A robust landing method for drones on mobile platforms

Abstract

This invention relates to the field of unmanned aerial vehicle (UAV) technology, and more particularly to a robust landing method for UAVs on a mobile platform, comprising the following steps: Step S1: Constructing a vision-based landing system, which consists of an underactuated quadcopter UAV and an unknown mobile platform. Step S2: The UAV is equipped with a downward-looking camera, and pixel points are acquired as image features through corner detection. Image errors are constructed based on a predetermined performance control method to ensure that the pixel points remain within the camera's field of view during the UAV landing process. Step S3: An image-based landing control law is designed using an asymmetric time-varying Lyapunov function. Step S4: The thrust and torque of the quadcopter are calculated. This invention can complete the landing task on a mobile platform while ensuring the field of view constraint, and has the advantages of predetermined steady-state and transient landing performance, computational simplicity, and strong robustness.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, specifically to a robust landing method for UAVs on a mobile platform. Background Technology

[0002] Due to increasingly precise and reliable technologies in sensing, computing, and communication systems, the automation of unmanned aerial vehicles (UAVs) is an important research area, and public interest in UAVs is growing daily. Although UAVs have been around for decades, their functionality and effectiveness have improved, giving them characteristics such as hovering, vertical takeoff and landing, and low cost. However, they are still limited by battery technology, making it difficult for them to perform missions for extended periods. In recent years, the combined use of UAVs and unmanned vehicles (UAVs) has become a major focus. UAVs undertake mobile tasks, and when tasks such as geographic reconnaissance, power grid inspection, and aerial surveillance are required, UAVs can take off directly from the UAVs and recharge on the UAVs, greatly extending their usage time and range, and enhancing their ability to autonomously perform tasks in complex environments. However, numerous accidents during landing indicate that without solving the problem of autonomous landing capabilities for UAVs on UAVs, they cannot effectively cooperate with UAVs. Therefore, this paper proposes a robust landing method for UAVs on mobile platforms to address these issues. Summary of the Invention

[0003] The purpose of this invention is to provide a robust landing method for unmanned aerial vehicles (UAVs) on a mobile platform, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A robust landing method for unmanned aerial vehicles (UAVs) on a mobile platform includes the following steps:

[0006] Step S1: Construct a vision-based landing system consisting of an underactuated quadcopter drone and an unknown mobile platform;

[0007] Step S2: The drone is equipped with a downward-looking camera, which uses corner detection to obtain 4 pixels as image features, and constructs image error based on a predetermined performance control method;

[0008] Step S3: Design an image-based landing control law to ensure that the pixels remain within the camera's field of view during the drone's landing process;

[0009] Step S4: Calculate the thrust and torque of the quadrotor using inversion control technology and asymmetric time-varying Lyapunov functions to drive the quadrotor to land autonomously.

[0010] As a preferred embodiment, step S1 specifically includes the following steps:

[0011] Step S101: The vision-based landing system consists of an underactuated quadcopter drone and an unknown mobile platform;

[0012] Step S102: The dynamic model of the underactuated quadcopter UAV is shown below:

[0013]

[0014]

[0015]

[0016]

[0017] F = fRe3

[0018] Among them, it is defined Here, J represents the position and velocity of the quadrotor in Cartesian coordinates, ω is the angular velocity of the quadrotor, m is its mass, J = diag(J1, J2, J3) represents the inertia matrix, f and Γ are the thrust and torque vectors required for quadrotor flight, respectively, and e3 = [0, 0, 1]. T Representing a unit vector, the rotation matrix R is determined by the roll angle. The pitch angle θ and yaw angle ψ together represent the rotational transformation from the quadrotor's body coordinate system to the world coordinate system, denoted by sk(·). Represents a skew-symmetric matrix;

[0019] Step S103: The mobile platform model is represented as follows:

[0020]

[0021]

[0022]

[0023] Where, v m , These represent the linear velocity and angular velocity of the mobile platform, respectively.

[0024] As a preferred embodiment, step S2 specifically includes the following steps:

[0025] Step S201: A monocular camera is mounted beneath the quadcopter drone, with its optical axis pointing towards the ground. A corner detection algorithm is used to capture an image point. Using a standard pinhole camera model, the 3D point Q = [X,Y,Z]T in the camera coordinate system is mapped to the image coordinate q = [u,v]. T The changes are represented as follows;

[0026]

[0027] Where, α u >0,α v >0 represents the scaling factor in the horizontal and vertical directions, respectively, (u0, v0). T These are the coordinates of the principal point;

[0028] The normalized image coordinates m of each feature point can be obtained from the mobile platform model. i =[x i ,y i ] T for Where i = 1, 2, 3, 4;

[0029] Step S202: The monocular camera has a fixed field of view. The image-based controller can only operate stably when the feature point remains within its field of view. Therefore, the image coordinates of the feature point need to ensure the following visibility constraints:

[0030]

[0031] in, It can be obtained from the formula in step S201:

[0032]

[0033]

[0034] Taking the derivative of the formula in step S201, we obtain the following expression, which correlates the normalized image coordinates with the velocity of the quadcopter / mobile platform:

[0035] s i =h i (V m -V)

[0036] Among them, S i =[x i ,y i ] T V = [v T ,w T ] T , and

[0037] Step S203: Define the normalized image error as... To ensure that image errors do not exceed the field of view, we construct a smoothly decreasing performance function to constrain the motion of feature points: Where, φ i (0), l i, Since all numbers are positive, we now define the following coordinate transformation:

[0038]

[0039]

[0040]

[0041] Therefore, by specifying the upper and lower bounds of the error, the normalized error is always kept within the time-varying, asymmetric upper and lower bounds. therefore:

[0042] Similarly, we obtain

[0043]

[0044]

[0045]

[0046] Step S204: From the previous definition, the transformed image error is obtained as follows:

[0047]

[0048] From step S202, we can see that the purpose of constructing the transformed error is not only to ensure that the error is within the time-varying upper and lower bound curves we set when converging, but also to ensure that the feature points are always within the field of view of the camera by selecting appropriate upper and lower bounds. Furthermore, when the error approaches zero, the transformed error also approaches zero.

[0049] As a preferred embodiment, step S3 specifically includes the following steps:

[0050] Step S301: Since a single feature point can only control two degrees of freedom, we choose four feature points to control the six degrees of freedom velocity of the quadcopter. Let's define a global image feature vector S = [S1, S2, S3, S4]. T Therefore, the image kinematics is represented as in, It is a stack of Jacobian matrices of all features;

[0051] Step S302: We design the image-based control law as V = K s H + ε E +V m , where ε E =[ε x,1 ,ε y,1 ,ε x,2,ε y,2 ,ε x,3 ,ε y,3 ,ε x,4 ,ε y,4 ] T H + It is represented as the generalized inverse of H.

[0052] As a preferred embodiment, S4 specifically includes the following steps:

[0053] Step S401: Based on the fundamental idea of ​​inversion control technology, the complex quadcopter dynamics are decomposed into two subsystems: velocity and force. Based on the calculated velocity, an asymmetric time-varying Lyapunov function and intermediate virtual control variables are designed for the force subsystem to calculate the thrust and torque. First, the velocity error is defined as: in, It is the intermediate virtual control quantity in the design;

[0054] Step S402: As can be seen from step S203, according to An asymmetric time-varying Lyapunov function is constructed as follows:

[0055] in,

[0056] right Find the differential and get

[0057]

[0058] Define a control vector

[0059]

[0060] Based on the dynamic equations of the quadrotor, the thrust is designed as follows:

[0061] f = F T (Re3)

[0062] Where F = -K v D v e v -ge3,K v >0 is a constant;

[0063] Step S403: Because Recalling the previous step, an asymmetric time-varying Lyapunov function is defined and constructed as follows:

[0064]

[0065] Define a control vector

[0066]

[0067] Based on the dynamic equations of the quadrotor, the design torque is as follows:

[0068] Γ=-K ω J T D ω e ω +ω×Jω

[0069] Among them, K a >0 is a constant.

[0070] The calculated torque and thrust are simple to calculate, yet can guarantee the desired steady-state and transient landing performance of the quadcopter UAV. It also solves the inherent limitations of the camera's field of view. When the initial feature points are all within the camera's field of view, by selecting appropriate upper and lower bounds, it can ensure that the feature points are always constrained within the camera's field of view.

[0071] As can be seen from the technical solution provided by the present invention above, the robust landing method for UAVs on mobile platforms provided by the present invention has the following advantages:

[0072] 1. Feature points are always within the field of view. By constructing a predetermined performance function, the convergence of the normalized image error is constrained to an asymmetric upper and lower bound that converges over time, so as not to violate the camera's field of view constraints during landing.

[0073] 2. The proposed landing scheme has advantages such as strong robustness, low complexity, and accurate landing. Compared with the traditional GPS-based landing scheme, our scheme is more robust, meets the requirements for indoor and outdoor landing, and the landing accuracy far exceeds that of GPS.

[0074] 3. It has controllable transient and steady-state responses. Whether it is characteristic error or velocity error, the overshoot, convergence time and steady-state value during the error convergence process are all controllable. Attached Figure Description

[0075] Figure 1 This is a flowchart illustrating the robust landing method for unmanned aerial vehicles on a mobile platform according to the present invention.

[0076] Figure 2 This is a quadcopter perspective camera model diagram of a robust landing method for unmanned aerial vehicles on a mobile platform according to the present invention.

[0077] Figure 3 This is an algorithm flowchart of a robust landing method for unmanned aerial vehicles on a mobile platform according to the present invention. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0079] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0080] like Figure 1 As shown, this embodiment of the invention provides a robust landing method for unmanned aerial vehicles (UAVs) on a mobile platform, comprising the following steps:

[0081] Step S1: Construct a vision-based landing system consisting of an underactuated quadcopter drone and an unknown mobile platform;

[0082] Step S2: The drone is equipped with a downward-looking camera, which uses corner detection to obtain 4 pixels as image features, and constructs image error based on a predetermined performance control method;

[0083] Step S3: Design an image-based landing control law to ensure that the pixels remain within the camera's field of view during the drone's landing process;

[0084] Step S4: Calculate the thrust and torque of the quadrotor using inversion control technology and asymmetric time-varying Lyapunov functions to drive the quadrotor to land autonomously.

[0085] In this embodiment, step S1 specifically includes the following steps:

[0086] Step S101: The vision-based landing system consists of an underactuated quadcopter drone and an unknown mobile platform;

[0087] Step S102: The dynamic model of the underactuated quadcopter UAV is shown below:

[0088]

[0089]

[0090]

[0091]

[0092] F = fRe3

[0093] Among them, it is defined Here, J represents the position and velocity of the quadrotor in Cartesian coordinates, ω is the angular velocity of the quadrotor, m is its mass, J = diag(J1, J2, J3) represents the inertia matrix, f and Γ are the thrust and torque vectors required for quadrotor flight, respectively, and e3 = [0, 0, 1]. T Representing a unit vector, the rotation matrix R is determined by the roll angle. The pitch angle θ and yaw angle ψ together represent the rotational transformation from the quadrotor's body coordinate system to the world coordinate system, denoted by sk(·). Represents a skew-symmetric matrix;

[0094] Step S103: The mobile platform model is represented as follows:

[0095]

[0096]

[0097]

[0098] Where, v m , These represent the linear velocity and angular velocity of the mobile platform, respectively.

[0099] In this embodiment, step S2 specifically includes the following steps:

[0100] Step S201: A monocular camera is mounted beneath the quadcopter drone, with its optical axis pointing towards the ground. A corner detection algorithm is used to capture an image point. Using a standard pinhole camera model, the 3D point Q = [X,Y,Z]T in the camera coordinate system is mapped to the image coordinate q = [u,v]. T The changes are represented as follows;

[0101]

[0102] Where, α u >0,α v >0 represents the scaling factor in the horizontal and vertical directions, respectively, (u0, v0). T These are the coordinates of the principal point;

[0103] The normalized image coordinates m of each feature point can be obtained from the mobile platform model. i =[x i ,y i ] T for Where i = 1, 2, 3, 4;

[0104] Step S202: The monocular camera has a fixed field of view. The image-based controller can only operate stably when the feature point remains within its field of view. Therefore, the image coordinates of the feature point need to ensure the following visibility constraints:

[0105]

[0106] in, It can be obtained from the formula in step S201:

[0107]

[0108]

[0109] Taking the derivative of the formula in step S201, we obtain the following expression, which correlates the normalized image coordinates with the velocity of the quadcopter / mobile platform:

[0110] s i =h i (V m -V)

[0111] Among them, S i =[x i ,y i ] T V = [v T ,w T ] T , and

[0112] Step S203: Define the normalized image error as... To ensure that image errors do not exceed the field of view, we construct a smoothly decreasing performance function to constrain the motion of feature points: Where, φ i (0), l i , Since all numbers are positive, we now define the following coordinate transformation:

[0113]

[0114]

[0115]

[0116] Therefore, by specifying the upper and lower bounds of the error, the normalized error is always kept within the time-varying, asymmetric upper and lower bounds. therefore:

[0117] Similarly, we obtain

[0118]

[0119]

[0120]

[0121] Step S204: From the previous definition, the transformed image error is obtained as follows:

[0122]

[0123] From step S202, we can see that the purpose of constructing the transformed error is not only to ensure that the error is within the time-varying upper and lower bound curves we set when converging, but also to ensure that the feature points are always within the field of view of the camera by selecting appropriate upper and lower bounds. Furthermore, when the error approaches zero, the transformed error also approaches zero.

[0124] In this embodiment, step S3 specifically includes the following steps:

[0125] Step S301: Since a single feature point can only control two degrees of freedom, we choose four feature points to control the six degrees of freedom velocity of the quadcopter. Let's define a global image feature vector S = [S1, S2, S3, S4]. T Therefore, the image kinematics is represented as in, It is a stack of Jacobian matrices of all features;

[0126] Step S302: We design the image-based control law as V = K s H + ε E +V m , where ε E =[ε x,1 ,ε y,1 ,ε x,2 ,ε y,2 ,ε x,3 ,ε y,3 ,ε x,4 ,ε y,4 ] T H + It is represented as the generalized inverse of H.

[0127] In this embodiment, S4 specifically includes the following steps:

[0128] Step S401: Based on the fundamental idea of ​​inversion control technology, the complex quadcopter dynamics are decomposed into two subsystems: velocity and force. Based on the calculated velocity, an asymmetric time-varying Lyapunov function and intermediate virtual control variables are designed for the force subsystem to calculate the thrust and torque. First, the velocity error is defined as: in, It is the intermediate virtual control quantity in the design;

[0129] Step S402: As can be seen from step S203, according to An asymmetric time-varying Lyapunov function is constructed as follows:

[0130] in,

[0131] right Find the differential and get

[0132]

[0133] Define a control vector

[0134]

[0135] Based on the dynamic equations of the quadrotor, the thrust is designed as follows:

[0136] f = F T (Re3)

[0137] Where F = -K v D v e v -ge3,K v >0 is a constant;

[0138] Step S403: Because Recalling the previous step, an asymmetric time-varying Lyapunov function is defined and constructed as follows:

[0139]

[0140] Define a control vector

[0141]

[0142] Based on the dynamic equations of the quadcopter, the design torque is as follows:

[0143] Γ=-K ω J T D ω e ω +ω×Jω

[0144] Where, Ka >0 is a constant.

[0145] In this invention example, the quadcopter drone uses only a simple vision kit, including an IMU and a downward-looking camera, to acquire the drone's linear / angular velocity and the quadcopter's attitude angle. A QR code is attached to the mobile platform, and four feature points in the image are obtained through a corner detection algorithm.

[0146] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A robust landing method for unmanned aerial vehicles (UAVs) on a mobile platform, characterized in that: Includes the following steps: Step S1: Construct a vision-based landing system consisting of an underactuated quadcopter drone and an unknown mobile platform; Step S2: The drone is equipped with a downward-looking camera. Corner detection is used to acquire 4 pixels as image features. Image error is constructed based on a predetermined performance control method, specifically including: Step S201: A monocular camera is mounted beneath the quadcopter drone, with its optical axis pointing towards the ground. A corner detection algorithm is used to capture an image point. Using a standard pinhole camera model, the three-dimensional point in the camera coordinate system... to image coordinates The changes are represented as follows; ； in, These are the scaling factors for the horizontal and vertical directions, respectively. These are the coordinates of the principal point; The normalized image coordinates of each feature point can be obtained from the mobile platform model. for ,in, ; Step S202: The monocular camera has a fixed field of view. The image-based controller can only operate stably when the feature point remains within its field of view. Therefore, the image coordinates of the feature point need to ensure the following visibility constraints: ； in, It can be obtained from the formula in step S201: ； Taking the derivative of the formula in step S201, we obtain the following expression, which correlates the normalized image coordinates with the velocity of the quadcopter / mobile platform: ； in, , and ; Step S203: Define the normalized image error as... To ensure that image errors do not exceed the field of view, a smoothly decreasing performance function is constructed to constrain the motion of feature points: ,in, All numbers are positive constants. The following coordinate transformation is defined: ； Therefore, by specifying the upper and lower bounds of the error, the normalized error is always kept within the time-varying, asymmetric upper and lower bounds. ,therefore: Similarly, we obtain ； ； ； Step S204: From the previous definition, the transformed image error is obtained as follows: ； From step S202, we can see that the purpose of constructing the transformed error is not only to ensure that the error is within the set time-varying upper and lower bound curves when converging, but also to ensure that the feature points are always within the field of view of the camera by selecting appropriate upper and lower bounds. Furthermore, when the error approaches zero, the transformed error also approaches zero. Step S3: Design an image-based landing control law to ensure that the pixels remain within the camera's field of view during the drone's landing process; Step S4: Calculate the thrust and torque of the quadrotor using inversion control technology and asymmetric time-varying Lyapunov functions to drive the quadrotor to land autonomously. Step S4 specifically includes: Step S401: Based on the fundamental idea of ​​inversion control technology, the complex quadcopter dynamics are decomposed into two subsystems: velocity and force. Based on the calculated velocity, an asymmetric time-varying Lyapunov function and intermediate virtual control variables are designed for the force subsystem to calculate the thrust and torque. First, the velocity error is defined as: , ,in, It is the intermediate virtual control quantity in the design; Step S402: As can be seen from step S203, according to An asymmetric time-varying Lyapunov function is constructed as follows: ,in, ; right Taking the differential, we get: ； Define a control vector: ； Based on the dynamic equations of the quadrotor, the thrust is designed as follows: ； in, It is a constant; Step S403: Because Recalling the previous step, an asymmetric time-varying Lyapunov function is defined and constructed as follows: ； Define a control vector: ； Based on the dynamic equations of the quadcopter, the design torque is as follows: ; in, It is a constant.

2. The robust landing method for a UAV on a mobile platform according to claim 1, characterized in that: Step S1 specifically includes the following steps: Step S101: The vision-based landing system consists of an underactuated quadcopter drone and an unknown mobile platform; Step S102: The dynamic model of the underactuated quadcopter UAV is shown below: ； Among them, the definition is These represent the position and velocity of the quadcopter in a Cartesian coordinate system. It is the angular velocity of the quadcopter. It is its quality. Represents the inertia matrix. and These are the thrust and torque vectors required for quadcopter flight. Represents a unit vector, rotation matrix By roll angle Pitch angle Yaw angle Composition, representing the rotational transformation from the quadrotor's body coordinate system to the world coordinate system, symbol. Represents a skew-symmetric matrix; Step S103: The mobile platform model is represented as follows: ； in, These represent the linear velocity and angular velocity of the mobile platform, respectively.

3. The robust landing method for a drone on a mobile platform according to claim 1, characterized in that: Step S3 specifically includes the following steps: Step S301: Since a single feature point can only control two degrees of freedom, four feature points are selected to control the six degrees of freedom velocity of the quadcopter, and a global image feature vector is defined. Therefore, the image kinematics is represented as ,in, , is a stack of Jacobian matrices of all features; Step S302: We design the image-based control law as follows: ,in, Represented as The generalized inverse.

Images