Unmanned aerial vehicle low-altitude flight intelligent management method and system based on cloud control

Abstract

The invention provides an unmanned aerial vehicle low-altitude flight intelligent management method and system based on cloud control, and the method comprises the steps: obtaining static obstacle information through a three-dimensional environment map, generating a global guidance corridor in combination with a task starting point, a task ending point and a flight energy consumption model, and fusing the data of a GPS, an IMU and a visual sensor in real time. The method comprises the following steps: estimating the pose and covariance matrix of an aircraft by using an extended Kalman filter, guiding to detect local obstacles in a corridor, predicting a flight path, constructing a space-time obstacle avoidance area and calculating an environment complexity index in a flight process, and triggering global path re-planning when a conflict risk exceeds a threshold value, adjusting the safety margin of the static obstacle repulsive force potential field according to the pose covariance matrix, optimizing the multi-target weight based on the complexity index, solving the optimal local velocity vector and the optimal cost value, adjusting the controller gain through the deviation between the velocity vector and the current pose and combining the optimal cost value, and generating a control instruction. And the aircraft is ensured to stably follow the optimal path in a complex environment.

Description

Technical Field

[0001] This application belongs to the field of control, and in particular relates to a cloud-based intelligent management method and system for low-altitude flight of unmanned aerial vehicles. Background Technology

[0002] Unmanned aerial vehicles (UAVs) are increasingly used in surveying, logistics, and inspection, and typically rely on cloud platforms for unified monitoring and task scheduling. Currently, most cloud-based UAV flight control methods employ a layered framework, combining global path planning in the cloud with onboard real-time obstacle avoidance. The system pre-acquires static map information of the cloud environment and uses search algorithms such as A* and RRT* to plan a collision-free path from the starting point to the destination. However, these methods mostly plan based on a static environment, resulting in a rigid optimal path generated by the cloud rather than a flexible guidance area. Onboard obstacle avoidance commonly uses methods such as artificial potential fields and dynamic windows to address unknown obstacles detected by sensors. However, artificial potential fields are prone to getting trapped in local minima, and most methods still rely on reactive obstacle avoidance, lacking reliable prediction of obstacle movement trends. This leads to short-sighted obstacle avoidance behavior and difficulty in ensuring safety in high-density scenarios. Overly conservative obstacle avoidance strategies sacrifice flight efficiency; and when pose estimation accuracy decreases, the safety margin may be insufficient to mitigate risks. Whether the drone is in an open and simple environment or faces obstacle avoidance situations requiring complex maneuvers, the interaction between the cloud and the airborne terminal is often limited to simple command issuance, and the airborne control response characteristics remain unchanged. The control strategy cannot be adjusted according to the current cloud task requirements and the real-time difficulty or cost of the locally planned task, which limits the performance of the aircraft in different environments and makes it difficult to achieve a balance between safety, stability, and maneuverability. Summary of the Invention

[0003] To address the above problems, this invention proposes a cloud-based intelligent management method and system for low-altitude flight of unmanned aerial vehicles, comprising the following steps: The system receives a 3D environmental map containing static obstacle information and flight mission instructions from the cloud control platform. Based on the mission start point, end point, and flight energy consumption model, it performs a global path search to generate a 3D global guidance corridor. It also integrates data from the Global Positioning System, Inertial Measurement Unit, and Visual Sensor in real time, uses an Extended Kalman Filter to estimate the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle, and uploads the current pose state to the cloud control platform for monitoring in real time. Within the three-dimensional global guidance corridor, airborne sensors are used to detect local and static obstacles, predict the future trajectories of the obstacles to construct a spatiotemporal obstacle avoidance zone, and calculate the local environmental complexity index. When the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, global path replanning is triggered with the current position of the unmanned aerial vehicle as a new starting point, and the replanned path result is updated to the cloud control platform. Based on the determinant value of the pose estimation covariance matrix, the safety margin of the static obstacle repulsive potential field is adjusted; and based on the local environment complexity index, the weight coefficients of the global guiding corridor gravity, static obstacle repulsive force, and spatiotemporal obstacle avoidance term in the multi-objective optimization problem are adjusted to obtain the optimal local expected velocity vector and the optimization cost of the solution. Based on the deviation between the optimal local desired velocity vector and the current pose state of the unmanned aerial vehicle, and according to the optimization cost, the gain parameters of the nonlinear controller are adjusted to generate control commands.

[0004] Optionally, the step of performing a global path search based on the mission start point, end point, and flight energy consumption model to generate a three-dimensional global guidance corridor includes: In the cloud control platform or the onboard computer of the unmanned aerial vehicle, the three-dimensional environment map is discretized into grid cells, and a search algorithm is used with path integral cost. The search is performed with the minimum as the objective, where P is the path and ds is the path element. and Preset weights configured for the cloud control platform; Energy consumption per unit distance Based on the current payload m and preset cruise speed of the unmanned aerial vehicle and climb angle Calculation, in the form of ,in Where g is air resistance and g is the acceleration due to gravity. The optimal path point sequence found is expanded outward to form a tubular region, which serves as the three-dimensional global guidance corridor and is then sent to the local planning module.

[0005] Optionally, the step of estimating the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle using an extended Kalman filter includes: In the prediction step, based on the six-degree-of-freedom kinematic model of the unmanned aerial vehicle and the pose state at the previous moment, the prior pose state and prior covariance at the current moment are predicted. In the update step, the three-dimensional position provided by the Global Positioning System, the three-axis acceleration and angular velocity provided by the Inertial Measurement Unit, and the feature point depth information provided by the visual sensor are used as measurement values ​​to calculate the Kalman gain and update the prior pose state and prior covariance to obtain the posterior pose state estimate and pose estimation covariance matrix at the current time.

[0006] Optionally, predicting the future trajectory of the obstacle to construct a spatiotemporal obstacle avoidance area includes: For each detected obstacle, a motion model is used to predict the sequence of the obstacle's positions over a future period of time; Each predicted location point is expanded into a safety envelope, and the set of safety envelopes for all predicted time steps constitutes the spatiotemporal obstacle avoidance region of the obstacle.

[0007] Optionally, when the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, global path replanning is triggered with the current position of the unmanned aerial vehicle as a new starting point, including: Calculate the minimum distance between the predicted trajectory of the obstacle and the centerline of the three-dimensional global guidance corridor; A conflict risk value is calculated based on the minimum distance, and the conflict risk value increases as the minimum distance decreases; When the conflict risk value exceeds the first preset threshold issued by the cloud control platform, global path replanning is triggered.

[0008] Optionally, adjusting the safety margin of the static obstacle repulsive potential field based on the determinant value of the pose estimation covariance matrix includes: Based on the pose estimation covariance matrix determinant value Adjust the safety margin of the repulsive potential field of the static obstacle according to the following formula. ,in Based on the basic safety margin, k and These are the positive tuning parameters.

[0009] Optionally, adjusting the weight coefficients of the global guiding corridor gravity, static obstacle repulsion, and spatiotemporal obstacle avoidance term in the multi-objective optimization problem based on the local environment complexity index includes: Local environment complexity index This is the sum of the speeds of all obstacles within the sensor's detection range: Where N is the number of obstacles, Let be the velocity vector of the i-th obstacle; According to the index The weighting coefficients are adjusted such that when the exponent increases, the weighting coefficient of the spatiotemporal obstacle avoidance term increases, while the weighting coefficient of the global guidance corridor gravity term decreases.

[0010] Optionally, adjusting the gain parameter of the nonlinear controller according to the optimized cost to generate control commands includes: A proportional-integral-derivative (PID) controller is used as the nonlinear controller. According to the optimized cost value The proportional gain of the PID controller is adjusted inversely using the following functional relationship. : ,in The baseline proportional gain set for the cloud control platform. It is a positive adjustment coefficient.

[0011] Furthermore, this invention also relates to a cloud-based intelligent management system for low-altitude flight of unmanned aerial vehicles, comprising the following modules: The first generation module is used to receive a 3D environment map containing static obstacle information and flight mission instructions issued by the cloud control platform, perform a global path search based on the mission start point, end point and flight energy consumption model, and generate a 3D global guidance corridor; it integrates data from the global positioning system, inertial measurement unit and visual sensor in real time, uses an extended Kalman filter to estimate the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle, and uploads the current pose state to the cloud control platform for monitoring in real time; The calculation module is used to detect local and static obstacles within the three-dimensional global guidance corridor using airborne sensors, predict the future trajectory of the obstacles to construct a spatiotemporal obstacle avoidance area, and calculate the local environmental complexity index. When the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, the module triggers global path replanning with the current position of the UAV as a new starting point and updates the replanned path result to the cloud control platform. The adjustment module is used to adjust the safety margin of the static obstacle repulsive potential field according to the determinant value of the pose estimation covariance matrix; and based on the local environment complexity index, adjust the weight coefficients of the global guiding corridor gravity, static obstacle repulsive force and spatiotemporal obstacle avoidance term in the multi-objective optimization problem, and solve for the optimal local expected velocity vector and the optimization cost of the solution. The second generation module is used to generate control commands based on the deviation between the optimal local desired velocity vector and the current pose state of the unmanned aerial vehicle, and by adjusting the gain parameters of the nonlinear controller according to the optimization cost.

[0012] Preferably, the step of performing a global path search based on the mission start point, end point, and flight energy consumption model to generate a three-dimensional global guidance corridor includes: In the cloud control platform or the onboard computer of the unmanned aerial vehicle, the three-dimensional environment map is discretized into grid cells, and a search algorithm is used with path integral cost. The search is performed with the minimum as the target, where Let ds be the path element, and ds be the path element. and Preset weights configured for the cloud control platform; Energy consumption per unit distance Based on the current payload m and preset cruise speed of the unmanned aerial vehicle and climb angle Calculation, in the form of ,in Where g is air resistance and g is the acceleration due to gravity. The optimal path point sequence found is expanded outward to form a tubular region, which serves as the three-dimensional global guidance corridor and is then sent to the local planning module.

[0013] Preferably, the step of estimating the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle using an extended Kalman filter includes: In the prediction step, based on the six-degree-of-freedom kinematic model of the unmanned aerial vehicle and the pose state at the previous moment, the prior pose state and prior covariance at the current moment are predicted. In the update step, the three-dimensional position provided by the Global Positioning System, the three-axis acceleration and angular velocity provided by the Inertial Measurement Unit, and the feature point depth information provided by the visual sensor are used as measurement values ​​to calculate the Kalman gain and update the prior pose state and prior covariance to obtain the posterior pose state estimate and pose estimation covariance matrix at the current time.

[0014] Preferably, predicting the future trajectory of the obstacle to construct a spatiotemporal obstacle avoidance area includes: For each detected obstacle, a motion model is used to predict the sequence of the obstacle's positions over a future period of time; Each predicted location point is expanded into a safety envelope, and the set of safety envelopes for all predicted time steps constitutes the spatiotemporal obstacle avoidance region of the obstacle.

[0015] Preferably, when the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, global path replanning is triggered with the current position of the unmanned aerial vehicle as a new starting point, including: Calculate the minimum distance between the predicted trajectory of the obstacle and the centerline of the three-dimensional global guidance corridor; A conflict risk value is calculated based on the minimum distance, and the conflict risk value increases as the minimum distance decreases; When the conflict risk value exceeds the first preset threshold issued by the cloud control platform, global path replanning is triggered.

[0016] Preferably, adjusting the safety margin of the static obstacle repulsive potential field based on the determinant value of the pose estimation covariance matrix includes: Based on the pose estimation covariance matrix determinant value Adjust the safety margin of the repulsive potential field of the static obstacle according to the following formula. ,in Based on the basic safety margin, k and These are the positive tuning parameters.

[0017] Preferably, adjusting the weight coefficients of the global guiding corridor gravity, static obstacle repulsion, and spatiotemporal obstacle avoidance term in the multi-objective optimization problem based on the local environment complexity index includes: Local environment complexity index This is the sum of the speeds of all obstacles within the sensor's detection range: Where N is the number of obstacles, Let be the velocity vector of the i-th obstacle; According to the index The weighting coefficients are adjusted such that when the exponent increases, the weighting coefficient of the spatiotemporal obstacle avoidance term increases, while the weighting coefficient of the global guidance corridor gravity term decreases.

[0018] Preferably, the step of adjusting the gain parameter of the nonlinear controller according to the optimized cost to generate control commands includes: A proportional-integral-derivative (PID) controller is used as the nonlinear controller. According to the optimized cost value The proportional gain of the PID controller is adjusted inversely using the following functional relationship. : ,in The baseline proportional gain set for the cloud control platform. It is a positive adjustment coefficient.

[0019] This invention plans a three-dimensional guidance corridor that balances flight energy consumption. It correlates the certainty of the UAV's own attitude estimation with the safety margin of static obstacles. When the attitude information is reliable, the UAV is allowed to pass through with smaller gaps, while the safety distance is increased when uncertainty increases. This improves flight efficiency and space utilization while ensuring safety. Furthermore, by predicting the future trajectory of moving targets and constructing a spatiotemporal avoidance zone, and by balancing the priority between global path tracking and local obstacle avoidance based on the complexity of the local environment, local decisions are made more context-aware. The cost of local planning is fed back to the controller, employing a smoother control response for high-cost complex maneuvers and a more agile response for low-cost simple cruises. This coupling of planning difficulty and control behavior improves the safety, stability, and mission execution efficiency of the UAV in complex environments. Attached Figure Description

[0020] Figure 1 A flowchart of the first embodiment; Figure 2 This is a schematic diagram of pose state estimation based on the extended Kalman filter. Figure 3 This is a schematic diagram of gain adjustment for a nonlinear controller based on optimization cost; Figure 4 A schematic diagram illustrating the adjustment logic for optimizing weights and control gain. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the first embodiment, the present invention proposes a cloud-based intelligent management method for low-altitude flight of unmanned aerial vehicles, such as... Figure 1 This includes the following steps: S1 receives a 3D environment map containing static obstacle information and flight mission instructions from the cloud control platform. Based on the mission start point, end point and flight energy consumption model, it performs a global path search to generate a 3D global guidance corridor. It integrates data from the Global Positioning System, Inertial Measurement Unit and visual sensor in real time, uses an extended Kalman filter to estimate the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle, and uploads the current pose state to the cloud control platform for monitoring in real time. Specifically, the cloud-based control platform generates an octree map of the environment using airborne LiDAR or offline methods and distributes it to the unmanned aerial vehicle (UAV). The flight energy consumption model is a function of flight power, flight speed, and acceleration, comprehensively considering hovering power, power to overcome air resistance, and power to change kinetic energy. An improved A-Star search algorithm is used in the cloud, replacing the traditional path length cost function with accumulated energy consumption. A sequence of energy-optimal path points from the starting point to the endpoint is obtained by searching the 3D rasterized map. Centered on this path point sequence, a series of interconnected convex polyhedra are generated by expanding outwards to the boundary of environmental obstacles. These convex polyhedra collectively constitute a 3D global guidance corridor.

[0023] The three-dimensional position, three-dimensional velocity, attitude quaternions of the unmanned aerial vehicle (UAV), and the bias of the inertial measurement unit (IMU) are used as the system's state vector. In the prediction phase of the extended Kalman filter (EPF), the angular velocity and acceleration measurements from the IMU are used to predict the state vector and covariance for the next moment using the UAV's kinematics and dynamics model. In the update phase, when the absolute position measurement from the GPS or the relative pose transformation provided by the visual odometry is received, the residual between the measured and predicted values ​​is calculated, and the Kalman gain is calculated accordingly. This corrects the estimated state vector and reduces the covariance matrix, resulting in a fused high-precision pose state and a covariance matrix representing the current estimation uncertainty, such as... Figure 2 The merged status data is then transmitted back to the cloud in real time via the communication module.

[0024] In an optional embodiment, the step of performing a global path search based on the mission start point, end point, and flight energy consumption model to generate a three-dimensional global guidance corridor includes: In the cloud control platform or the onboard computer of the unmanned aerial vehicle, the three-dimensional environment map is discretized into grid cells, and a search algorithm is used with path integral cost. The search is performed with the minimum as the target, where Let ds be the path element, and ds be the path element. and Preset weights configured for the cloud control platform; Energy consumption per unit distance Based on the current payload m and preset cruise speed of the unmanned aerial vehicle and climb angle Calculation, in the form of ,in Where g is air resistance and g is the acceleration due to gravity. The optimal path point sequence found is expanded outward to form a tubular region, which serves as the three-dimensional global guidance corridor and is then sent to the local planning module.

[0025] A 100m × 100m × 50m three-dimensional mission space is discretized into a 1m square grid map. Unmanned aerial vehicle (UAV) parameters are set, such as a payload of 2kg and a cruising speed. The air resistance is 10 m / s. The weight is 0.5N. It is 0.1. The value is 0.9. Based on the aforementioned parameters, the energy consumption per unit distance under different flight conditions is calculated, such as the climb angle during level flight. The value is 0 degrees, and the calculation yields... The cost is 0.5 J / m, and the cost of moving from one grid cell to an adjacent horizontal grid cell is 0.55; when flying at a climb angle of 15 degrees, The cost is approximately 5.57 J / m, and the cost of moving to the adjacent upper grid is approximately 5.11.

[0026] The A* search algorithm is employed, using the mission's starting grid as the initial point and the ending grid as the target point. The algorithm searches for the path with the minimum cumulative cost in the grid map, using the calculated inter-grid movement cost as the evaluation criterion. It prioritizes exploring directions with the lowest combined energy consumption and flight distance cost, generating an optimal path point sequence composed of grid center points, for example... , , until The path point sequence is then expanded, for example, by generating a sphere with a radius of 5m centered on each point on the path. All spheres are connected and enclosed to form a smooth, tubular safe flight area, i.e., a three-dimensional global guidance corridor.

[0027] In an optional embodiment, estimating the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle using an extended Kalman filter includes: In the prediction step, based on the six-degree-of-freedom kinematic model of the unmanned aerial vehicle and the pose state at the previous moment, the prior pose state and prior covariance at the current moment are predicted. In the update step, the three-dimensional position provided by the Global Positioning System, the three-axis acceleration and angular velocity provided by the Inertial Measurement Unit, and the feature point depth information provided by the visual sensor are used as measurement values ​​to calculate the Kalman gain and update the prior pose state and prior covariance to obtain the posterior pose state estimate and pose estimation covariance matrix at the current time.

[0028] The state vector of the extended Kalman filter It includes the UAV's three-dimensional position, three-dimensional velocity, and a quaternion representing its attitude. In the prediction step, the posterior pose state from the previous time step, i.e., time k-1, is obtained. and covariance matrix The acceleration and angular velocity readings at the current time (k) obtained from the inertial measurement unit are used as input. By integrating the state from the previous time step using the six-degree-of-freedom kinematic equation f, the prior pose state at the current time step can be predicted. Simultaneously, the Jacobian matrix is ​​obtained by linearizing the kinematic equations. And combined with process noise covariance To predict the prior covariance matrix .

[0029] In the update step, measurements from multiple sensors are acquired. For example, the three-dimensional position (latitude, longitude, and altitude) provided by the Global Positioning System (GPS), or the three-dimensional position calculated by visual odometry based on the depth of image feature points. The residual between the measured value and the predicted value from the prior state is calculated. , where h is the measurement function. Calculate the Kalman gain. ,in It is the Jacobian matrix of the measurement function. It measures the noise covariance matrix. Kalman gain is used to correct the prior state and covariance, yielding the optimal posterior state estimate at the current time that incorporates information from all sensors. and the corresponding pose estimation covariance matrix .

[0030] S2, within the three-dimensional global guidance corridor, airborne sensors are used to detect local and static obstacles, predict the future trajectory of the obstacles to construct a spatiotemporal obstacle avoidance area, and calculate the local environmental complexity index; when the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, global path replanning is triggered with the current position of the unmanned aerial vehicle as a new starting point, and the replanned path result is updated to the cloud control platform; Specifically, the local environment is perceived using LiDAR or depth cameras, and the obstacle point cloud is segmented into individual units using point cloud clustering algorithms such as DBSCAN. For identified obstacles, a Kalman filter is used to continuously track them, estimating their current velocity and acceleration, and predicting their trajectory over the next few seconds based on a uniform or uniformly accelerated motion model. This predicted trajectory is then subjected to a Minkowski sum operation with the geometric dimensions to form a four-dimensional spatiotemporal channel serving as the obstacle avoidance area. In one embodiment, the local environment complexity index is calculated by weighting the number of obstacles within the field of view, their average velocity, and their average distance from the UAV. The conflict risk value is represented by calculating the volume or probability of the intersection between the predicted spatiotemporal channel of the obstacle and the convex polyhedron of the global guidance corridor. Once this value exceeds a threshold set in the cloud, it indicates that the global path has severely failed, and a re-search for the global path is immediately initiated, with a new path synchronized to the cloud.

[0031] In an optional embodiment, predicting the future trajectory of the obstacle to construct a spatiotemporal obstacle avoidance region includes: For each detected obstacle, a motion model is used to predict the sequence of the obstacle's positions over a future period of time; Each predicted location point is expanded into a safety envelope, and the set of safety envelopes for all predicted time steps constitutes the spatiotemporal obstacle avoidance region of the obstacle.

[0032] For each obstacle detected by the sensor, such as a car, a constant-speed motion model is used for trajectory prediction. Assuming at the current moment... The location of the car was detected as follows: m, estimated velocity is m / s. Define a prediction time domain, such as the next 5 seconds, and set the prediction time step to 0.5 seconds.

[0033] Calculate the future position iteratively by step size. At what time, the predicted position is .exist At what time, the predicted position is And so on, generating a time series containing 10 future location points. To account for the volume of obstacles and the uncertainty of the prediction, each predicted location point... Expanding to a safety envelope, such as a sphere with a radius of 3m. The spatiotemporal obstacle avoidance zone of this obstacle is determined by the series of future moments. Composed of spheres occupying different spatial positions, it forms a tubular no-fly zone that extends in space and time.

[0034] In an optional embodiment, when the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, global path replanning is triggered with the current position of the unmanned aerial vehicle as a new starting point, including: Calculate the minimum distance between the predicted trajectory of the obstacle and the centerline of the three-dimensional global guidance corridor; A conflict risk value is calculated based on the minimum distance, and the conflict risk value increases as the minimum distance decreases; When the conflict risk value exceeds the first preset threshold issued by the cloud control platform, global path replanning is triggered.

[0035] The predicted trajectory of the obstacle over the next 5 seconds is obtained, which is a 3D polyline composed of multiple line segments. Simultaneously, the centerline of the planned 3D global guidance corridor is also obtained, which is a 3D polyline. Through geometric calculations, all line segment pairs of these two polylines are traversed to find their shortest distance in 3D space. For example, calculations showed that the predicted trajectory of the obstacle would be closest to the center line of the guiding corridor in the next 3 seconds, with a minimum distance of 12m.

[0036] The conflict risk value is calculated based on this minimum distance, and an inverse proportional function or a linear function is used to assess the risk. For example, a warning distance of 25m can be set. The formula for calculating the risk value is: For a minimum distance of 12m, the calculated risk value is... The calculated risk value is compared with a preset threshold issued by the cloud, for example, the first preset threshold is 0.5. Since the current risk value of 0.52 is greater than the threshold of 0.5, the system determines that there is a high risk of collision and immediately triggers global path replanning. The replanning process will use the current real-time position of the UAV as the new path starting point, and re-invoke the global path search algorithm to generate a new 3D global guidance corridor that can avoid the predicted conflict area.

[0037] S3. Based on the determinant value of the pose estimation covariance matrix, adjust the safety margin of the static obstacle repulsive potential field; and based on the local environment complexity index, adjust the weight coefficients of the global guiding corridor gravity, static obstacle repulsive force, and spatiotemporal obstacle avoidance term in the multi-objective optimization problem, and solve for the optimal local expected velocity vector and the optimization cost of the solution. Specifically, the determinant of the pose estimation covariance matrix represents the magnitude of pose uncertainty. The safety margin for static obstacles is set as a base value plus a term positively correlated with this determinant value; that is, the greater the pose uncertainty, the larger the required safety margin. A multi-objective cost function is constructed, comprising three terms: a guidance term, which penalizes the deviation of the desired velocity direction from the direction pointing to the centerline of the global guidance corridor; a static obstacle avoidance term, based on the principle of the artificial potential field method, which generates a large repulsive force cost when the distance between the UAV and the static obstacle is less than the adjusted safety margin; and an obstacle avoidance term, which penalizes the probability of the desired velocity causing a collision between the future trajectory and the obstacle's spatiotemporal region. Based on the aforementioned complexity index, a high index increases the weight of the obstacle avoidance term to prioritize safety; a low index increases the weight of the guidance term to prioritize flight efficiency. The minimum value of this cost function is obtained through numerical optimization methods, yielding the optimal local desired velocity vector. The minimum value of this cost function is the optimization cost.

[0038] In an optional embodiment, adjusting the safety margin of the static obstacle repulsive potential field based on the determinant value of the pose estimation covariance matrix includes: Based on the pose estimation covariance matrix determinant value Adjust the safety margin of the repulsive potential field of the static obstacle according to the following formula. ,in Based on the basic safety margin, k and These are the positive tuning parameters.

[0039] Obtain the pose estimation covariance matrix at the current time step from the extended Kalman filter module. And calculate the determinant value. The magnitude of this determinant directly reflects the degree of uncertainty in the unmanned aerial vehicle's position estimation; the larger the value, the higher the uncertainty. For example, when the GPS signal is good... It might be a small value, such as 0.01; however, when flying in canyons where signals are blocked, relying on visual and inertial navigation, uncertainty accumulates. It may increase to 0.8.

[0040] Use preset parameters, such as base safety margin. Set to 2m, and the tuning parameter k to 5. Set it to 0.5, and calculate the safety margin according to the formula. When the signal is good, m. When the signal is poor. m. Calculated The value is input into the obstacle repulsion potential field model of the local path planner in real time, indicating that when the UAV is not very certain about its own position, it will automatically maintain a greater distance from static obstacles, thereby improving flight safety.

[0041] In an optional embodiment, adjusting the weight coefficients of the global guiding corridor gravity, static obstacle repulsion, and spatiotemporal obstacle avoidance term in the multi-objective optimization problem based on the local environment complexity index includes: Local environment complexity index This is the sum of the speeds of all obstacles within the sensor's detection range: Where N is the number of obstacles, Let be the velocity vector of the i-th obstacle; According to the index The weighting coefficients are adjusted such that when the exponent increases, the weighting coefficient of the spatiotemporal obstacle avoidance term increases, while the weighting coefficient of the global guidance corridor gravity term decreases.

[0042] Calculate the complexity index of the local environment in real time. Assume the UAV's sensors detect two obstacles ahead: obstacle A has a velocity of 3 m / s and obstacle B has a velocity of 5 m / s. What is the complexity index at this point? =8. If entering a more complex traffic intersection, four obstacles are detected with speeds of 2, 4, 5, and 6 m / s respectively, then the complexity index is... =17.

[0043] The weights of each term in the multi-objective optimization cost function of the local path planner are adjusted based on this index. For example, the base weight of the spatiotemporal obstacle avoidance term configured in the cloud is 0.4, and the base weight of the global guidance corridor gravity term is 0.5. The weight adjustment function can be planned as follows: Linear changes, for example, the new weights of the obstacle avoidance term are The new weights of the corridor gravity term are In the first case, , Adjusted to 0.56. Adjusted to 0.42. In the second complex case, , Adjusted to 0.74, The value has been adjusted to 0.33. This adjustment allows the UAV to prioritize threat avoidance when the environment becomes complex, even if the described path temporarily deviates from the optimal global path.

[0044] S4. Based on the deviation between the optimal local desired velocity vector and the current pose state of the unmanned aerial vehicle, and according to the optimization cost, adjust the gain parameter of the nonlinear controller to generate control commands.

[0045] Specifically, a cascaded PID controller is used as the nonlinear controller, with the outer loop being a position controller and the inner loop an attitude controller. The optimal local desired velocity vector obtained by solving is used as the input target of the outer loop controller, and the deviation between the optimal local desired velocity vector and the current estimated velocity is used as the velocity error. The proportional gain of the controller... Integral gain and differential gain Instead of being a fixed value, it is a function that optimizes the cost value, for example, Where J is the optimization cost and a is the adjustment coefficient. A high optimization cost indicates a complex and difficult obstacle avoidance task, resulting in a corresponding decrease in controller gain. This leads to a smoother and more conservative UAV response, avoiding instability caused by violent maneuvers. Conversely, a low optimization cost indicates a simple environment, leading to an increase in controller gain, allowing the UAV to track the desired speed more quickly. The controller calculates the desired thrust and attitude angle based on the adjusted gain and speed error, and then translates this into control commands for the motor speeds of the UAV's four rotors, such as... Figure 3 .

[0046] In an optional embodiment, adjusting the gain parameter of the nonlinear controller according to the optimized cost to generate control commands includes: A proportional-integral-derivative (PID) controller is used as the nonlinear controller. According to the optimized cost value The proportional gain of the PID controller is adjusted inversely using the following functional relationship. : ,in The baseline proportional gain set for the cloud control platform. It is a positive adjustment coefficient.

[0047] In each planning cycle, the local path planner outputs a comprehensive optimization cost after solving for the optimal trajectory. The value reflects the quality and execution difficulty of the generated trajectory. For example, when a drone flies smoothly along a guide corridor in an open area, the penalties in the cost function are very small, resulting in... It might be version 2.0. However, when a drone needs to perform emergency obstacle avoidance maneuvers between two nearby buildings, the trajectory curvature is large and it's close to the obstacle, resulting in accumulated penalties. Increase, for example, to 10.0.

[0048] Use this value to adjust the proportional gain of the PID controller Set the reference proportional gain. The adjustment coefficient is 2.5. It is 0.08. In low-cost scenarios, The calculated new proportional gain is At this point, the controller remains relatively responsive and can accurately track simple trajectories. However, in high-cost scenarios... The new proportional gain is At this point, the proportional gain is lowered, resulting in a smoother controller response. This avoids flight attitude instability or overshoot in narrow and complex environments caused by overly aggressive control commands, thus ensuring flight smoothness and safety under demanding maneuvers. Figure 4 .

[0049] In the second embodiment, the present invention also proposes a cloud-based intelligent management system for low-altitude flight of unmanned aerial vehicles, comprising the following modules: The first generation module is used to receive a 3D environment map containing static obstacle information and flight mission instructions issued by the cloud control platform, perform a global path search based on the mission start point, end point and flight energy consumption model, and generate a 3D global guidance corridor; it integrates data from the global positioning system, inertial measurement unit and visual sensor in real time, uses an extended Kalman filter to estimate the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle, and uploads the current pose state to the cloud control platform for monitoring in real time; The calculation module is used to detect local and static obstacles within the three-dimensional global guidance corridor using airborne sensors, predict the future trajectory of the obstacles to construct a spatiotemporal obstacle avoidance area, and calculate the local environmental complexity index. When the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, the module triggers global path replanning with the current position of the UAV as a new starting point and updates the replanned path result to the cloud control platform. The adjustment module is used to adjust the safety margin of the static obstacle repulsive potential field according to the determinant value of the pose estimation covariance matrix; and based on the local environment complexity index, adjust the weight coefficients of the global guiding corridor gravity, static obstacle repulsive force and spatiotemporal obstacle avoidance term in the multi-objective optimization problem, and solve for the optimal local expected velocity vector and the optimization cost of the solution. The second generation module is used to generate control commands based on the deviation between the optimal local desired velocity vector and the current pose state of the unmanned aerial vehicle, and by adjusting the gain parameters of the nonlinear controller according to the optimization cost.

[0050] In this specification, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise limited, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the associated listed items.

[0051] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0052] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cloud-based intelligent management method for low-altitude flight of unmanned aerial vehicles, characterized in that, Includes the following steps: The system receives a 3D environmental map containing static obstacle information and flight mission instructions from the cloud control platform. Based on the mission start point, end point, and flight energy consumption model, it performs a global path search to generate a 3D global guidance corridor. It also integrates data from the Global Positioning System, Inertial Measurement Unit, and Visual Sensor in real time, uses an Extended Kalman Filter to estimate the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle, and uploads the current pose state to the cloud control platform for monitoring in real time. Within the three-dimensional global guidance corridor, airborne sensors are used to detect local and static obstacles, predict the future trajectories of the obstacles to construct a spatiotemporal obstacle avoidance zone, and calculate the local environmental complexity index. When the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, global path replanning is triggered with the current position of the unmanned aerial vehicle as a new starting point, and the replanned path result is updated to the cloud control platform. Based on the determinant value of the pose estimation covariance matrix, the safety margin of the static obstacle repulsive potential field is adjusted; and based on the local environment complexity index, the weight coefficients of the global guiding corridor gravity, static obstacle repulsive force, and spatiotemporal obstacle avoidance term in the multi-objective optimization problem are adjusted to obtain the optimal local expected velocity vector and the optimization cost of the solution. Based on the deviation between the optimal local desired velocity vector and the current pose state of the unmanned aerial vehicle, and according to the optimization cost, the gain parameters of the nonlinear controller are adjusted to generate control commands.

2. The method according to claim 1, characterized in that, The process of performing a global path search based on the mission start point, end point, and flight energy consumption model to generate a three-dimensional global guidance corridor includes: In the cloud control platform or the onboard computer of the unmanned aerial vehicle, the three-dimensional environment map is discretized into grid cells, and a search algorithm is used with path integral cost. The search is performed with the minimum as the objective, where P is the path and ds is the path element. and Preset weights configured for the cloud control platform; Energy consumption per unit distance Based on the current payload m and preset cruise speed of the unmanned aerial vehicle and climb angle Calculation, in the form of ,in Where g is air resistance and g is the acceleration due to gravity. The optimal path point sequence found is expanded outward to form a tubular region, which serves as the three-dimensional global guidance corridor and is then sent to the local planning module.

3. The method according to claim 1, characterized in that, The method of estimating the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle using an extended Kalman filter includes: In the prediction step, based on the six-degree-of-freedom kinematic model of the unmanned aerial vehicle and the pose state at the previous moment, the prior pose state and prior covariance at the current moment are predicted. In the update step, the three-dimensional position provided by the Global Positioning System, the three-axis acceleration and angular velocity provided by the Inertial Measurement Unit, and the feature point depth information provided by the visual sensor are used as measurement values ​​to calculate the Kalman gain and update the prior pose state and prior covariance to obtain the posterior pose state estimate and pose estimation covariance matrix at the current time.

4. The method according to claim 1, characterized in that, The prediction of the future trajectory of the obstacle to construct a spatiotemporal obstacle avoidance area includes: For each detected obstacle, a motion model is used to predict the sequence of the obstacle's positions over a future period of time; Each predicted location point is expanded into a safety envelope, and the set of safety envelopes for all predicted time steps constitutes the spatiotemporal obstacle avoidance region of the obstacle.

5. The method according to claim 1, characterized in that, When the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, a global path replanning is triggered with the current position of the unmanned aerial vehicle as the new starting point, including: Calculate the minimum distance between the predicted trajectory of the obstacle and the centerline of the three-dimensional global guidance corridor; A conflict risk value is calculated based on the minimum distance, and the conflict risk value increases as the minimum distance decreases; When the conflict risk value exceeds the first preset threshold issued by the cloud control platform, global path replanning is triggered.

6. The method according to claim 1, characterized in that, The step of adjusting the safety margin of the static obstacle repulsive potential field based on the determinant value of the pose estimation covariance matrix includes: Based on the pose estimation covariance matrix determinant value Adjust the safety margin of the repulsive potential field of the static obstacle according to the following formula. ,in Based on the basic safety margin, k and These are the positive tuning parameters.

7. The method according to claim 1, characterized in that, The adjustment of the weight coefficients of the global guiding corridor gravity, static obstacle repulsion, and spatiotemporal obstacle avoidance term in the multi-objective optimization problem based on the local environment complexity index includes: Local environment complexity index This is the sum of the speeds of all obstacles within the sensor's detection range: Where N is the number of obstacles, Let be the velocity vector of the i-th obstacle; According to the index The weighting coefficients are adjusted such that when the exponent increases, the weighting coefficient of the spatiotemporal obstacle avoidance term increases, while the weighting coefficient of the global guidance corridor gravity term decreases.

8. The method according to claim 1, characterized in that, The step of adjusting the gain parameters of the nonlinear controller according to the optimized cost to generate control commands includes: A proportional-integral-derivative (PID) controller is used as the nonlinear controller. According to the optimized cost value The proportional gain of the PID controller is adjusted inversely using the following functional relationship. : ,in The baseline proportional gain set for the cloud control platform. It is a positive adjustment coefficient.

9. A cloud-based intelligent management system for low-altitude flight of unmanned aerial vehicles, characterized in that, Includes the following modules: The first generation module is used to receive a 3D environment map containing static obstacle information and flight mission instructions issued by the cloud control platform, perform a global path search based on the mission start point, end point and flight energy consumption model, and generate a 3D global guidance corridor; it integrates data from the global positioning system, inertial measurement unit and visual sensor in real time, uses an extended Kalman filter to estimate the current pose state and pose estimation covariance matrix of the unmanned aerial vehicle, and uploads the current pose state to the cloud control platform for monitoring in real time; The calculation module is used to detect local and static obstacles within the three-dimensional global guidance corridor using airborne sensors, predict the future trajectory of the obstacles to construct a spatiotemporal obstacle avoidance area, and calculate the local environmental complexity index. When the conflict risk value between the predicted trajectory of the obstacle and the guidance corridor exceeds a first preset threshold, the module triggers global path replanning with the current position of the UAV as a new starting point and updates the replanned path result to the cloud control platform. The adjustment module is used to adjust the safety margin of the static obstacle repulsive potential field according to the determinant value of the pose estimation covariance matrix; and based on the local environment complexity index, adjust the weight coefficients of the global guiding corridor gravity, static obstacle repulsive force and spatiotemporal obstacle avoidance term in the multi-objective optimization problem, and solve for the optimal local expected velocity vector and the optimization cost of the solution. The second generation module is used to generate control commands based on the deviation between the optimal local desired velocity vector and the current pose state of the unmanned aerial vehicle, and by adjusting the gain parameters of the nonlinear controller according to the optimization cost.

10. The system according to claim 9, characterized in that, The process of performing a global path search based on the mission start point, end point, and flight energy consumption model to generate a three-dimensional global guidance corridor includes: In the cloud control platform or the onboard computer of the unmanned aerial vehicle, the three-dimensional environment map is discretized into grid cells, and a search algorithm is used with path integral cost. The search is performed with the minimum as the target, where Let ds be the path element, and ds be the path element. and Preset weights configured for the cloud control platform; Energy consumption per unit distance Based on the current payload m and preset cruise speed of the unmanned aerial vehicle and climb angle Calculation, in the form of ,in Where g is air resistance and g is the acceleration due to gravity. The optimal path point sequence found is expanded outward to form a tubular region, which serves as the three-dimensional global guidance corridor and is then sent to the local planning module.

Images