Multi-dimensional flight navigation system and method for unmanned aerial vehicle

Abstract

The invention discloses an unmanned aerial vehicle multi-dimensional flight navigation system and method, and belongs to the technical field of unmanned aerial vehicle navigation control. Aiming at the problems of unstable positioning and difficulty in considering safety and energy consumption of a path under the conditions of indoor and outdoor complex scenes and limited satellite signals, a multi-source sensing module is used for synchronously acquiring data such as inertia, height and course, a multi-dimensional state is constructed, target weights such as distance, energy consumption, attitude stability, time and safety are adaptively adjusted according to environmental uncertainty, and the positioning accuracy is improved. Re-planning is triggered when the environment changes; the system can be expanded to a multi-machine cooperation sharing environment model and path. According to the invention, the navigation safety and the energy efficiency are enhanced, and the real-time performance and the embedded deployment value are good.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) flight control and navigation technology, and in particular to a multi-dimensional flight navigation system for UAVs. Background Technology

[0002] With the widespread application of unmanned aerial vehicle (UAV) technology in surveying and inspection, emergency rescue, logistics and transportation, agricultural monitoring, and urban management, the autonomous navigation performance of UAVs has become an important indicator of their intelligence and mission execution capabilities. Traditional UAV navigation systems typically rely on the Global Positioning System (GPS) or the BeiDou Satellite System (BDS) for positioning, and combine them with sensors such as inertial measurement units (IMUs), barometers, and magnetometers to achieve attitude estimation and altitude measurement. However, in complex environments such as urban canyons, forests, tunnels, and indoor spaces, satellite signals often attenuate or are lost due to obstruction and multipath effects, leading to a significant decrease in positioning accuracy or even complete failure, thereby affecting the flight stability and mission safety of the UAV.

[0003] To address scenarios with limited satellite signals, researchers and engineers have proposed various multi-sensor fusion navigation methods, such as combining visual sensors with lidar, or employing visual inertial odometry (IMU) and laser inertial odometry (LAO) to fuse IMU data with visual or lidar data to improve attitude calculation accuracy. These methods have improved the navigation robustness of UAVs in complex environments to some extent, but several limitations remain: Most current navigation systems still model their states in the three-dimensional space of position and velocity, lacking joint consideration of high-dimensional state information such as energy consumption and environmental disturbances, thus limiting the system's ability to balance flight performance, safety, and energy consumption optimization. Simultaneously, existing 3D mapping and environmental perception technologies primarily focus on geometric reconstruction, failing to fully utilize the semantic information of the scene, making it difficult for UAVs to achieve semantic understanding of avionics, hazardous areas, and dynamic obstacles. In terms of path planning, traditional navigation algorithms generally use the shortest distance or shortest time as a single optimization objective, lacking a comprehensive consideration of multiple objectives such as energy efficiency, attitude stability, and safety margin, making it difficult for UAVs to achieve globally optimal performance in actual flight. Furthermore, when faced with dynamic environments such as wind disturbances, obstacle movement, or changes in mission objectives, the path planning mechanism of existing navigation systems is slow to respond and lacks an adaptive mechanism for weight adjustment and path replanning, which can easily lead to flight trajectory deviation or abnormal energy consumption.

[0004] In summary, existing UAV navigation systems generally suffer from problems such as limited navigation dimensions, incomplete perception information, single optimization target, insufficient adaptability to dynamic environments, and a lack of multi-UAV collaborative mechanisms. These shortcomings severely restrict the autonomous navigation capabilities and mission execution efficiency of UAVs in complex environments. Summary of the Invention

[0005] At present, there is an urgent need for a flight navigation system that can integrate information from multiple sensor sources, construct a semantic 3D environment model, and achieve information entropy adaptive multi-objective optimization in a multi-dimensional state space, so as to realize high-precision positioning, dynamic path planning and safe and reliable autonomous flight control of UAVs in complex scenarios.

[0006] To achieve the above objectives, the present invention provides a multi-dimensional flight navigation system for unmanned aerial vehicles (UAVs), comprising:

[0007] A multi-sensor module is used to collect flight status data and environmental data of the UAV. The multi-sensor module includes an inertial measurement unit, a lidar sensor, a vision camera, an ultra-wideband positioning sensor, a barometer, and a magnetometer.

[0008] An environmental perception and modeling module, connected to the multi-sensor module, is used to construct a 3D environment model with semantic information and its confidence level based on a vision-geometry fusion network. The vision-geometry fusion network includes a visual feature extraction branch and a geometric structure reconstruction branch. The visual feature extraction branch extracts scene semantic features based on image data collected by the vision camera and outputs semantic confidence level. The geometric structure reconstruction branch reconstructs the scene geometric structure based on point cloud data collected by the lidar sensor. The outputs of the two branches are processed by a feature fusion layer to generate the 3D environment model. The 3D environment model includes the geometric shape, spatial location, semantic category, and probability distribution parameters of obstacles.

[0009] The positioning module, connected to the multi-sensor module, is used to determine the real-time position and attitude of the UAV by fusing data from the inertial measurement unit, the ultra-wideband positioning sensor, the visual camera, and the lidar sensor in a GPS-restricted environment. The positioning module uses a deterministic observer based on the SE(3) group to perform state estimation on the Lie group manifold. The deterministic observer based on the SE(3) group outputs the position vector, attitude rotation matrix, and state estimation uncertainty of the UAV in the world coordinate system by fusing the acceleration and angular velocity data provided by the inertial measurement unit, the distance measurement data provided by the ultra-wideband positioning sensor, and the relative pose data provided by the visual camera or the lidar sensor.

[0010] The path planning module, connected to the environment perception and modeling module and the positioning module, is used to generate the optimal flight path in the three-dimensional environment model based on a multi-objective optimization navigation model in a multi-dimensional state space. The multi-objective optimization navigation model in the multi-dimensional state space uses state vectors. Describe the multidimensional state of the drone;

[0011] Where x, y, and z are the three-dimensional spatial coordinates, and v_x, v_y, and v_z are the three-dimensional velocity components. θ and ψ are attitude angles, E is residual energy, R is safety margin, and W is environmental disturbance parameter; the path planning module determines the control vector by solving for the minimum value of a multi-objective cost function with an adaptive weighting mechanism based on information entropy. Where T is the thrust. ω_θ and f_ψ are the roll rate control, pitch rate control, and yaw rate control, respectively; the information entropy adaptive weighting mechanism dynamically adjusts the weight coefficients of each cost term in the multi-objective cost function based on the semantic confidence and the state estimation uncertainty;

[0012] A control execution module, connected to the path planning module, is used to realize trajectory tracking and dynamic obstacle avoidance based on the control vector;

[0013] The collaboration module, connected to the path planning module and the environment perception and modeling module, is used to realize information sharing, collaborative environmental modeling, and cluster path coordination in a multi-UAV system.

[0014] The multi-objective cost function is expressed as:

[0015]

[0016] Where D is the path length cost term, E c Let S be the energy consumption cost, and T be the attitude stability cost. f R is the flight time cost term. -1 The reciprocal cost term for safety margin, w1, w2, w3, w4, and w5 are the corresponding weighting coefficients, and T is the total flight time.

[0017] The mathematical expression for the information entropy adaptive weighting mechanism is:

[0018] w i (t+1)=exp(-H i (t)) / ∑ j exp(-H j (t))

[0019] Among them, H i (t) represents the information entropy of the i-th cost term at time t, calculated using the following formula:

[0020] H i (t)=-∑ k p ik (t)log p ik (t)

[0021] Where, p ik(t) represents the probability distribution value of the i-th cost term in the k-th state at time t, which is obtained by comprehensively calculating the semantic confidence and the state estimation uncertainty.

[0022] The path length cost term D is calculated by summing the Euclidean distances between each waypoint on the planned path.

[0023] The energy consumption cost item E C Calculated based on the product of the thrust T and the flight time;

[0024] The attitude stability cost term S is based on the attitude angle. The deviations of θ and ψ from the desired attitude and the angular velocity Calculation of the combined measurement of ω_θ and ω_ψ;

[0025] The flight time cost term T f This represents the estimated time of arrival from the current moment to the target location.

[0026] The safety margin reciprocal cost term R-1 is calculated based on the reciprocal of the distance between the drone and the nearest obstacle.

[0027] The path planning module uses a multi-objective reinforcement learning algorithm or an improved RRT algorithm to generate the optimal flight path. The multi-objective reinforcement learning algorithm trains the agent to learn a strategy that balances path length, energy consumption, attitude stability, flight time, and safety margin in the multi-dimensional state space. The improved RRT algorithm introduces the multi-objective cost function as a path evaluation criterion during the random tree expansion process, selects the node that minimizes the cost function value for tree expansion, and realizes path optimization in three-dimensional space.

[0028] The control execution module uses model predictive control to achieve trajectory tracking and dynamic obstacle avoidance; the model predictive control method uses the multi-objective cost function as the optimization objective in each control cycle and outputs control commands that satisfy the dynamic constraints and environmental safety constraints of the UAV.

[0029] The collaboration module shares the state vector, weight coefficients, local 3D environment models, and obstacle location information among multiple UAVs via a wireless communication network. Each UAV merges multiple local 3D environment models into a global environment model based on the received information and updates its own weight coefficients and multi-objective cost function parameters. During path planning, the module coordinates the flight paths of each UAV to avoid mid-air collisions and mission conflicts.

[0030] The system supports dynamic switching between indoor and outdoor environments. In the indoor environment, the positioning module mainly relies on the ultra-wideband positioning sensor, the visual camera, and the lidar sensor for positioning. In the outdoor environment, the positioning module further integrates GPS signals and barometer data. The path planning module adjusts the initial value of the weight coefficient and the update frequency of the information entropy adaptive weight mechanism according to the environment type identifier.

[0031] The path planning module monitors the UAV's flight status and environmental changes in real time. When any of the following conditions are met, the path replanning process is triggered: a new obstacle is detected within the safe range of the currently planned path, the environmental disturbance parameter W exceeds a preset threshold, the safety margin R is lower than a preset minimum value, or the mission target position changes. After triggering replanning, the information entropy adaptive weighting mechanism is re-executed to update the weight coefficients, the minimum value of the multi-objective cost function is recalculated, and an updated flight path is generated.

[0032] A UAV flight navigation method based on a multi-objective optimization navigation model in a multi-dimensional state space includes the following steps:

[0033] Step S1: Collect flight status data and environmental data of the UAV through a multi-sensor module, which includes an inertial measurement unit, a lidar sensor, a vision camera, an ultra-wideband positioning sensor, a barometer, and a magnetometer.

[0034] Step S2: Process the environmental data based on the visual-geometric fusion network to construct a three-dimensional environment model with semantic information and confidence. The visual feature extraction branch of the visual-geometric fusion network extracts scene semantic features and outputs semantic confidence, while the geometric structure reconstruction branch reconstructs the scene geometric structure. The outputs of the two branches are processed by the feature fusion layer to generate the three-dimensional environment model.

[0035] Step S3: Using a deterministic observer based on the SE(3) group to fuse the flight state data, perform state estimation on the Lie group manifold, and determine the current state vector of the UAV. The state estimation uncertainty is given by x, y, z, which are the three-dimensional spatial coordinates, and v_x, v_y, v_z, which are the three-dimensional velocity components. θ and ψ are attitude angles, E is remaining energy, R is safety margin, and W is environmental disturbance parameter;

[0036] Step S4: Calculate the information entropy H of each cost term based on the semantic confidence and the state estimation uncertainty. i (t)=-Σ k p ij (t)log p ik(t), and updates the weight coefficients in the multi-objective cost function through an information entropy adaptive weighting mechanism, with the update formula being w. i (t+1)=exp(-H i (t)) / ∑ j exp(-H j (t));

[0037] Step S5: Construct the multi-objective cost function Where D is the path length cost term, E C Let S be the energy consumption cost, and T be the attitude stability cost. f R is the flight time cost term. -1 The cost term is the inverse of the safety margin.

[0038] Step S6: Using a multi-objective reinforcement learning algorithm or an improved RRT algorithm, solve for the optimal flight path that minimizes the multi-objective cost function in the three-dimensional environment model, and calculate the corresponding control vector. Where T is the thrust. ω_θ and ω_ψ are the roll rate control, pitch rate control, and yaw rate control, respectively.

[0039] Step S7: Using the model predictive control method, flight control commands are generated based on the control vector to drive the UAV to achieve trajectory tracking and dynamic obstacle avoidance;

[0040] Step S8: Monitor the drone's status and environmental changes in real time. When a new obstacle is detected within the safe range, the environmental disturbance parameter exceeds the threshold, the safety margin is lower than the minimum value, or the mission objective changes, return to step S2 to perform path replanning. Otherwise, return to step S1 to continue execution until the target location is reached.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0042] The multi-dimensional flight navigation system for unmanned aerial vehicles (UAVs) proposed in this invention can achieve higher precision and stability in autonomous flight navigation in complex or changing environments, representing a significant improvement over existing technologies. This invention establishes a multi-dimensional navigation model that includes parameters such as position, attitude, velocity, energy, safety margin, and environmental disturbances. This allows the UAV to simultaneously consider the balance between flight performance, safety, and energy consumption during flight, thereby maintaining good control stability and mission execution efficiency in different scenarios.

[0043] In terms of environmental perception, this invention integrates visual images and laser ranging information to establish a three-dimensional environmental model with semantic information. This enables the UAV to not only identify the location and shape of obstacles but also distinguish between flyable and hazardous areas, thereby improving environmental understanding and obstacle avoidance accuracy. In terms of positioning, this invention integrates an inertial measurement unit, an ultra-wideband positioning sensor, and visual or laser ranging data to achieve joint estimation of the UAV's position and attitude. Even in environments with weak satellite signals, it maintains stable flight positioning accuracy, avoiding the attitude drift problem that is common in traditional methods.

[0044] In terms of path planning and control, this invention introduces a multi-objective optimization method. When calculating the flight path, it comprehensively considers factors such as distance, time, energy consumption, safety, and flight attitude. By dynamically adjusting the weights of each factor, the system can automatically select a safer or more energy-efficient flight path based on environmental changes. Simultaneously, the control module of this invention can adjust the flight trajectory in real time based on the planning results, responding quickly to sudden obstacles or external disturbances to ensure stable and reliable flight. Furthermore, this invention designs a multi-UAV cooperation mechanism, enabling multiple UAVs to share flight information and environmental models, automatically coordinating flight paths during joint missions to avoid collisions and mission overlap. Through this cooperative approach, the system's overall perception range is wider, and mission efficiency is higher.

[0045] In summary, this invention has achieved comprehensive improvements in flight navigation dimension expansion, path optimization, and collaborative capabilities, and can significantly enhance the autonomous navigation capability, safety, and energy efficiency of UAVs in complex environments, thus possessing strong practical value and promotional significance. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the accompanying drawings used in the following description of the embodiments or examples will be briefly introduced. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the technical solutions shown in these drawings without creative effort.

[0047] Figure 1 This is a module architecture diagram of a multi-dimensional flight navigation system for unmanned aerial vehicles (UAVs).

[0048] Figure 2 This is a flowchart of a multi-dimensional flight navigation method for unmanned aerial vehicles (UAVs). Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0051] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0052] This embodiment verifies the multi-dimensional flight navigation system for unmanned aerial vehicles (UAVs) of the present invention within a warehouse space. The warehouse has internal dimensions of 80m × 40m × 8m, with multiple rows of fixed shelves, each 6m high, and aisles 3m wide between the shelves. The UAV needs to fly from point A to point B, avoiding obstacles such as shelves, chandeliers, and mobile forklifts during flight, and hovering to take pictures at three pre-set checkpoints. Dynamic interference exists in the environment, including forklift movement and personnel movement.

[0053] The system consists of a multi-sensor module, an environmental perception and modeling module, a localization module, a path planning module, a control execution module, and a collaboration module. This embodiment is described using a standalone mode. The multi-sensor module includes an inertial measurement unit, a lidar, a vision camera, a broadband positioning sensor, a barometer, and a magnetometer. Each sensor is triggered uniformly by a time synchronization module, and the sampling frequency is set to 200Hz. The control and computing unit uses an onboard embedded processor to run fusion, modeling, planning, and control algorithms.

[0054] After the system is powered on, data acquisition begins. The inertial measurement unit outputs linear acceleration and angular velocity; the lidar outputs point cloud data; the visual camera outputs synchronized RGB and depth images; the ultra-wideband module performs distance measurements with four base stations; the barometer outputs relative altitude; and the magnetometer outputs magnetic field vector. Data from each sensor is then timestamped and fed into the subsequent processing chain.

[0055] The environmental perception and modeling module employs a vision-geometric fusion framework. The vision branch extracts semantic features from images and outputs pixel-level semantic labels and confidence scores for identifying the ground, shelves, walls, and dynamic targets. The geometry branch performs ground separation, outlier removal, and voxel downsampling on the point cloud, and identifies the shapes of the ground, walls, and shelves through plane fitting. Subsequently, the visual semantic results are projected onto the LiDAR coordinate system, and semantic occupancy probability fusion is performed on the octet voxel map. The fusion rule adopts a weighted model of occupancy probability and semantic confidence score, with the fusion weight α set to 0.6 in this embodiment. Finally, a 3D environment model with semantic labels and their confidence scores is obtained, which can provide the spatial layout of the shelves. In this embodiment, the current position and velocity estimates of three rows of parallel shelves, passable aisles, and dynamic obstacles are identified.

[0056] The localization module performs nonlinear fusion estimation of the UAV pose on the SE(3) Lie group manifold. The system uses the acceleration and angular velocity of the inertial measurement unit for short-time pose and velocity prediction, and updates the position and relative pose observations provided by ultra-wideband ranging and visual / LiDAR odometry, respectively. The estimation is performed in the group space to ensure the orthogonality of the rotation matrix and improve stability under drastic attitude changes. The state vector is obtained by combining the barometer altitude and magnetometer heading assistance.

[0057]

[0058] Position, velocity, and attitude are represented in world coordinates, E is the normalized value of the electrical charge, R is the safety margin estimate of the nearest obstacle, and W is the normalized estimate of the intensity of environmental disturbances. This state vector, along with its estimated uncertainty, is then fed into the subsequent planning and control modules.

[0059] The path planning module constructs probability distributions related to each optimization objective based on the semantic confidence output from the 3D environment model and the estimation uncertainty from the localization module, calculates the corresponding information entropy values, and obtains the dynamic weight coefficients of each objective according to the entropy-driven soft normalization rule. Taking a warehouse scenario as an example, under the condition that the task has no strict time limit and attitude stability is more critical, the result reflects the system's tendency to prioritize attitude stability and energy efficiency over the shortest time in the current environment.

[0060] Under the aforementioned weights, the system constructs a multi-objective cost function:

[0061]

[0062] In this embodiment, the safe distance threshold is set to 1m. Under the joint constraints of the 3D environment model and the cost function, the path planning module uses an improved RRT algorithm to generate a 3D trajectory from the starting point A to the ending point B, and uses the aforementioned cost function as the path segment evaluation criterion during the expansion and reconnection phases. The algorithm increases the sampling density near the predicted trajectory of dynamic obstacles to enhance obstacle avoidance feasibility. When the environment model is updated, such as when changes in forklift pose cause the feasible passage to shrink, or when the W index increases, the module triggers local replanning to maintain the safety and cost optimization of the trajectory. The trajectory generated in this embodiment contains approximately 45 waypoints, a path length of approximately 78.3m, an estimated flight time of approximately 26.1s, an average speed of approximately 3.0m / s, an estimated power consumption of approximately 4.2%, and a minimum safe distance of approximately 1.8m. At the three checkpoints, the system automatically performs short-term hovering and completes the photo-taking task.

[0063] The control execution module employs model predictive control to solve for control variables within a finite time domain, combining aircraft dynamics and obstacle avoidance constraints to achieve trajectory tracking and disturbance rejection control. For near-field small-range corrections or low-speed passage, the system can switch to vector field control to reduce computational burden. The controller periodically performs rolling optimization; when a new obstacle is detected or environmental disturbances increase, it automatically strengthens the weighted response of safety and attitude stability terms, outputting corresponding thrust and three-axis angular velocity commands to ensure smooth passage and safe obstacle avoidance in storage aisles and densely packed racking areas.

[0064] In some embodiments, the localization module employs a deterministic observer based on the SE(3) group to perform state estimation on a Lie group manifold. The SE(3) group is a three-dimensional rigid body pose group, whose elements are used to simultaneously describe the UAV's position and planar attitude in three-dimensional space. An element of this group can be represented as a homogeneous transformation matrix.

[0065]

[0066] Where R represents the attitude rotation matrix of the UAV, belonging to the rotation group SO(3); p = [x, t, z] T This represents the position vector of the UAV in the coordinate system.

[0067] The deterministic observer is a nonlinear state estimation algorithm defined on a Lie group manifold. It estimates the attitude and position of the UAV in real time by fusing multi-source sensor data. The dynamic equation of the observer can be expressed as follows:

[0068]

[0069] in, For the current pose estimation, Let K(e) be the estimated velocity vector composed of angular velocity and linear acceleration obtained from the inertial measurement unit (IMU), and let K(e) be the error feedback term used to correct the estimation error in the Lie group space.

[0070] During implementation, the positioning module first uses an inertial measurement unit to collect the angular velocity and acceleration signals of the UAV, and obtains the initial pose prediction result through integration. Subsequently, based on the distance or relative position information obtained by the ultra-wideband positioning sensor, and the relative pose change data provided by the visual camera or lidar, the prediction result is corrected. The deterministic observer performs pose calculations directly on the Lie group SE(3), thereby maintaining the orthogonality constraint of the rotation matrix and avoiding anomalies in the Euler angle representation.

[0071] Through the aforementioned fusion mechanism, the positioning module can achieve high-precision state estimation of the UAV without relying on linearization assumptions and statistical noise models. This algorithm guarantees that the estimation error converges asymptotically in the sense of a Lie group, maintaining the stability and consistency of attitude estimation even in the presence of sensor noise or dynamic disturbances. Compared to traditional Kalman filter-based positioning methods, the deterministic observer based on the SE(3) group is mathematically more consistent with the geometric characteristics of rigid body motion of the UAV, effectively improving positioning robustness and convergence accuracy, and is suitable for flight missions in environments with limited GPS signals or complex conditions.

Claims

1. A multi-dimensional flight navigation system for unmanned aerial vehicles (UAVs), characterized in that, include: A multi-sensor module is used to collect flight status data and environmental data of the UAV. The multi-sensor module includes an inertial measurement unit, a lidar sensor, a vision camera, an ultra-wideband positioning sensor, a barometer, and a magnetometer. An environmental perception and modeling module, connected to the multi-sensor module, is used to construct a 3D environment model with semantic information and its confidence level based on a vision-geometry fusion network. The vision-geometry fusion network includes a visual feature extraction branch and a geometric structure reconstruction branch. The visual feature extraction branch extracts scene semantic features based on image data collected by the vision camera and outputs semantic confidence level. The geometric structure reconstruction branch reconstructs the scene geometric structure based on point cloud data collected by the lidar sensor. The outputs of the two branches are processed by a feature fusion layer to generate the 3D environment model. The 3D environment model includes the geometric shape, spatial location, semantic category, and probability distribution parameters of obstacles. The positioning module, connected to the multi-sensor module, is used to determine the real-time position and attitude of the UAV by fusing data from the inertial measurement unit, the ultra-wideband positioning sensor, the visual camera, and the lidar sensor in a GPS-restricted environment. The positioning module uses a deterministic observer based on the SE(3) group to perform state estimation on the Lie group manifold. The deterministic observer based on the SE(3) group outputs the position vector, attitude rotation matrix, and state estimation uncertainty of the UAV in the world coordinate system by fusing the acceleration and angular velocity data provided by the inertial measurement unit, the distance measurement data provided by the ultra-wideband positioning sensor, and the relative pose data provided by the visual camera or the lidar sensor. The path planning module, connected to the environment perception and modeling module and the positioning module, is used to generate the optimal flight path in the three-dimensional environment model based on a multi-objective optimization navigation model in a multi-dimensional state space. The multi-objective optimization navigation model in the multi-dimensional state space uses state vectors. Describe the multidimensional state of the drone; Where x, y, and z are the three-dimensional spatial coordinates, and v_x, v_y, and v_z are the three-dimensional velocity components. θ and ψ are attitude angles, E is residual energy, R is safety margin, and W is environmental disturbance parameter; the path planning module determines the control vector by solving for the minimum value of a multi-objective cost function with an adaptive weighting mechanism based on information entropy. Where T is the thrust. ω_θ and ω_ψ are the roll rate control, pitch rate control, and yaw rate control, respectively; the information entropy adaptive weighting mechanism dynamically adjusts the weight coefficients of each cost term in the multi-objective cost function based on the semantic confidence and the state estimation uncertainty; A control execution module, connected to the path planning module, is used to realize trajectory tracking and dynamic obstacle avoidance based on the control vector; The collaboration module, connected to the path planning module and the environment perception and modeling module, is used to realize information sharing, collaborative environmental modeling, and cluster path coordination in a multi-UAV system.

2. The UAV multi-dimensional flight navigation system according to claim 1, characterized in that, The multi-objective cost function is expressed as: wherein D is a path length cost term, E C is an energy consumption cost term, S is a pose stability cost term, T f is a flight time cost term, R -1 is a reciprocal safety margin cost term, w1, w2, w3, w4, w5 are corresponding weight coefficients, and T is the total flight time.

3. The UAV multi-dimensional flight navigation system according to claim 2, characterized in that, The mathematical expression for the information entropy adaptive weighting mechanism is: w i (t+1) = exp(-H i (t)) / ∑ j exp(-H j (t)) where H i (t) is the information entropy of the i-th cost item at time t, and the calculation formula is: H i (t) = -∑ k p ik (t) log p ik (t) wherein p ik (t) is a probability distribution value of the kth state of the ith cost term at the tth time instant, which is obtained according to the comprehensive calculation of the semantic confidence and the state estimation uncertainty.

4. The UAV multi-dimensional flight navigation system according to claim 2, characterized in that: The path length cost term D is calculated by accumulating the Euclidean distances between each waypoint on the planned path. the energy consumption cost term E C calculated from the product of the thrust T and the time of flight; The attitude stability cost term S is based on the attitude angle. The deviations of θ and ψ from the desired attitude and the angular velocity Calculation of the combined measurement of ω_θ and ω_ψ; the time-of-flight cost term T f is the estimated time of arrival to the target location from the current time. The safety margin reciprocal cost term R -1 According to the reciprocal of the distance between the UAV and the nearest obstacle.

5. The UAV multi-dimensional flight navigation system according to claim 1, characterized in that, The path planning module uses a multi-objective reinforcement learning algorithm or an improved RRT algorithm to generate the optimal flight path. The multi-objective reinforcement learning algorithm trains the agent to learn a strategy that balances path length, energy consumption, attitude stability, flight time, and safety margin in the multi-dimensional state space. The improved RRT algorithm introduces the multi-objective cost function as a path evaluation criterion during the random tree expansion process, selects the node that minimizes the cost function value for tree expansion, and realizes path optimization in three-dimensional space.

6. The UAV multi-dimensional flight navigation system according to claim 1, characterized in that, The control execution module uses model predictive control to achieve trajectory tracking and dynamic obstacle avoidance; the model predictive control method uses the multi-objective cost function as the optimization objective in each control cycle and outputs control commands that satisfy the dynamic constraints and environmental safety constraints of the UAV.

7. The UAV multi-dimensional flight navigation system according to claim 1, characterized in that, The collaboration module shares the state vector, weight coefficients, local 3D environment models, and obstacle location information among multiple UAVs via a wireless communication network. Each UAV merges multiple local 3D environment models into a global environment model based on the received information and updates its own weight coefficients and multi-objective cost function parameters. During path planning, it coordinates the flight paths of each UAV to avoid mid-air collisions and mission conflicts.

8. The UAV multi-dimensional flight navigation system according to claim 1, characterized in that, The system supports dynamic switching between indoor and outdoor environments. In the indoor environment, the positioning module mainly relies on the ultra-wideband positioning sensor, the visual camera, and the lidar sensor for positioning. In the outdoor environment, the positioning module further integrates GPS signals and barometer data. The path planning module adjusts the initial value of the weight coefficient and the update frequency of the information entropy adaptive weight mechanism according to the environment type identifier.

9. The UAV multi-dimensional flight navigation system according to claim 1, characterized in that, The path planning module monitors the UAV's flight status and environmental changes in real time. When any of the following conditions are met, the path replanning process is triggered: a new obstacle is detected within the safe range of the currently planned path, the environmental disturbance parameter W exceeds a preset threshold, the safety margin R is lower than a preset minimum value, or the mission target position changes. After triggering replanning, the information entropy adaptive weighting mechanism is re-executed to update the weight coefficients, the minimum value of the multi-objective cost function is recalculated, and an updated flight path is generated.

10. A UAV flight navigation method based on a multi-objective optimization navigation model in a multi-dimensional state space, characterized in that, Includes the following steps: Step S1: Collect flight status data and environmental data of the UAV through a multi-sensor module, which includes an inertial measurement unit, a lidar sensor, a vision camera, an ultra-wideband positioning sensor, a barometer, and a magnetometer. Step S2: Process the environmental data based on the visual-geometric fusion network to construct a three-dimensional environment model with semantic information and confidence. The visual feature extraction branch of the visual-geometric fusion network extracts scene semantic features and outputs semantic confidence, while the geometric structure reconstruction branch reconstructs the scene geometric structure. The outputs of the two branches are processed by the feature fusion layer to generate the three-dimensional environment model. Step S3: Using a deterministic observer based on the SE(3) group to fuse the flight state data, perform state estimation on the Lie group manifold, and determine the current state vector of the UAV. The state estimation uncertainty is given by x, y, z, which are the three-dimensional spatial coordinates, and v_x, v_y, v_z, which are the three-dimensional velocity components. θ and ψ are attitude angles, E is remaining energy, R is safety margin, and W is environmental disturbance parameter; Step S4: calculating information entropy H of each cost term according to the semantic confidence and the state estimation uncertainty i (t) = -∑ k p ik (t) log p ik (t), and updating each weight coefficient in the multi-objective cost function through an information entropy adaptive weight mechanism, with an update formula being w i (t+1) = exp(-H i (t)) / ∑ j exp(-H j (t)); Step S5: Construct the multi-objective cost function Where D is the path length cost term, E C Let S be the energy consumption cost, and T be the attitude stability cost. f R is the flight time cost term. -1 The cost term is the inverse of the safety margin. Step S6: Using a multi-objective reinforcement learning algorithm or an improved RRT algorithm, solve for the optimal flight path that minimizes the multi-objective cost function in the three-dimensional environment model, and calculate the corresponding control vector. Where T is the thrust. ω_θ and ω_ψ are the roll rate control, pitch rate control, and yaw rate control, respectively. Step S7: Using the model predictive control method, flight control commands are generated based on the control vector to drive the UAV to achieve trajectory tracking and dynamic obstacle avoidance; Step S8: Monitor the drone's status and environmental changes in real time. When a new obstacle is detected within the safe range, the environmental disturbance parameter exceeds the threshold, the safety margin is lower than the minimum value, or the mission objective changes, return to step S2 to perform path replanning. Otherwise, return to step S1 to continue execution until the target location is reached.

Images