schematic diagram for calculating the passable gap measurement index of an embodiment of the present invention;

[0039] Figure 3 Flow chart of the obstacle segmentation and counting method according to an embodiment of the present invention;

[0040] Figure 4 Structural framework diagram of the reinforcement learning model according to an embodiment of the present invention;

[0041] Figure 5 Flow chart of the reinforcement learning model training according to an embodiment of the present invention;

[0042] Figure 6 Flow chart of a speed adaptive adjustment method for a quadrotor UAV according to an embodiment of the present invention. Detailed implementation manners

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0044] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0045] This embodiment provides a speed adaptive adjustment method for a quadrotor UAV, as Figure 6 shown, including:

[0046] During the flight of the quadrotor UAV, continuously perform environmental perception and environmental assessment to obtain environmental information and obstacle information;

[0047] Input the environmental information and obstacle information into the reinforcement learning model, and output the speed change amount and direction selection;

[0048] Based on the speed change amount and direction selection, calculate the speed and angular velocity of the quadrotor UAV, and calculate the actual control amount based on the speed and angular velocity and send it to the quadrotor UAV to complete the speed adaptive adjustment of the quadrotor UAV.

[0049] Specifically, this embodiment performs reinforcement learning through a reinforcement learning model constructed by combining a Q function and a reward function, which can achieve adaptive control, enabling the quadrotor UAV to automatically adjust the flight speed and direction in different environments and achieve safe and efficient autonomous navigation.

[0050] Further, performing the environmental perception includes:

[0051] A sensing module is carried on the quadrotor UAV, and the sensing module includes a vision sensor and an image processing unit;

[0052] The environmental image is collected through the vision sensor, and after preprocessing the environmental image, it is input into the image processing unit for image feature extraction to identify obstacles. Among them, the sensing area of the sensing module is:

[0053]

[0054] where ω is the sensing area of the quadrotor UAV, is the forward field of view angle of the quadrotor UAV, and d is the forward field of view distance of the quadrotor UAV.

[0055] Furthermore, performing the environmental assessment includes:

[0056] Based on the environmental image and the identified obstacles, a preset environmental assessment index is used for environmental assessment. Among them, the environmental assessment index includes a field of view brightness measurement index, a field of view clarity measurement index, a passable gap measurement index, and an obstacle density measurement index.

[0057] Furthermore, the field of view brightness measurement index is used to quantify the field of view brightness values in different environments; the field of view clarity measurement index is used to calculate the image quality of the environmental image and quantify the clarity; the passable gap measurement index is used to calculate the distance between the quadrotor UAV and the obstacles, as well as the size of the gaps between the obstacles; the obstacle density measurement index is used to segment and count the obstacles in the environmental image to evaluate the density of the obstacles.

[0058] Furthermore, inputting the environmental information and obstacle information into the reinforcement learning model, and the output speed change amount and direction selection include:

[0059] Taking the environmental information and obstacle information as the state input of the reinforcement learning model, performing behavior selection through the Q function combined with the reward function, and obtaining the action output, that is, the speed change amount and direction selection of the quadrotor UAV.

[0060] Furthermore, the reward function is:

[0061] J = F(a1, a2, a3, loss);

[0062] where J is the comprehensive reward, F(·) is the reward function, a1 is whether the quadrotor UAV reaches the end point, a2 is the energy consumption of the quadrotor UAV, a3 is the distance between the quadrotor UAV and the obstacles, and loss represents the environmental loss rate.

[0063] Furthermore, calculating the speed of the quadrotor UAV is:

[0064]

[0065] Among them, and respectively represent the velocities of the drone in the north, east, and down directions in the world coordinate system, g represents the acceleration due to gravity, m represents the mass of the drone, u1 represents the total thrust perpendicular to the body of the drone, θ represents the pitch angle, φ represents the roll angle, and ψ represents the yaw angle.

[0066] Furthermore, the angular velocity of the quadrotor drone is calculated as:

[0067]

[0068] Among them, τ x is the rolling moment of the propeller about the O b x b axis, τ y is the pitching moment of the propeller about the O b y b axis, τ z is the yawing moment of the propeller about the O b z b axis; J RP represents the total moment of inertia of the entire motor rotor and propeller about the body axis of rotation, p, q, r represent the three components of the angular velocity on the body axis, represents the three components of the updated angular velocity on the body axis, I xx 、I yy 、I zz respectively represent the inertia tensors of the drone in the north, east, and down directions in the world coordinate system, and Ω represents the angular velocity vector of the drone in the world coordinate system.

[0069] Next, in combination with Figures 1-6 a detailed description of a speed adaptive adjustment method for a quadrotor drone proposed in this embodiment is given, which specifically includes:

[0070] Step 1: Sense the environment.

[0071] In the autonomous flight of the quadrotor drone, sensing the environment is the crucial first step. In this embodiment, a high-resolution camera carried by the quadrotor drone is used as a visual sensor, and the front environment is sensed through image processing technology. The sensing system of the quadrotor drone covers a 120-degree field of view in the front, and can identify obstacles in the unobstructed area within 3 meters (drone sensing area ω), as Figure 1 shown. This sensing ability is achieved through the following steps:

[0072]

[0073] 1.1 Image acquisition: The camera of the quadcopter drone captures real-time images in front of the flight at a high frame rate.

[0074] 1.2 Preprocessing: The acquired images are preprocessed such as denoising, grayscale conversion, and binarization to improve the image quality.

[0075] 1.3 Feature extraction and obstacle recognition: Deep learning methods, such as convolutional neural network (CNN), are used to extract features from the preprocessed images and identify potential obstacles.

[0076] Step 2: Environmental evaluation metrics.

[0077] The environmental evaluation metrics proposed in this embodiment aim to quantify the flight environment of the quadcopter drone and provide a basis for subsequent decisions:

[0078] 2.1 Visual field brightness measurement metric: In this embodiment, the average pixel brightness X avg , root mean square pixel height X rms , and perceived brightness X p and other methods are used to quantify the visual field brightness values in different environments respectively, and a comprehensive brightness quantization index B1 is obtained, B1 = B(X avg , X rms , X p ).

[0079] 2.2 Visual field clarity measurement metric: An image quality metric such as peak signal-to-noise ratio is used to quantify the clarity. The peak signal-to-noise ratio is an index based on MSE, and the larger the value, the better the image quality. A visual field clarity index B2 = C(I, K, L) is obtained, where I represents the original image, K represents the degraded image, and L represents the maximum possible pixel value in the image.

[0080] 2.3 Passable gap measurement metric: Based on monocular ranging technology, the distance between the quadcopter drone and the obstacles, as well as the gap size between the obstacles, are calculated. This method first determines the camera focal length F, and then uses the principle of similar triangles to calculate the distance B3 according to the pixel P change of the object at different positions and the actual width W of the object, B3 = D(F, P, W). As Figure 2 shown.

[0081] 2.4 Obstacle density measurement metric: A deep learning image segmentation network, such as SegNet, is used to segment and count the obstacles in the image, evaluate the density of the obstacles, and obtain the index B4 = U(G), where G represents the input image. As Figure 3 shown.

[0082] Step 3: Construct and train a reinforcement learning model for decision-making.

[0083] 3.1 Basic Structure of the Reinforcement Learning Model:

[0084] State: It includes the real-time coordinate position of the quadrotor UAV, the positions of obstacles within the UAV's sensing area, and environmental information (including field of view brightness, field of view clarity, available clearance, and obstacle density).

[0085] Action: The actions that the quadrotor UAV can perform include adjusting the magnitude of the flight speed and changing the flight direction. Specifically, it includes the components of the UAV's speed and angular velocity on each axis of the world coordinate system and the body coordinate system obtained from its flight control equations.

[0086] Reward Function: Based on factors such as whether the UAV reaches the end point a1, energy consumption a2, distance from obstacles a3, and environmental loss rate loss, where the environmental loss rate is obtained according to environmental evaluation indicators, design a comprehensive reward function J, where J = F(a1, a2, a3, loss).

[0087] Agent: Make decisions according to the Q function and use the ε-greedy strategy to generate an action.

[0088] The basic structure framework is as Figure 4 shown.

[0089] 3.2 Training Process of the Reinforcement Learning Model:

[0090] During the reinforcement learning training process of the quadrotor UAV, multiple loops are carried out. Each loop represents a complete flight mission. A complete flight mission means starting from the starting point and reaching the end point or colliding. The specific steps are as follows:

[0091] 3.2.1: Initialization. Use a deep reinforcement learning algorithm, such as Deep Q Network, to initialize the parameters of the value function Q(s, a) of the DQN network and set a target Q network Q tar , whose parameters are periodically copied from the Q network and used to calculate the target Q value to stabilize the training process.

[0092] 3.2.2: State Input. Randomly generate the obstacle density and positions within the map, start a new loop, and at each time point (step), obtain the image within the UAV's sensing area, and use a convolutional neural network to identify the obstacles to obtain information such as the obstacle density and positions as the state s.

[0093] 3.2.3: Action Selection. The DQN algorithm predicts the value function through a neural network. For example, the Q-function in DQN, i.e., Q(s,a), represents the expected return of taking action a in state s. The drone agent selects an action a according to the current Q-function. To balance exploration and exploitation in reinforcement learning, the ε-greedy strategy is adopted to select actions. That is, in the current state, a random number is generated. If the random number is less than ε, the agent will randomly select an action for exploration; otherwise, if the random number is greater than or equal to ε, the agent will select the action that can maximize the current Q-value for exploitation. Under this strategy, the probability of exploration is ε, and the probability of exploitation is 1-ε.

[0094] 3.2.4: Execute the Action and Obtain the Reward. The drone executes the selected action a and interacts with the environment. According to action a and the environmental state s' after executing the action, a reward function J is designed to calculate the reward value j.

[0095] 3.2.5: State Transition. After executing the action, the environmental state changes, and a new state s' is generated through the state transition function q(s'|s,a). The drone enters the new state s'.

[0096] 3.2.6: Store the Experience: Store the current state s, action, reward J, and the next state s' in the experience replay pool.

[0097] 3.2.7: Learning and Update. At the end of each loop, a batch of experiences is randomly sampled from the experience replay pool for learning. For each sample, by comparing the actually obtained reward j and Q tar calculate the target Q-value, Q = H(Q tar , s', a', j). Calculate the loss function and perform backpropagation based on an optimization algorithm, such as the Adam algorithm, to update the deep neural network part in deep reinforcement learning, such as the deep neural network that calculates the Q-value in DQN.

[0098] 3.2.8: Target Network Update. Regularly copy the parameters of the Q network to the target network.

[0099] 3.2.9: End. When the drone collides with an obstacle, the current loop ends; otherwise, when the drone successfully reaches the destination, the current loop ends; otherwise, when the flight time exceeds the preset value, the current loop ends.

[0100] Through a large number of iterative trainings of the above processes, the Q-function is continuously updated, and the agent selects actions according to the updated Q-value, so that it can learn a better strategy and can achieve speed adjustment and direction adjustment of autonomous flight in a complex environment.

[0101] The above iterative process is as Figure 5 shown.

[0102] 3.3 Design of the Reward Function for the Reinforcement Learning Model:

[0103] The reward function is designed as:

[0104] J = F(a1, a2, a3, loss);

[0105] Where a1, a2, and a3 are whether the UAV reaches the end point, energy consumption, and distance from the obstacle respectively, and loss represents the environmental loss rate. Reaching the end point is a positive reward, while energy consumption, distance from the obstacle, and environmental loss rate are negative rewards.

[0106] The energy consumption a2 is a function of flight time t and flight speed, expressed as a2 = A(t, v).

[0107] The distance from the obstacle a3 = P(x1, y1, z1, x2, y2, z2), where x1, y1, z1 are the UAV position parameters and x2, y2, z2 are the obstacle position parameters.

[0108] The environmental loss rate loss is a comprehensive index that integrates the effects of field of view brightness, field of view clarity, passable gap (all positive reward factors) and obstacle density (negative reward factor). Its specific expression is:

[0109] loss = α * A + β * b + γ * C + δ * D;

[0110] In the formula, α, β, γ, δ are weight factors obtained based on experiments or experience. A, B, C, D represent the four indexes of field of view brightness, field of view clarity, passable gap, and obstacle density respectively.

[0111] 3.4 Output of the Reinforcement Learning Model: The change in the UAV's speed and direction selection.

[0112] 3.5 Application of the Reinforcement Learning Model: Input the environmental information and obstacle information obtained in Step 1 into the trained reinforcement learning model, and output the change in the UAV's speed and direction selection, that is, the decision result given by the reinforcement learning model, and enter Step Four.

[0113] Step Four: Execute the decision.

[0114] 4.1 Kinematic Equation. Based on the four-rotor UAV motion model, a flight control equation is established to describe the dynamic response of the UAV when receiving control inputs, including the speed of the four-rotor UAV in the x, y, z axes and the angular velocity of the four-rotor UAV in the x, y, z axes.

[0115] 4.2 Calculation of actual control quantity. According to the decision result, the expected speed and direction adjustment quantities are given. Advanced control algorithms such as model predictive control (MPC) are used to calculate the actual control quantities such as propeller rotation speed, inclination angle, etc., and the control quantities are sent to the UAV to achieve precise control of the UAV.

[0116] Through this adaptive control method, the UAV can automatically adjust its flight speed and direction in different environments to achieve safe and efficient autonomous navigation.

[0117] The speed calculation formulas of the quadrotor UAV in the x, y, and z-axis directions in the above 4.1 are as follows:

[0118]

[0119] Among them, and respectively represent the speeds of the UAV in the north, east, and downward directions in the world coordinate system, g represents the acceleration due to gravity, m represents the mass of the UAV, u1 represents the total thrust perpendicular to the UAV fuselage direction, θ represents the pitch angle, φ represents the roll angle, and ψ represents the yaw angle.

[0120] The angular velocity calculation formulas of the quadrotor UAV in the x, y, and z-axis directions are as follows:

[0121]

[0122] Among them, τ x is the rolling moment of the propeller around the O b x b axis, τ y is the pitching moment of the propeller around the O b y b axis, τ z is the yawing moment of the propeller around the O b z b axis; J RP represents the total moment of inertia of the entire motor rotor and propeller around the body rotation axis, p, q, and r represent the three components of the angular velocity on the body axis, represents the three components of the updated angular velocity on the body axis, I xx 、I yy 、I zz respectively represent the inertia tensors of the UAV in the north, east, and downward directions in the world coordinate system, and Ω represents the angular velocity vector of the UAV in the world coordinate system.

[0123] The above-described embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for adaptively adjusting the speed of a quad-rotor drone, characterized in that: include: During the flight of the quadrotor drone, the drone continuously performs environmental perception and environmental assessment to obtain environmental and obstacle information; Input the environmental information and obstacle information into a reinforcement learning model, and output a speed change and a direction selection; The speed and angular velocity of the quad-rotor drone are calculated based on the speed change and direction selection, and the actual control amount is calculated based on the speed and angular velocity and sent to the quad-rotor drone, thereby completing the speed adaptive adjustment of the quad-rotor drone.

2. The method for adaptively adjusting the speed of a quad-rotor drone according to claim 1, characterized in that: The environmental perception includes: A perception module is mounted on the quadrotor drone, wherein the perception module includes a visual sensor and an image processing unit; The visual sensor collects environmental images, pre-processes the environmental images, and then inputs them into the image processing unit for image feature extraction to identify obstacles, wherein the perception area of ​​the perception module is: Among them, ω is the sensing area of ​​the quadrotor drone, is the front field of view angle of the quadrotor drone, and d is the front field of view distance of the quadrotor drone.

3. The method for adaptively adjusting the speed of a quad-rotor drone according to claim 2, characterized in that: Conducting the environmental assessment includes: Based on the environmental image and the identified obstacles, an environmental assessment is performed using preset environmental assessment indicators, wherein the environmental assessment indicators include a field of view brightness measurement indicator, a field of view clarity measurement indicator, a passable gap measurement indicator, and an obstacle density measurement indicator.

4. The method for adaptively adjusting the speed of a quad-rotor drone according to claim 3, characterized in that: The field of view brightness measurement index is used to quantify the field of view brightness value in different environments; the field of view clarity measurement index is used to calculate the image quality of the environment image and quantify the clarity; the clearance measurement index is used to calculate the distance between the quadcopter drone and the obstacle, and the size of the clearance between obstacles; The obstacle density measurement index is used to segment and count obstacles in the environment image and evaluate the density of obstacles.

5. The method for adaptively adjusting the speed of a quad-rotor drone according to claim 1, characterized in that: Inputting the environmental information and obstacle information into the reinforcement learning model, and outputting the speed change and direction selection include: The environmental information and obstacle information are used as the state input of the reinforcement learning model, and behavior selection is performed through the Q function combined with the reward function to obtain the action output, that is, the speed change and direction selection of the quadrotor drone.

6. The method for adaptively adjusting the speed of a quad-rotor drone according to claim 5, characterized in that: The reward function is: J = F(a1, a2, a3, loss); Among them, J is the comprehensive reward, F(·) is the reward function, a1 is whether the quadrotor drone reaches the destination, a2 is the energy consumption of the quadrotor drone, a3 is the distance between the quadrotor drone and the obstacle, and loss represents the environmental loss rate.

7. The method for adaptively adjusting the speed of a quad-rotor drone according to claim 1, characterized in that: The speed of the quadrotor drone is calculated as: in, and They represent the velocities of the UAV in the north, east and down directions in the world coordinate system, respectively; g represents the acceleration of gravity; m represents the mass of the UAV; u1 represents the total thrust perpendicular to the direction of the UAV body; θ represents the pitch angle; φ represents the roll angle; and ψ represents the yaw angle.

8. The method for adaptively adjusting the speed of a quad-rotor drone according to claim 1, characterized in that: The angular velocity of the quadrotor drone is calculated as: Among them, τ x For the propeller around O b x b Rolling moment about the axis, τ y For the propeller around O b y b Pitch moment about the axis, τ z For the propeller around O b z b Yaw moment of the axis; J RP It represents the total moment of inertia of the entire motor rotor and propeller around the body axis, p, q, r represent the three components of angular velocity on the body axis, Represents the three components of the updated angular velocity on the body axis, I xx ,I yy ,I zz They represent the inertia tensors of the UAV in the north, east and down directions in the world coordinate system respectively, and Ω represents the angular velocity vector of the UAV in the world coordinate system.

Images