Unmanned aerial vehicle phased autonomous landing method for air-ground cooperation

Abstract

The application discloses a kind of unmanned aerial vehicle phased autonomous landing methods for air-ground cooperation, belong to artificial intelligence and robot field.This method includes the first stage based on trajectory tracking fast landing, and the second stage based on visual servo precision landing.First stage uses an improved environment adaptive weighting ant colony algorithm (EAW-ACO) to carry out global path search, and combined with Minimum Snap method generates smooth, safe flight trajectory, guide unmanned aerial vehicle to quickly drop to unmanned vehicle platform nearby.When unmanned aerial vehicle enters the landing area of pre-set, second stage uses the servo control based on visual marker, realizes high-precision terminal alignment and landing.The application has the unique advantage of fast landing in front and accurate docking in the end by the organic combination of two-stage control strategy, provides key technical support for unmanned aerial vehicle-unmanned vehicle air-ground cooperation and cluster continuous operation, has wide application prospect in intelligent warehousing, autonomous inspection and other fields.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and robotics, specifically to a phased autonomous landing method for unmanned aerial vehicles (UAVs) in air-ground collaborative scenarios. Background Technology

[0002] In modern warehousing and logistics, autonomous inspection, and other fields, the collaborative operation of unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) is a key technological path to improve system automation and operational efficiency. Within this collaborative framework, UAVs perform aerial tasks using their three-dimensional maneuverability, while UGVs provide ground support and significantly enhance the UAVs' endurance by providing autonomous charging. However, the core technological bottleneck in achieving this collaborative closed loop lies in whether the UAVs can autonomously, accurately, and reliably land on the ground-based UGV platform after the mission.

[0003] This landing process faces severe challenges, especially in environments such as indoor warehouses, where there are common problems such as narrow spaces, dense obstacles like shelves, lack of GPS signals, and small landing windows. Existing drone trajectory planning technologies are insufficient to perfectly address these challenges. Traditional path planning algorithms, such as the A* algorithm, have enormous computational costs in complex 3D environments; the Rapid Expanding Random Tree (RRT) algorithm generates paths with poor optimality and smoothness; and the Artificial Potential Field (APF) method is prone to getting trapped in local optima. Although swarm intelligence algorithms, represented by the Ant Colony Algorithm (ACO), have demonstrated good global search capabilities, the underlying algorithms suffer from slow convergence speed and insufficient path safety. While the latest end-to-end methods, such as Deep Reinforcement Learning (DRL), have great potential in decision-making intelligence, their training costs are high, sample requirements are large, and their reliability and interpretability in safety-critical scenarios still face challenges in practical application.

[0004] Therefore, existing technologies still lack a comprehensive solution that can efficiently and accurately adapt to complex indoor environments while simultaneously ensuring landing efficiency, obstacle avoidance safety, and end-point docking accuracy, in order to meet the needs of air-ground collaborative autonomous charging. Summary of the Invention

[0005] To address the technical challenge of simultaneously achieving landing efficiency, obstacle avoidance safety, and terminal landing accuracy when UAVs autonomously land on unmanned vehicles in complex indoor environments, this invention proposes a phased autonomous landing method for UAVs based on air-ground collaboration. This invention decouples the landing process into two stages: "rapid landing" and "precise landing," organically combining the speed advantages of global trajectory planning with the precision advantages of local visual servoing, thereby achieving rapid, accurate, and reliable landing of UAVs in complex environments.

[0006] This invention is achieved through the following technical solution:

[0007] A phased autonomous landing method for UAVs oriented towards air-to-ground cooperation includes the following steps:

[0008] Step 1: Based on the differential flatness characteristics of the UAV dynamics model, the UAV flight control problem is transformed into a trajectory planning problem in a flat output space;

[0009] Step 2: Establish a trajectory discretization model corresponding to the trajectory planning problem, and use a trajectory planning algorithm to search for a smooth and safe trajectory based on this model;

[0010] Step 3: The first stage is a rapid descent based on trajectory tracking. An Environment Adaptive Weighted Ant Colony Optimization (EAW-ACO) algorithm is first used to search for a discrete path point sequence connecting the starting point and the target point in the three-dimensional discrete grid model.

[0011] Step 4: Then, the Minimum Snap optimization method is used to take the discrete path point sequence as a constraint and generate a globally optimal trajectory by solving a quadratic programming (QP) problem.

[0012] Step 5: Then use the sliding mode cascade controller to control the drone to track the trajectory and perform the rapid descent phase until it reaches the end of the trajectory and enters the preset switching altitude;

[0013] Step 6: The second stage is precise landing based on visual servoing. In each sampling period, the airborne camera is used to measure the three-dimensional relative pose data of the UAV and the visual markers on the landing platform in real time.

[0014] Step 7: In each control cycle, the relative pose data calculated in Step 6 is used as feedback, and a sliding mode cascade control method based on RBF disturbance compensation is adopted to generate control commands to drive the UAV to complete a precise landing.

[0015] Furthermore, the environment-adaptive weighted ant colony optimization algorithm specifically includes the following steps:

[0016] Step 3.1: Parameter initialization: Initialize the algorithm parameters, including the number of ants, pheromone importance factor, heuristic function importance factor, pheromone evaporation coefficient, maximum number of iterations, and initialize the pheromone distribution of each path in the three-dimensional discrete grid model;

[0017] Step 3.2: Path construction based on environmental feedback: In the first step... In this iteration, all ants are placed at the starting point, and each ant repeats the following operations until it reaches the target point:

[0018] a) Calculate the environmental feedback factors from the current node to all neighboring nodes. Fixed obstacle influence factor and path shift influencing factors The product is composed of the following formulas:

[0019]

[0020] Among them, fixed obstacle influence factor Based on the distance between the path segment and the obstacle Perform piecewise calculations as shown in the following formula:

[0021]

[0022] in, Indicates the minimum safe interval distance. Indicates the obstacle buffer distance;

[0023] Path shift influencing factors Based on the yaw angle deviation of the path The calculation formula is as follows:

[0024]

[0025] in, Defined as the maximum permissible yaw angle deviation, therefore the deviation The smaller the value, the larger the factor value, so that the algorithm can prioritize the path with smoother course changes;

[0026] b) Based on the path selection strategy for the feedback factors in the fusion environment, the calculation formula is as follows:

[0027]

[0028] in, It is the ants from the node To the node The transition probability, It is a path The concentration of pheromones on the surface It is a heuristic function. , These are the pheromone importance factor and the heuristic function importance factor, respectively.

[0029] c) Use the roulette wheel selection method to move to the next node and mark this node as visited to construct a complete path;

[0030] Step 3.3: Adaptive Pheromones Update: After all ants have completed path construction in this iteration, an adaptive weighted pheromone update strategy is adopted to update the global pheromone distribution in the 3D discrete grid model based on the path quality constructed by all ants. The calculation formula for the adaptive weighted pheromone update strategy is as follows:

[0031]

[0032] It is a path The concentration of pheromones on the surface Indicates the first Only ants on the path The amount of pheromones released. It is a pheromone volatile factor. , For the first An adaptive weighting factor for each ant, used to adjust pheromone gain, is derived from environmental feedback factors along each path. It is calculated using an exponential function, and its formula is as follows:

[0033]

[0034] Step 3.4: Iteration Termination: Determine whether the maximum number of iterations has been reached. If yes, end the algorithm and output the optimal path as the discrete path point sequence; otherwise, return to step 3.2.

[0035] Furthermore, the transformation of the flat output space specifically includes the following steps:

[0036] Step 1.1: Select Flat Output: For a quadcopter UAV, select its position coordinates in three-dimensional space. and yaw angle As a flat output;

[0037] Step 1.2: Representation of State and Control Variables: All state variables and control inputs of the UAV are represented as functions of the flat output and its finite-order time derivative:

[0038]

[0039] In addition to the flat output, the state variables also include the roll angle. Pitch angle The control inputs include total tension. and control torque in three directions , , , For the quality of drones, It is the acceleration due to gravity. , , They are respectively around , , Moment of inertia of the shaft.

[0040] Furthermore, the purpose of establishing the trajectory discretization model is to transform the continuous flight trajectory planning problem in three-dimensional space into a path search problem for a discrete set of path points. By dividing the trajectory into equal time steps, each time step corresponds to an optional motion strategy. The motion strategy consists of a flat output and its derivative, and includes attributes such as position, yaw angle, linear velocity, and yaw angular velocity. Based on this model, a trajectory planning algorithm is used to search for a smooth and safe trajectory.

[0041] Furthermore, the Minimum Snap optimization method is specifically implemented by including the following steps:

[0042] Step 4.1: Piecewise polynomial trajectory modeling: [The sentence is incomplete and requires further context to be translated accurately.] Composed of a sequence of discrete path points The segments of the path are sequentially spliced ​​together to form a trajectory. , of which Segment trajectory It can be parameterized using a set of fifth-order polynomials:

[0043]

[0044] Among them, parameters Indicates time, Let be the coefficient vector of this trajectory;

[0045] Step 4.2: Optimization Objective Construction: Establish an objective function that aims to minimize the square integral of the fourth derivative of the entire trajectory. The objective function can be expressed as the segmented trajectory. The sum of the corresponding integrals:

[0046]

[0047] The objective function is about all unknown polynomial coefficients. The quadratic form can therefore be transformed into a standard quadratic programming cost function. ,in The matrix is ​​a constant matrix pre-calculated based on the optimization objective;

[0048] Step 4.3: Constraint Construction: Establish linear equation constraints including path point constraints, trajectory endpoint constraints, and continuity constraints between each segmented trajectory. The path point constraints require that the trajectory must precisely pass through all discrete path points planned by the environment-adaptive weighted ant colony optimization algorithm. The trajectory endpoint constraints require that the position, velocity, acceleration, and higher-order derivatives of the starting and ending points of the trajectory meet preset values. The continuity constraints require that at the connection points of each discrete path point, the ending state of the previous trajectory segment and the starting state of the next trajectory segment remain continuous in terms of position, velocity, acceleration, and even the fourth-order derivative. All of the above constraints are about the polynomial coefficient vector. Linear equality constraints are uniformly expressed in a standard form:

[0049]

[0050] in, This represents the set of time matrices corresponding to all constraints. Represents the set of all constraints;

[0051] Step 4.4: Solving the Quadratic Programming Problem: The optimization objective and constraints are combined to construct a standard quadratic programming (QP) problem, and the unique optimal trajectory coefficients are obtained by solving this problem in a closed loop.

[0052]

[0053] Step 4.5: Trajectory Generation: The obtained optimal coefficients... Substituting the values ​​into the piecewise polynomial model from step 4.1, the final smooth trajectory that can be executed by the UAV flight control system can be generated. .

[0054] Furthermore, the trajectory tracking specifically includes the following steps:

[0055] Step 5.1: Status Acquisition: During the rapid descent, the UAV continuously acquires its actual operational status variables in three-dimensional space at a high frequency through the global positioning module on board. These variables include real-time position, velocity, attitude, and angular velocity data.

[0056] Step 5.2: Command Query: In each control cycle, the onboard computing unit queries the command based on the current time. The location trajectory generated in step 4 and yaw angle trajectory The query retrieves the expected flat output at that moment, including the expected position. and desired yaw angle Then, other state variables are calculated from the flat output;

[0057] Step 5.3: Deviation Calculation: Compare the actual operating state obtained in Step 5.1 with the expected operating state queried in Step 5.2, and calculate the real-time tracking deviation, including position loop deviation and attitude loop deviation;

[0058] Step 5.4: Control Calculation and Execution: The calculated tracking deviation is input into a sliding mode cascade controller for calculation; this controller includes an outer loop position controller and an inner loop attitude controller. The position controller calculates the desired attitude that the UAV needs to achieve based on the position and velocity deviations. , ) and total tension Subsequently, the inner loop employs a global fast terminal sliding mode control attitude controller, which is responsible for quickly and accurately tracking the desired attitude given by the outer loop and calculating the final three-axis control torque to be applied. Finally, total tensile force and three-axis control torque ( , , The calculation and allocation of the UAV inverse dynamics model is transformed into speed commands for the four rotor motors and then issued for execution, driving the UAV to follow the predetermined trajectory.

[0059] Furthermore, the visual marker is a nested positioning marker, with the outer layer consisting of two concentric circles of different sizes nested eccentrically, and the inner layer consisting of a nested QR code. During the calculation process, for the outer marker, the target ellipse set is first obtained by an ellipse detection method based on arc segment detection and classification, and then the target concentric ellipse set is extracted by combining concentric ellipse constraints. The projection points of the concentric circle centers are then located to complete the marker positioning calculation. For the inner marker, after recognizing the target QR code, an improved EPnP pose calculation method is used to obtain the relative pose of the nested QR code.

[0060] Furthermore, the sliding mode cascade control method based on RBF disturbance compensation specifically includes the following steps:

[0061] Step 7.1: RBF Neural Network Perturbation Estimation: An RBF neural network is designed as the perturbation observer, using a Gaussian radial basis function as the activation function. It has a three-layer structure: an input layer, a hidden layer, and an output layer. The input layer selects the position tracking error and its rate of change. As input Hidden layer The activation function for each neuron is:

[0062]

[0063] in, These are the central parameters of each neuron. It is the width of the Gaussian function;

[0064] The output layer weight vector is Interference It can be approximated by an RBF neural network as follows:

[0065]

[0066] in, These are the output parameters of each neuron. It is an approximation error; this is used to estimate and approximate the disturbances caused by the ground effect in real time. ;

[0067] Step 7.2: Outer Loop Position Control: Based on the three-dimensional relative pose deviation, a sliding mode position controller is designed. This controller calculates the total tension required to eliminate the three-dimensional relative pose deviation. And desired posture, including roll angle Pitch angle And send this desired attitude as an instruction to the inner loop;

[0068] Step 7.3: Inner Loop Attitude Control: Based on the attitude error between the desired attitude output from the outer loop and the actual attitude obtained from the inertial measurement unit, a global fast terminal sliding mode attitude controller is designed. The control torque is synthesized by the sliding mode feedback control part and the RBF feedforward compensation part. .

[0069] Compared with the prior art, the present invention has at least the following beneficial effects or advantages:

[0070] This invention proposes a phased autonomous landing method for UAVs in air-to-ground collaborative scenarios, aiming to address the core technical bottleneck of simultaneously achieving rapid approach efficiency, obstacle avoidance safety in complex environments, and end-point docking accuracy when UAVs autonomously land on ground-based unmanned vehicle (UGV) platforms to obtain charging and refueling. This invention decouples the landing process into "rapid landing under global trajectory guidance" and "precise landing under local visual servoing," specifically combining the speed and safety advantages of global trajectory planning with the accuracy advantages of local visual servoing. This not only achieves rapid and precise landing but also establishes a key closed loop for autonomous collaborative operation between UAVs and UGVs, providing core technical support for significantly improving the continuous operation capability and endurance of UAV swarms through autonomous charging. Attached Figure Description

[0071] The present invention will now be described in further detail with reference to the accompanying drawings:

[0072] Figure 1This is a system flowchart of the phased autonomous landing method for UAVs oriented towards air-ground cooperation according to the present invention.

[0073] Figure 2 This is a schematic diagram of the landing process of the present invention;

[0074] Figure 3 This is a flowchart of the EAW-ACO-MS method of the present invention;

[0075] Figure 4 This is a diagram of the sliding mode cascade control architecture of the present invention. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0077] Reference Figures 1 to 3 As shown, the core idea of ​​the phased autonomous landing method for UAVs oriented towards air-ground cooperation in this invention is to decouple the complex landing task into two consecutive stages: the first stage is a rapid landing based on trajectory tracking, and the second stage is a precise landing based on visual servoing.

[0078] Phase 1: Rapid landing based on trajectory tracking. The goal of this phase is to guide the UAV safely and quickly from the starting point to a switching point near the target landing platform. The trajectory for this phase is planned and generated by the EAW-ACO-MS algorithm proposed in this invention, and its implementation includes the following steps:

[0079] Step 1: Based on the differential flatness characteristics of the UAV dynamics model, the UAV flight control problem is transformed into a trajectory planning problem in a flat output space;

[0080] Step 2: Establish a trajectory discretization model corresponding to the trajectory planning problem, and use a trajectory planning algorithm to search for a smooth and safe trajectory based on this model;

[0081] Step 3: First, use an Environment Adaptive Weighted Ant Colony Optimization (EAW-ACO) algorithm to search for a discrete path point sequence connecting the starting point and the target point in the three-dimensional discrete grid model;

[0082] Step 4: Then, the Minimum Snap optimization method is used to take the discrete path point sequence as a constraint and generate a globally optimal trajectory by solving a quadratic programming (QP) problem.

[0083] Step 5: Then use the sliding mode cascade controller to control the drone to track the trajectory and perform the rapid descent phase until it reaches the end of the trajectory and enters the preset switching altitude;

[0084] Phase Two: Precision Landing Based on Visual Servo. The goal of this phase is to accurately land from the switching point to the landing point. The specific implementation includes the following steps:

[0085] Step 6: Precise landing based on visual servoing. In each sampling cycle, the airborne camera is used to measure the three-dimensional relative pose data of the UAV and the visual markers on the landing platform in real time.

[0086] Step 7: In each control cycle, the relative pose data calculated in Step 6 is used as feedback. A sliding mode cascade control method based on RBF disturbance compensation is adopted to generate control commands to drive the UAV to complete a precise landing.

[0087] First, the rapid descent phase needs to be completed. The specific steps are as follows:

[0088] For trajectory planning of unmanned aerial vehicles (UAVs), a differential flat space transformation is used to convert the complex dynamic constraints of the UAV into a simplified trajectory planning method with a flat output space. The differential flat space transformation includes the following steps:

[0089] Step 1.1: Select Flat Output: For a quadcopter UAV, select its position coordinates in three-dimensional space. and yaw angle As a flat output;

[0090] Step 1.2: Representation of State and Control Variables: All state variables and control inputs of the UAV are represented as functions of the flat output and its finite-order time derivative:

[0091]

[0092] In addition to the flat output, the state variables also include the roll angle. Pitch angle The control inputs include total tension. and control torque in three directions , , , For the quality of drones, It is the acceleration due to gravity. , , They are respectively around , , Moment of inertia of the shaft.

[0093] A trajectory discretization model is established for the trajectory planning problem, transforming the continuous flight trajectory planning problem in three-dimensional space into a path search problem for a discrete set of path points. The trajectory is divided into equal time steps, with each time step corresponding to an optional motion strategy. The motion strategy consists of a flat output and its derivative, including attributes such as position, yaw angle, linear velocity, and yaw angular velocity. Based on this model, a trajectory planning algorithm is used to search for a smooth and safe trajectory.

[0094] For path planning during the rapid descent phase, the Environment Adaptive Weighted Ant Colony (EAW-ACO) algorithm is employed. This algorithm aims to overcome the bottlenecks of traditional ACO algorithms, such as slow convergence, susceptibility to local optima, and neglect of flight safety in complex 3D environments. The Environment Adaptive Weighted Ant Colony optimization algorithm includes the following steps:

[0095] 1) Parameter initialization: Set algorithm parameters, including the number of ants. Pheromones are important factors Heuristic function importance factor pheromone volatility coefficient Maximum number of iterations and all path segments on the map. pheromone concentration ;

[0096] 2) Path construction based on environmental feedback: In each iteration In the middle, All ants are placed at the starting point, and each ant independently repeats the following operations until it reaches the target point:

[0097] a) Calculation of environmental feedback factors: for the current node Each reachable adjacent node Calculate the environmental feedback factor according to the following formula. :

[0098]

[0099] Among them, fixed obstacle influence factor Based on the distance between the path segment and the obstacle Perform piecewise calculations as shown in the following formula:

[0100]

[0101] in, Indicates the minimum safe interval distance. Indicates the obstacle buffer distance;

[0102] Path shift influencing factors Based on the yaw angle deviation of the path The calculation formula is as follows:

[0103]

[0104] in, Defined as the maximum permissible yaw angle deviation, therefore the deviation The smaller the value, the larger the factor value, so that the algorithm can prioritize the path with smoother course changes;

[0105] b) Based on the path selection strategy for the feedback factors in the fusion environment, the calculation formula is as follows:

[0106]

[0107] in, It is the ants from the node To the node The transition probability, It is a path The concentration of pheromones on the surface It is a heuristic function. , These are the pheromone importance factor and the heuristic function importance factor, respectively. Representing the Only an ant;

[0108] c) Use the roulette wheel selection method to move to the next node and mark this node as visited to construct a complete path;

[0109] 3) Adaptive pheromone update: After all ants have completed path construction in this iteration, an adaptive weighted pheromone update strategy is adopted to update the global pheromone distribution in the 3D discrete grid model based on the quality of the paths constructed by all ants. The calculation formula for the adaptive weighted pheromone update strategy is as follows:

[0110]

[0111] in, It is a path The concentration of pheromones on the surface Indicates the first Only ants on the path The amount of pheromones released. It is a pheromone volatile factor. , For the first An adaptive weighting factor for each ant, used to adjust pheromone gain, is derived from environmental feedback factors along each path. It is calculated using an exponential function, and its formula is as follows:

[0112]

[0113] This strategy enables ants with better path quality to release more pheromones, thereby accelerating the convergence of the algorithm to a high-quality solution.

[0114] 4) Iteration Termination: Repeat steps 2) and 3) until the preset maximum number of iterations is reached. After the algorithm terminates, the optimal path recorded during all iterations is taken as the final result, which is the discrete path point sequence output by the EAW-ACO algorithm.

[0115] For discrete waypoint sequences, trajectory optimization is performed using the Minimum Snap method to generate a continuous, smooth, and dynamically feasible flight trajectory. The basic principle is to parameterize the discrete waypoints using piecewise polynomials and find a trajectory that passes through all waypoints and minimizes the square integral of its fourth derivative by solving a quadratic programming (QP) problem, thus ensuring the high smoothness of the trajectory. The Minimum Snap method includes the following steps:

[0116] 1) Piecewise polynomial trajectory modeling: The output of the EAW-ACO algorithm... Discrete path points The composition The path segment is parameterized using a set of piecewise fifth-order polynomials. Segment trajectory It can be represented as:

[0117]

[0118] Among them, parameters Indicates time, Let be the coefficient vector of this trajectory;

[0119] 2) Optimization objective construction: Establish an objective function that aims to minimize the square integral of the fourth derivative of the entire trajectory. The aim is to suppress abrupt changes in the control system and generate a smoother trajectory. This cost function can be expressed as the sum of the integrals of each segmented trajectory:

[0120]

[0121] The objective function is about all unknown polynomial coefficients. The quadratic form can therefore be transformed into a standard quadratic programming cost function. ,in The matrix is ​​a constant matrix pre-calculated based on the optimization objective;

[0122] 3) Constraint Construction: To ensure the effectiveness and smoothness of the trajectory, the following three types of linear equality constraints are established: a) Path point constraints: The trajectory must precisely pass through all discrete path points planned by the EAW-ACO algorithm at a specified time; b) Endpoint constraints: The position, velocity, acceleration, and higher-order derivatives of the starting and ending points of the entire trajectory must meet preset values; c) Continuity constraints: At the connection points of discrete path points, the ending state of the previous trajectory segment and the starting state of the next trajectory segment must remain continuous in terms of position, velocity, acceleration, and even the fourth-order derivative.

[0123] 4) Solving the quadratic programming problem: Represent all the linear equality constraints established in step 3) in a standard matrix form:

[0124]

[0125] Subsequently, the optimization objective of step 2) and the constraints of step 3) are combined to construct a standard quadratic programming (QP) problem:

[0126]

[0127] 5) Trajectory Generation: The obtained optimal coefficients Substituting the result into the piecewise polynomial model of step 1), the final smooth trajectory that can be executed by the UAV flight control can be generated.

[0128] For trajectory tracking during the rapid descent phase, the trajectory tracking method includes the following steps:

[0129] 1) Status acquisition: During rapid descent, the UAV continuously acquires its actual operational status variables in three-dimensional space at a high frequency through the global positioning module on board. These variables include real-time position, speed, attitude, and angular velocity data.

[0130] 2) Command query: In each control cycle, the airborne computing unit queries the command based on the current time. The location trajectory generated in step 4 and yaw angle trajectory The query retrieves the expected flat output at that moment, including the expected position. and desired yaw angle Then, other state variables are calculated from the flat output;

[0131] 3) Deviation calculation: Compare the actual operating state obtained in step 5.1 with the expected operating state queried in step 5.2, and calculate the real-time tracking deviation, including position loop deviation and attitude loop deviation;

[0132] 4) Control Calculation and Execution: The calculated tracking deviation is input into a sliding mode cascade controller for calculation; this controller includes an outer-loop position controller and an inner-loop attitude controller. The position controller calculates the desired attitude that the UAV needs to achieve based on the position and velocity deviations. , ) and total tension Subsequently, the inner loop employs a global fast terminal sliding mode control attitude controller, which is responsible for quickly and accurately tracking the desired attitude given by the outer loop and calculating the final three-axis control torque to be applied. Finally, total tensile force and three-axis control torque ( , , The calculation and allocation of the UAV inverse dynamics model is transformed into speed commands for the four rotor motors and then issued for execution, driving the UAV to follow the predetermined trajectory.

[0133] Next, the precise landing phase needs to be completed, and the specific steps are as follows:

[0134] For the precise landing phase, a nested positioning marker is used for relative pose calculation. The outer layer consists of two concentric circles of different sizes nested eccentrically, while the inner layer uses a nested QR code. During the calculation process, for the outer marker, the target ellipse set is first obtained through an ellipse detection method based on arc segment detection and classification. Then, the target concentric ellipse set is extracted by combining concentric ellipse constraints, and the projection points of the concentric circle centers are located to complete the marker positioning calculation. For the inner marker, after recognizing the target QR code, an improved EPnP pose calculation method is used to obtain the relative pose of the nested QR code.

[0135] For ground effect compensation during the precision landing phase, a sliding mode cascaded control method based on RBF disturbance compensation is adopted. This method utilizes an RBF neural network to approximate and estimate unknown disturbances such as ground effects online, and the estimated values ​​are input as feedforward compensation terms into the sliding mode controller to enhance the robustness and accuracy of the control system. The sliding mode cascaded control method based on RBF disturbance compensation includes the following steps:

[0136] 1) RBF Neural Network Perturbation Estimation: An RBF neural network is designed as the perturbation observer, using a Gaussian radial basis function as the activation function. It has a three-layer structure: input layer, hidden layer, and output layer. The input layer selects the position tracking error and its rate of change. As input Hidden layer The activation function for each neuron is:

[0137]

[0138] in, These are the central parameters of each neuron. It is the width of the Gaussian function;

[0139] The output layer weight vector is Interference It can be approximated by an RBF neural network as follows:

[0140]

[0141] in, These are the output parameters of each neuron. It is an approximation error, used to estimate and approximate the disturbances caused by the ground effect in real time. ;

[0142] 2) Outer loop position control: Based on the three-dimensional relative pose deviation, a sliding mode position controller is designed. This controller calculates the total tension required to eliminate the three-dimensional relative pose deviation. And desired posture, including roll angle Pitch angle And send this desired attitude as an instruction to the inner loop;

[0143] 3) Inner-loop attitude control: Based on the attitude error between the desired attitude output from the outer loop and the actual attitude obtained from the inertial measurement unit, a global fast terminal sliding mode attitude controller is designed. The control torque is synthesized by the sliding mode feedback control part and the RBF feedforward compensation part. .

[0144] The above describes in detail specific implementation examples of the present invention. The present invention has many specific applications. The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A phased autonomous landing method for unmanned aerial vehicles (UAVs) oriented towards air-ground cooperation, characterized in that... Includes the following steps: Step 1: Based on the differential flatness characteristics of the UAV dynamics model, the UAV flight control problem is transformed into a trajectory planning problem in a flat output space; Step 2: Establish a trajectory discretization model corresponding to the trajectory planning problem, and use a trajectory planning algorithm to search for a smooth and safe trajectory based on this model; Step 3: The first stage is a rapid descent based on trajectory tracking. An Environment Adaptive Weighted Ant Colony Optimization (EAW-ACO) algorithm is first used to search for a discrete path point sequence connecting the starting point and the target point in the three-dimensional discrete grid model. Step 4: Then, the Minimum Snap optimization method is used to take the discrete path point sequence as a constraint and generate a globally optimal trajectory by solving a quadratic programming (QP) problem. Step 5: Then use the sliding mode cascade controller to control the drone to track the trajectory and perform the rapid descent phase until it reaches the end of the trajectory and enters the preset switching altitude; Step 6: The second stage is precise landing based on visual servoing. In each sampling period, the airborne camera is used to measure the three-dimensional relative pose data of the UAV and the visual markers on the landing platform in real time. Step 7: In each control cycle, the relative pose data calculated in Step 6 is used as feedback. A sliding mode cascade control method based on RBF disturbance compensation is adopted to generate control commands to drive the UAV to complete a precise landing. The environment adaptive weighted ant colony optimization algorithm specifically includes the following steps: Step 3.1: Parameter initialization: Initialize the algorithm parameters, including the number of ants, pheromone importance factor, heuristic function importance factor, pheromone evaporation coefficient, maximum number of iterations, and initialize the pheromone distribution of each path in the three-dimensional discrete grid model; Step 3.2: Path construction based on environmental feedback: In the first step... In this iteration, all ants are placed at the starting point, and each ant repeats the following operations until it reaches the target point: a) Calculate the environmental feedback factors from the current node to all neighboring nodes. Fixed obstacle influence factor and path shift influencing factors The product is composed of the following formulas: Among them, fixed obstacle influence factor Based on the distance between the path segment and the obstacle Perform piecewise calculations as shown in the following formula: in, Indicates the minimum safe interval distance. Indicates the obstacle buffer distance; Path shift influencing factors Based on the yaw angle deviation of the path The calculation formula is as follows: in, Defined as the maximum permissible yaw angle deviation, therefore the deviation The smaller the value, the larger the factor value, so that the algorithm can prioritize the path with smoother course changes; b) Based on the path selection strategy for the feedback factors in the fusion environment, the calculation formula is as follows: in, It is the ants from the node To the node The transition probability, It is a path The concentration of pheromones on the surface It is a heuristic function. , These are the pheromone importance factor and the heuristic function importance factor, respectively. c) Use the roulette wheel selection method to move to the next node and mark this node as visited to construct a complete path; Step 3.3: Adaptive Pheromones Update: After all ants have completed path construction in this iteration, an adaptive weighted pheromone update strategy is adopted to update the global pheromone distribution in the 3D discrete grid model based on the path quality constructed by all ants. The calculation formula for the adaptive weighted pheromone update strategy is as follows: It is a path The concentration of pheromones on the surface Indicates the first Only ants on the path The amount of pheromones released. It is a pheromone volatile factor. , For the first An adaptive weighting factor for each ant, used to adjust pheromone gain, is derived from environmental feedback factors along each path. It is calculated using an exponential function, and its formula is as follows: Step 3.4: Iteration Termination: Determine whether the maximum number of iterations has been reached. If yes, end the algorithm and output the optimal path as the discrete path point sequence; otherwise, return to step 3.

2.

2. The phased autonomous landing method for UAVs oriented towards air-ground cooperation according to claim 1, characterized in that, The flat output space transformation involves representing all state variables and control inputs of the system algebraically using a set of flat output variables and their finite-order time derivatives. Specifically, it includes the following steps: Step 1.1: Select Flat Output: For a quadcopter UAV, select its position coordinates in three-dimensional space. and yaw angle As a flat output; Step 1.2: Representation of State and Control Variables: All state variables and control inputs of the UAV are represented as functions of the flat output and its finite-order time derivative: In addition to the flat output, the state variables also include the roll angle. Pitch angle The control inputs include total tension. and control torque in three directions , , , For the quality of drones, It is the acceleration due to gravity. , , They are respectively around , , Moment of inertia of the shaft.

3. The phased autonomous landing method for UAVs oriented towards air-ground cooperation according to claim 1, characterized in that, The purpose of establishing the trajectory discretization model is to transform the continuous flight trajectory planning problem in three-dimensional space into a path search problem for a discrete set of path points. By dividing the trajectory into equal time steps, each time step corresponds to an optional motion strategy. The motion strategy consists of a flat output and its derivative, and includes attributes such as position, yaw angle, linear velocity, and yaw angular velocity. Based on this model, a trajectory planning algorithm is used to search for a smooth and safe trajectory.

4. The phased autonomous landing method for UAVs oriented towards air-ground cooperation according to claim 1, characterized in that, The Minimum Snap optimization method specifically includes the following steps: Step 4.1: Piecewise polynomial trajectory modeling: [The sentence is incomplete and requires further context to be translated accurately.] Composed of a sequence of discrete path points The segments of the path are sequentially spliced ​​together to form a trajectory. , of which Segment trajectory Represented using a set of fifth-order polynomials: Among them, parameters Indicates time, Let be the coefficient vector of this trajectory; Step 4.2: Optimization Objective Construction: Establish an objective function that aims to minimize the square integral of the fourth derivative of the entire trajectory. The objective function can be expressed as the segmented trajectory. The sum of the corresponding integrals: The objective function is about all unknown polynomial coefficients. The quadratic form can therefore be transformed into a standard quadratic programming cost function. ,in The matrix is ​​a constant matrix pre-calculated based on the optimization objective; Step 4.3: Constraint Construction: Establish linear equation constraints including path point constraints, trajectory endpoint constraints, and continuity constraints between each segmented trajectory. The path point constraints require that the trajectory must precisely pass through all discrete path points planned by the environment-adaptive weighted ant colony optimization algorithm. The trajectory endpoint constraints require that the position, velocity, acceleration, and higher-order derivatives of the starting and ending points of the trajectory meet preset values. The continuity constraints require that at the connection points of each discrete path point, the ending state of the previous trajectory segment and the starting state of the next trajectory segment remain continuous in terms of position, velocity, acceleration, and even the fourth-order derivative. All of the above constraints are about the polynomial coefficient vector. Linear equality constraints are uniformly expressed in a standard form: in, This represents the set of time matrices corresponding to all constraints. Represents the set of all constraints; Step 4.4: Solving the Quadratic Programming Problem: The optimization objective and constraints are combined to construct a standard quadratic programming (QP) problem, and the unique optimal trajectory coefficients are obtained by solving this problem in a closed loop. Step 4.5: Trajectory Generation: The obtained optimal coefficients... Substituting the values ​​into the piecewise polynomial model from step 4.1, the final smooth trajectory that can be executed by the UAV flight control system can be generated. .

5. The phased autonomous landing method for UAVs oriented towards air-ground cooperation according to claim 1, characterized in that, The trajectory tracking specifically includes the following steps: Step 5.1: Status Acquisition: During the rapid descent, the UAV continuously acquires its actual operational status variables in three-dimensional space at a high frequency through the global positioning module on board. These variables include real-time position, velocity, attitude, and angular velocity data. Step 5.2: Command Query: In each control cycle, the onboard computing unit queries the command based on the current time. The location trajectory generated in step 4 and yaw angle trajectory The query retrieves the expected flat output at that moment, including the expected position. and desired yaw angle Then, other state variables are calculated from the flat output; Step 5.3: Deviation Calculation: Compare the actual operating state obtained in Step 5.1 with the expected operating state queried in Step 5.2, and calculate the real-time tracking deviation, including position loop deviation and attitude loop deviation; Step 5.4: Control Calculation and Execution: The calculated tracking deviation is input into a sliding mode cascade controller for calculation; this controller includes an outer loop position controller and an inner loop attitude controller. The position controller calculates the desired attitude that the UAV needs to achieve based on the position and velocity deviations. , ) and total tension Subsequently, the inner loop employs a global fast terminal sliding mode control attitude controller, which is responsible for quickly and accurately tracking the desired attitude given by the outer loop and calculating the final three-axis control torque to be applied. Finally, total tensile force and three-axis control torque ( , , The calculation and allocation of the UAV inverse dynamics model is transformed into speed commands for the four rotor motors and then issued for execution, driving the UAV to follow the predetermined trajectory.

6. The phased autonomous landing method for UAVs oriented towards air-ground cooperation according to claim 1, characterized in that, The visual marker is a nested positioning marker, consisting of two concentric circles of different sizes nested eccentrically on the outer layer, and a nested QR code on the inner layer. During the calculation process, for the outer marker, the target ellipse set is first obtained by an ellipse detection method based on arc segment detection and classification, and then the target concentric ellipse set is extracted by combining concentric ellipse constraints. The center projection points of the concentric circles are then located to complete the marker positioning calculation. For the inner marker, after recognizing the target QR code, an improved EPnP pose calculation method is used to obtain the relative pose of the nested QR code.

7. The phased autonomous landing method for UAVs oriented towards air-ground cooperation according to claim 1, characterized in that, The sliding mode cascade control method based on RBF disturbance compensation specifically includes the following steps: Step 7.1: RBF Neural Network Perturbation Estimation: An RBF neural network is designed as the perturbation observer, using a Gaussian radial basis function as the activation function. It has a three-layer structure: an input layer, a hidden layer, and an output layer. The input layer selects the position tracking error and its rate of change. As input Hidden layer The activation function for each neuron is: in, These are the central parameters of each neuron. It is the width of the Gaussian function; The output layer weight vector is Interference It can be approximated by an RBF neural network as follows: in, These are the output parameters of each neuron. It is an approximation error; this is used to estimate and approximate the disturbances caused by the ground effect in real time. ; Step 7.2: Outer Loop Position Control: Based on the three-dimensional relative pose deviation, a sliding mode position controller is designed. This controller calculates the total tension required to eliminate the three-dimensional relative pose deviation. And desired posture, including roll angle Pitch angle And send this desired attitude as an instruction to the inner loop; Step 7.3: Inner Loop Attitude Control: Based on the attitude error between the desired attitude output from the outer loop and the actual attitude obtained from the inertial measurement unit, a global fast terminal sliding mode attitude controller is designed. The control torque is synthesized by the sliding mode feedback control part and the RBF feedforward compensation part. .

Images