A path planning method for library unmanned aerial vehicle inventory

Abstract

The application discloses a kind of library unmanned plane inventory path planning method, it is related to unmanned plane path planning technical field, the application solves the indoor GPS failure of library and causes positioning drift, complex environment is not enough to sense, path planning does not consider priority and safety efficiency, poor unmanned plane attitude stability, breakpoint reflight precision low technical problem.The application adopts multi-modal dynamic environment modeling, dynamic priority multi-objective path planning, scene intelligent obstacle avoidance, anti-disturbance adaptive attitude control, breakpoint five core methods of five-step method, fusion multi-source sensor data, in combination with quantitative fusion formula, genetic algorithm, PID+PPO hybrid control, A* algorithm etc.Each link optimization is realized, and corresponding planning model is also constructed.The application realizes the precise positioning of library unmanned plane inventory, full-scene safe obstacle avoidance, anti-disturbance stable flight and breakpoint accurate reflight, greatly improves inventory efficiency and task continuity, reduces operation interference, adapts to the special operation scene of library indoor.

Description

Technical Field

[0001] This invention relates to the field of drone path planning technology, specifically a path planning method for drone inventory in a library. Background Technology

[0002] In the daily management of libraries, book inventory is an important but tedious task. Traditional manual inventory methods are inefficient, time-consuming, labor-intensive, and prone to omissions and errors. With the development of drone technology, drone inventory is gradually becoming an important alternative to manual labor. However, the unique indoor environment of libraries presents many challenges to drone path planning:

[0003] 1. Indoor GPS signals in the library are ineffective, the anti-interference ability of single vision or LiDAR is poor, the combined positioning efficiency is low, positioning drift is prone to occur in low texture environment, and there is a lack of quantitative fusion algorithms to ensure positioning accuracy.

[0004] 2. There are static obstacles such as bookshelves, tables and chairs, and glass doors in the room, as well as dynamic pedestrian flow. It is difficult to identify transparent obstacles. Complex scenes such as changes in lighting and dark environments further increase the difficulty of perception. Existing technologies lack multi-source data quantification and fusion solutions.

[0005] 3. Traditional path planning often pursues the geometric shortest path, without fully considering factors such as mission urgency and regional importance, and it is difficult to balance flight safety and efficiency through quantitative formulas;

[0006] 4. The indoor space is small and the bookshelves are dense. The flight of drones is easily affected by airflow disturbances, resulting in insufficient attitude stability. Traditional PID control parameters are fixed and lack a dynamic adjustment quantification model, making it difficult to adapt to the low-interference and high-safety operation requirements of libraries.

[0007] 5. During drone operations, operations may be interrupted due to reasons such as low battery, obstacle avoidance hovering, or mission pause. Traditional breakpoint resume technology has insufficient positioning accuracy in indoor environments without GPS, and lacks quantitative support for anchor point calibration and path replanning, which can easily lead to problems such as deviation of the re-flight path, mission duplication, or omission.

[0008] To address the aforementioned issues, there is an urgent need for a drone inventory path planning method that incorporates multiple sets of quantitative formulas and is adapted to the indoor environment of libraries, in order to achieve accurate, efficient, safe, and continuous inventory operations. Summary of the Invention

[0009] The purpose of this application is to address the following technical issues in existing technologies: GPS failure in library rooms leads to positioning drift; lack of quantitative fusion algorithms; insufficient perception of obstacles and dynamic pedestrian flow in complex environments, and lack of multi-source data quantization processing schemes; path planning does not take into account both task priority and safety efficiency through formulas; poor attitude stability and lack of quantitative models for dynamic parameter adjustment; and low accuracy of go-around after operation interruption, lacking quantitative support for anchor point calibration and path replanning.

[0010] To achieve the above objectives, this application adopts the following technical solution:

[0011] A path planning method for library drone inventory, characterized by comprising the following steps:

[0012] S1: Multimodal dynamic environment modeling, based on library environment data obtained from multi-source sensors carried by UAVs, and a spatiotemporal network model for implementation and updating is constructed through feature point extraction and data fusion formulas;

[0013] S2: Dynamic Priority Multi-Objective Path Planning: Based on the spatiotemporal network model of S1, the surveying sub-regions are divided. The urgency of the public-private partnership task, regional importance and real-time environmental status are calculated through dynamic priority. Combined with optimization algorithms, the optimal self-access sequence and local routes are generated.

[0014] S3: Scenario-based intelligent obstacle avoidance: The detection data is processed through a cross-validation formula based on multi-sensor data, and the obstacle avoidance strategy is dynamically adjusted based on hierarchical decision logic and distance judgment formula to achieve safe flight in three-dimensional space without blind spots;

[0015] S4: Disturbance-resistant adaptive attitude control: Adopting a hybrid control architecture, the control parameters are dynamically optimized by combining the PID parameter dynamic adjustment formula and the reinforcement learning reward calculation formula with multi-source sensor data to ensure the stability of the UAV attitude;

[0016] S5: Resume Flight After Disconnection: When the drone operation is interrupted, the position and attitude information are recorded by the four-dimensional spatiotemporal anchor point calibration formula. Combined with the task status snapshot storage logic, when the operation is resumed, the flight path is accurately backtracked and replanned based on the environmental update data and dynamic priority replanning formula.

[0017] Preferably, the specific steps of S1 are as follows:

[0018] First, a basic map is constructed: environmental feature points are extracted using ORB-SLAM3 and fused with UWB centimeter-level positioning data to generate an initial dense 3D point cloud map. The positioning fusion formula is as follows:

[0019]

[0020] in, The merged positioning coordinates , These are the weights for UWB positioning and ORB-SLAM3 positioning, respectively. , To locate the original coordinates for UWB, Locate the original coordinates for ORB-SLAM3;

[0021] Secondly, static obstacles are labeled: static obstacles are identified using image segmentation algorithms, and their size and position are calculated using geometric feature extraction formulas.

[0022]

[0023] in, , , These are the length, width, and height of the obstacle, respectively. The coordinates of the obstacle boundary. The coordinates of the obstacle's center;

[0024] Then, the dynamic data is fused: the area of ​​human activity is captured by an infrared thermal imaging sensor, and the data of the dark environment is supplemented by a depth camera. The fusion formula is as follows:

[0025]

[0026] in, For dynamic obstacle fusion data, The fusion weight is (0.3-0.7). For infrared thermal imaging data, For depth camera data;

[0027] Last update of the spatiotemporal network: The embedded processor fuses static map and dynamic obstacle layers in real time to construct a spatiotemporal mesh model, and the update frequency formula is:

[0028]

[0029] in, For the update cycle, For update frequency.

[0030] Preferably, the specific steps of S2 are as follows:

[0031] S2-1: Pre-modeling and sub-region division: Based on the spatiotemporal grid model generated in S1, 5m×5m surveying sub-regions are divided according to the library bookshelf arrangement and regional functions. The formula for the area of ​​each sub-region is:

[0032]

[0033] in, , These are the length and width of the sub-region, respectively. , All are set to 5m;

[0034] S2-2: Dynamic Priority Calculation: Input task urgency (U) and region importance (I) parameters, and calculate the dynamic priority of sub-regions using a weighted summation formula:

[0035]

[0036] in, =0.6 (weight for high task urgency). The weighting is based on the importance of the area (0.4 for the core collection area and 0.2 for the general reading area).

[0037] S2-3: Subregion Visit Sequence Optimization: Using a genetic algorithm, with a population size of 50 and 100 iterations, the goal is to minimize visit time. The fitness function formula is:

[0038]

[0039] in, For sub-region To sub-region distance, Select variables for the path (1 indicates that the path passes through, 0 indicates that the path does not pass through). For sub-region The time for inventory checks;

[0040] S2-4: Local Flight Path Design: Generate local flight paths within a sub-region using a "Z" shaped scanning mode. The recursive formula for flight path coordinates is:

[0041]

[0042] in, The scanning step size is 0.5m. To maintain a fixed flight altitude;

[0043] S2-5: Safety and Smoothness Optimization: A 3D safety corridor is constructed based on static obstacle data, and the path is smoothed using a B-spline curve algorithm. The curve formula is:

[0044]

[0045] in, For the coordinates of the control points, Let k be the basis functions of the B-spline. ∈[0,1]; Set the maximum flight speed =2m / s, the velocity constraint formula is:

[0046] .

[0047] Preferably, step S3 is as follows:

[0048] S3-1 Data Processing: Interference suppression is performed on the sensor data using a Kalman filter algorithm. The filtering formula is as follows:

[0049]

[0050] in, This is the filtered state estimate. Here is the state transition matrix. To control the input matrix, To control the input, Let covariance matrix be the variance matrix. For process noise covariance, For Kalman gain, For the observation matrix, For the observed values, To observe the noise covariance,

[0051] The identity matrix is ​​used; cross-validation is performed using ultrasonic and infrared sensors, and the validation formula is as follows:

[0052]

[0053] in, To verify the distance to the obstacle, This refers to the ultrasonic detection distance. For infrared detection distance, The allowable error is 0.2m.

[0054] S3-2 Hierarchical Decision Making: Dividing decision-making levels using a distance-based formula:

[0055] Long distance: Cruise mode;

[0056] Mid-range: In fuzzy steering mode, the steering angle formula is:

[0057]

[0058] in, The distance coefficient is 6° / m. =5m;

[0059] Up close: Emergency hovering triggers the A* algorithm to replan the local detour path. The path cost function formula is:

[0060]

[0061] in, From the starting point to the node The actual cost, For nodes The estimated cost to reach the destination;

[0062] S3-3 Intelligent Environmental Linkage: Accesses the library management system via a Wi-Fi 6 Mesh network to receive real-time pedestrian flow data. The pedestrian density calculation formula is as follows:

[0063]

[0064] in, For human traffic density, For the number of people in the region, The area is the region; if If so, it is designated as a temporary obstacle avoidance zone.

[0065] Preferably, the specific steps of S4 are as follows:

[0066] S4-1 Hybrid Control Architecture: The basic control adopts a dual-closed-loop cascaded PID controller, and the outer loop angle control formula is:

[0067]

[0068] The inner loop angular rate control formula is:

[0069]

[0070] in, , These are the expected commands for angle and angular rate, respectively. , These are the actual angle and angular rate, respectively. , , These are the proportional, integral, and differential coefficients, respectively; an attention-based PPO reinforcement learning agent is introduced, with the reward function formula as follows:

[0071]

[0072] in, As a reward for stable posture, For speed tracking rewards, Penalty for collision;

[0073] S4-2 Model Deployment: The trained reinforcement learning model is embedded into the STM32F407 as a fixed-point library, working in conjunction with the PID controller. The parameter fusion formula is as follows:

[0074]

[0075] in, For the final control parameters, , These are the weights of the PID and PPO parameters, respectively.

[0076] S4-3 Multitasking Scheduling: The system is divided into three core tasks: data acquisition, algorithm calculation, and motor control, with an anti-disturbance response time. The response time constraint formula is:

[0077]

[0078] in, For data collection time, The algorithm's computation time, To control the execution time of instructions.

[0079] Preferably, the specific steps of S5 are as follows:

[0080] S5-1 Breakpoint Triggering and State Snapshot Storage: Breakpoint triggering conditions include both active and passive triggering. When triggered, the UAV performs the "hovering stabilization - state snapshot - anchor point calibration" action; the current centimeter-level coordinates are obtained through the UWB positioning module, and the surrounding static obstacle features are extracted as spatial anchor points in conjunction with the spatiotemporal grid model. The four-dimensional spatiotemporal anchor point calibration formula is:

[0081]

[0082] in,( () represents three-dimensional coordinates. For trigger time, , , These are pitch angle, roll angle, and yaw angle, respectively; in the operation progress snapshot, the formula for calculating sub-area scan coverage is:

[0083]

[0084] in, For scan coverage, The scanned area. This represents the total area of ​​the sub-region;

[0085] S5-2 Go-Ahead Preparation and Precise Backtracking: When the UAV restarts or resumes operations, after updating the environmental data, precise backtracking is achieved through a coordinate deviation correction formula.

[0086]

[0087]

[0088] in, For coordinate deviation, For anchor point coordinates, The current coordinates, For correction factor, To backtrack to the target coordinates; the backtracking error is required to be | |≤5cm;

[0089] S5-3 Dynamic Priority Replanning and Go-Around Execution: Based on the unfinished tasks, after recalculating the dynamic priority, the optimized formula for the go-around path length is as follows:

[0090]

[0091] in, This represents the total length of the go-around path. The distance between path nodes; the flight speed is adjusted based on the remaining battery power, and the adaptive speed formula is:

[0092]

[0093] in, Remaining battery power Fully charged For the speed of resuming flights;

[0094] S5-4 Seamless Operation Status Transition: After the drone reaches the starting position of the unfinished task, the operation parameters are restored, and the data merging formula is as follows:

[0095]

[0096] in, For complete operation data, For snapshot data, Added data for resumption of flights;

[0097] S5-5 Safety Redundancy Design: The formula for determining remaining power is:

[0098]

[0099] in, The amount of electricity needed to complete the remaining tasks. This represents the maximum flight distance on a full charge; if Then the return route will be automatically planned.

[0100] This application also provides a library inventory drone adapted to the path planning method described above, characterized in that: it includes a drone body, the drone body has a core control area inside, the core control area uses an STM32F4 MCU; the drone body has a Wi-Fi 6 Mesh antenna at the bottom, a lidar at the top, and an infrared thermal imaging sensor and a mono / dual-lens camera at the bottom; the drone body is equipped with an ORB-SLAM3 vision module and a UWB ultra-wideband positioning module.

[0101] Preferably, the four extended arms of the UAV are each equipped with a four-way obstacle avoidance module, which is a four-channel ultrasonic sensor; the core control area is also equipped with a UWB ultra-wideband positioning module, a low-noise motor and a flexible circuit board, and the operating noise of the UAV is controlled within 50dB.

[0102] Preferably, the UAV body is equipped with an attitude control module. The core control of the attitude control module uses an STM32F407 microcontroller and is equipped with the FreeRTOS operating system. The sensor combination of the attitude control module is connected to the MPU-6050, HMC5883L, MS5611 and UWB ultra-wideband positioning module through I2C / SPI interfaces.

[0103] Preferably, the actuator of the attitude control module is a low-noise motor, which supports PWM precise speed regulation, and a depth camera is also installed inside the main body of the UAV.

[0104] Compared with the prior art, this application has at least the following beneficial effects:

[0105] 1. Multimodal dynamic environment modeling integrates multiple sensor data through multiple sets of quantization formulas to accurately calculate positioning coordinates, obstacle features and dynamic data, solving problems such as GPS failure in indoor libraries and positioning drift in low-texture environments, making the environment model update cycle stable at 50ms and the positioning error ≤5cm;

[0106] 2. Dynamic priority multi-objective path planning clarifies task priorities and path parameters through quantitative formulas. Combined with genetic algorithms and B-spline curve smoothing techniques, it reduces sub-region access time by more than 30%, ensures continuous and stable flight speed, and avoids efficiency losses caused by sharp turns.

[0107] 3. Scenario-based intelligent obstacle avoidance improves the anti-interference capability of sensor data by 40% through quantitative processing formulas such as Kalman filtering and cross-validation, reduces the false alarm rate of transparent obstacles to below 5%, and improves the efficiency of detouring in narrow spaces by 25% by combining hierarchical decision-making logic with distance formulas.

[0108] 4. The anti-disturbance adaptive attitude control uses a dynamic adjustment formula for PID parameters and a reinforcement learning quantization model to ensure that the attitude error of the UAV under airflow disturbance of 0.5-1m / s is ≤0.5°, the anti-disturbance response time is <100ms, and the path execution accuracy is significantly improved.

[0109] 5. The breakpoint resume flight technology uses quantitative formulas such as anchor point calibration and deviation correction to achieve a go-around position error of ≤5cm. Combined with quantitative calculation of operation progress and path replanning formula, it avoids operation repetition and omission, improving mission continuity by more than 50%. Attached Figure Description

[0110] Figure 1 This is a structural modification diagram of the library drone of this invention;

[0111] Figure 2 This is an overall flowchart of the path planning method for library drone inventory in this invention;

[0112] Figure 3 This is a schematic diagram of a drone inventory operation in a library setting according to the present invention. Detailed Implementation

[0113] Please see Figure 2 This application provides a path planning method for library drone inventory, comprising the following steps:

[0114] S1: Multimodal dynamic environment modeling

[0115] In one implementation, library environment data is obtained based on multi-source sensors carried by a drone, and a spatiotemporal network model for implementation and updating is constructed through feature point extraction and data fusion formulas.

[0116] Specifically, S1 includes the following steps:

[0117] First, a basic map is constructed: environmental feature points are extracted using ORB-SLAM3, and UWB centimeter-level positioning data is fused to generate an initial dense 3D point cloud map. The positioning fusion formula is as follows:

[0118]

[0119] in, The merged positioning coordinates , The weights for UWB positioning and ORB-SLAM3 positioning are respectively (the sum is 1). To locate the original coordinates for UWB, Locate the original coordinates for ORB-SLAM3;

[0120] In one embodiment, a configuration is provided. .

[0121] Secondly, static obstacles are labeled: static obstacles are identified using image segmentation algorithms, and their size and position are calculated using geometric feature extraction formulas.

[0122]

[0123] in, , , These are the length, width, and height of the obstacle, respectively. Equal to the boundary coordinates of the obstacle. The coordinates of the obstacle's center;

[0124] Then, the dynamic data is fused: the area of ​​human activity is captured by an infrared thermal imaging sensor, and the data of the dark environment is supplemented by a depth camera. The fusion formula is as follows:

[0125]

[0126] in, For dynamic obstacle fusion data, The fusion weight is (0.3-0.7). For infrared thermal imaging data, For depth camera data; preferably, the .

[0127] Last update of the spatiotemporal network: The embedded processor fuses static map and dynamic obstacle layers in real time to construct a spatiotemporal mesh model, and the update frequency formula is:

[0128]

[0129] in, The update cycle is set to 50ms. The update frequency is 20Hz.

[0130] S2: Dynamic Priority Multi-Objective Path Planning

[0131] In one implementation, the mapping sub-region is divided based on the S1 spatiotemporal network model. The urgency of the public fusion task, regional importance and real-time environmental status are calculated through dynamic priority calculation. The optimal self-access sequence and local routes are generated by combining optimization algorithms.

[0132] Specifically, in one embodiment, the specific steps of S2 are as follows:

[0133] S2-1: Pre-modeling and sub-region division: Based on the spatiotemporal grid model generated in S1, 5m×5m surveying sub-regions are divided according to the library bookshelf arrangement and regional functions. The formula for the area of ​​each sub-region is:

[0134]

[0135] in, , These are the length and width of the sub-region (both set to 5m);

[0136] S2-2: Dynamic Priority Calculation: Input task urgency (U) and region importance (I) parameters, and calculate the dynamic priority of sub-regions using a weighted summation formula:

[0137]

[0138] in, =0.6 (weight for high task urgency). The weighting is based on the importance of the area (0.4 for the core collection area and 0.2 for the general reading area).

[0139] In one implementation, a task urgency level is set as "the inventory must be completed within one day". =1.0, Regional Importance of the Core Collection Area =0.4, I=0.2 for general reading area.

[0140] S2-3: Subregion Visit Sequence Optimization: Using a genetic algorithm, with a population size of 50 and 100 iterations, the goal is to minimize visit time. The fitness function formula is:

[0141]

[0142] in, For sub-region To sub-region distance, Select variables for the path (1 indicates that the path passes through, 0 indicates that the path does not pass through). For sub-region The time for inventory checks;

[0143] S2-4: Local route design: Generate local routes within sub-regions using a "Z" shaped scanning mode. The route coordinate recursion formula is as follows:

[0144]

[0145] in, The scanning step size is 0.5m. To maintain a fixed flight altitude;

[0146] S2-5: Safety and Smoothness Optimization: A 3D safety corridor is constructed based on static obstacle data, and the path is smoothed using a B-spline curve algorithm. The curve formula is:

[0147]

[0148] in, For the coordinates of the control points, For k-th order B-spline basis functions, ∈[0,1]; Set the maximum flight speed =2m / s, the velocity constraint formula is:

[0149]

[0150] Preferably, a three-dimensional safety corridor is constructed within a 1m radius around the planned path, and a cubic B-spline curve is used to smooth the path. The flight speed is ensured to be ≤2m / s through a speed constraint formula.

[0151] S3: Scenario-based intelligent obstacle avoidance

[0152] In one embodiment, the detection data is processed by a multi-sensor data cross-validation formula, and the obstacle avoidance strategy is dynamically adjusted based on hierarchical decision logic and distance determination formula to achieve safe flight in three-dimensional space without blind spots.

[0153] Specifically, in one embodiment, the above-mentioned S3 step is as follows:

[0154] S3-1 Data Processing: Interference suppression is performed on the sensor data using a Kalman filter algorithm. The filtering formula is as follows:

[0155]

[0156] in, This is the filtered state estimate. Here is the state transition matrix. To control the input matrix, To control the input, Let covariance matrix be the variance matrix. For process noise covariance, For Kalman gain, For the observation matrix, For the observed values, To observe the noise covariance,

[0157] The identity matrix is ​​used; cross-validation is performed using ultrasonic and infrared sensors, and the validation formula is as follows:

[0158]

[0159] in, To verify the distance to the obstacle, This refers to the ultrasonic detection distance. For infrared detection distance, The allowable error is 0.2m.

[0160] Preferably, Q=1e−6 and R=1e−3 are set; a cross-validation formula is used, with δ=0.2m set to eliminate false alarm data for transparent obstacles;

[0161] S3-2 Hierarchical Decision Making: Dividing decision-making levels using a distance-based formula:

[0162] Long distance: Cruise mode;

[0163] Mid-range: In fuzzy steering mode, the steering angle formula is:

[0164]

[0165] in, Distance factor (6° / m) =5m;

[0166] Close up: Emergency hovering triggers the A* algorithm to replan the local detour path. The path cost function formula is:

[0167]

[0168] in, From the starting point to the node The actual cost, For nodes The estimated cost to reach the destination;

[0169] S3-3 Intelligent Environmental Linkage: Accesses the library management system via a Wi-Fi 6 Mesh network to receive real-time pedestrian flow data. The pedestrian density calculation formula is as follows:

[0170]

[0171] in, For human traffic density, For the number of people in the region, The area is the region; if If so, it is designated as a temporary obstacle avoidance zone.

[0172] S4: Disturbance-resistant adaptive attitude control

[0173] In one implementation, a hybrid control architecture is adopted, which uses the PID parameter dynamic adjustment formula and the reinforcement learning reward calculation formula to dynamically optimize the control parameters by combining multi-source sensor data to ensure the attitude stability of the UAV.

[0174] S4-1 Hybrid Control Architecture: The basic control adopts a dual-closed-loop cascaded PID controller, and the outer loop angle control formula is:

[0175]

[0176] The inner loop angular rate control formula is:

[0177]

[0178] in, , These are the expected commands for angle and angular rate, respectively. , These are the actual angle and angular rate, respectively. , , These are the proportional, integral, and differential coefficients, respectively; an attention-based PPO reinforcement learning agent is introduced, with the reward function formula as follows:

[0179]

[0180] in, As a reward for stable posture, For speed tracking rewards, Penalty for collision;

[0181] In one embodiment, an outer ring angle ring is provided. Inner ring angular rate ring setting ; Build a library airflow disturbance simulation environment (wind speed 0.5-1m / s), and set the reinforcement learning reward function. ;

[0182] S4-2 Model Deployment: The trained reinforcement learning model is embedded into the STM32F407 as a fixed-point library, working in conjunction with the PID controller. The parameter fusion formula is as follows:

[0183]

[0184] in, For the final control parameters, , These are the weights of the PID and PPO parameters, respectively.

[0185] In one embodiment, a configuration is provided. .

[0186] S4-3 Multitasking Scheduling: The system is divided into three core tasks: data acquisition, algorithm calculation, and motor control, with an anti-disturbance response time. The response time constraint formula is:

[0187]

[0188] in, For data collection time, The algorithm's computation time, To control the execution time of instructions.

[0189] S5: Resume Flight After Disconnection:

[0190] In one implementation, when the drone operation is interrupted, the position and attitude information are recorded by the four-dimensional spatiotemporal anchor point calibration formula. Combined with the task status snapshot storage logic, when the operation is resumed, the re-planning path is accurately traced back and re-planned based on the environmental update data and the dynamic priority replanning formula.

[0191] The specific steps of S5 are as follows:

[0192] S5-1 Breakpoint Triggering and State Snapshot Storage: Breakpoint triggering conditions include both active and passive triggering. When triggered, the UAV performs the "hovering stabilization → state snapshot → anchor point calibration" action; the current centimeter-level coordinates are obtained through the UWB positioning module, and the surrounding static obstacle features are extracted as spatial anchor points through the spatiotemporal grid model. The four-dimensional spatiotemporal anchor point calibration formula is:

[0193]

[0194] in,( () represents three-dimensional coordinates. For trigger time, , , These are pitch angle, roll angle, and yaw angle, respectively; in the operation progress snapshot, the formula for calculating sub-area scan coverage is:

[0195]

[0196] in, For scan coverage, The scanned area. This represents the total area of ​​the sub-region;

[0197] S5-2 Go-Ahead Preparation and Precise Backtracking: When the UAV restarts or resumes operations, after updating the environmental data, precise backtracking is achieved through a coordinate deviation correction formula.

[0198]

[0199]

[0200] in, For coordinate deviation, For anchor point coordinates, The current coordinates, The correction factor is (0.8-1.2). To backtrack to the target coordinates; the backtracking error is required to be | |≤5cm;

[0201] S5-3 Dynamic Priority Replanning and Go-Around Execution: Based on the unfinished tasks, after recalculating the dynamic priority, the optimized formula for the go-around path length is as follows:

[0202]

[0203] in, This represents the total length of the go-around path. The distance between path nodes; the flight speed is adjusted based on the remaining battery power, and the adaptive speed formula is:

[0204]

[0205] in, Remaining battery power Fully charged The go-around speed is a minimum of 1.2 m / s.

[0206] S5-4 Seamless Operation Status Transition: After the drone reaches the starting position of the unfinished task, the operation parameters are restored, and the data merging formula is as follows:

[0207]

[0208] in, For complete operation data, For snapshot data, Added data for resumption of flights;

[0209] S5-5 Safety Redundancy Design: The formula for determining remaining power is:

[0210]

[0211] in, The amount of electricity needed to complete the remaining tasks. This represents the maximum flight distance on a full charge; if Then the return route will be automatically planned.

[0212] Based on the path planning method described above, this application also provides a drone for library inventory. Please refer to [link / reference]. Figure 1 The drone includes a main body 1, inside which is a core control area 2, which uses an STM32F4 MCU. To enable network connectivity, a Wi-Fi 6 Mesh antenna 8 is also provided below the main body 1.

[0213] The top of the drone body 1 is equipped with a lidar 10, and the bottom is equipped with an infrared thermal imaging sensor 4 and a monocular / binocular camera 6. The drone also carries an ORB-SLAM3 vision module and a UWB ultra-wideband positioning module, achieving full indoor coverage by deploying UWB base stations on the ground. Preferably, a SLAM binocular camera is used. In one embodiment, the drone uses an embedded processor as a processing platform to achieve environmental modeling through the above design.

[0214] The UAV is equipped with four-way obstacle avoidance modules 7 on its four extended arms. In one embodiment, each four-way obstacle avoidance module 7 consists of four ultrasonic sensors with a detection range of 0.1-10 meters, arranged in a three-dimensional layout. The four-way obstacle avoidance module 7, infrared thermal imaging sensor 4, and TOP laser sensor form a sensor system. Furthermore, the core control area 2 houses a UWB ultra-wide positioning module 3, a low-noise motor, and a flexible circuit board, keeping operating noise below 50dB and enabling scene-based intelligent obstacle avoidance.

[0215] In one embodiment, the UAV is also equipped with an attitude control module, the core of which is an STM32F407 microcontroller running the FreeRTOS operating system; the sensor assembly is connected to the MPU-6050, HMC5883L, MS5611 and UWB ultra-wide positioning module 3 via I2C / SPI interfaces; the actuator uses a low-noise motor that supports PWM precise speed regulation.

[0216] In one embodiment, an electronic speed controller 9 is provided inside the main body 1 of the drone in order to adjust the flight speed of the drone.

[0217] Figure 3 This is a schematic diagram of a drone inventory operation in a library setting, as described in this application. Based on the above description, the drone inventory path planning method for libraries provided by this community, through multiple sets of quantitative formulas supporting the collaborative work of various modules, achieves accurate positioning, full-scene perception, efficient path planning, safe obstacle avoidance, stable attitude control, and continuous operation for drone inventory in an indoor library environment. This significantly improves inventory efficiency and data integrity, reduces operational interference, and has good practical application results.

Claims

1. A path planning method for library drone inventory, characterized in that: Includes the following steps: S1: Multimodal dynamic environment modeling, based on library environment data obtained from multi-source sensors carried by UAVs, and a spatiotemporal network model for implementation and updating is constructed through feature point extraction and data fusion formulas; S2: Dynamic Priority Multi-Objective Path Planning: Based on the spatiotemporal network model of S1, the surveying sub-regions are divided. The urgency of the public-private partnership task, regional importance and real-time environmental status are calculated through dynamic priority. Combined with optimization algorithms, the optimal self-access sequence and local routes are generated. S3: Scenario-based intelligent obstacle avoidance: The detection data is processed through a cross-validation formula based on multi-sensor data, and the obstacle avoidance strategy is dynamically adjusted based on hierarchical decision logic and distance judgment formula to achieve safe flight in three-dimensional space without blind spots; S4: Disturbance-resistant adaptive attitude control: Adopting a hybrid control architecture, the control parameters are dynamically optimized by combining the PID parameter dynamic adjustment formula and the reinforcement learning reward calculation formula with multi-source sensor data to ensure the stability of the UAV attitude; S5: Resume Flight After Disconnection: When the drone operation is interrupted, the position and attitude information are recorded by the four-dimensional spatiotemporal anchor point calibration formula. Combined with the task status snapshot storage logic, when the operation is resumed, the flight path is accurately backtracked and replanned based on the environmental update data and dynamic priority replanning formula.

2. The path planning method for library drone inventory as described in claim 1, characterized in that: The specific steps of S1 are as follows: First, a basic map is constructed: environmental feature points are extracted using ORB-SLAM3, and UWB centimeter-level positioning data is fused to generate an initial dense 3D point cloud map. The positioning fusion formula is as follows: in, The merged positioning coordinates , These are the weights for UWB positioning and ORB-SLAM3 positioning, respectively. , To locate the original coordinates for UWB, Locate the original coordinates for ORB-SLAM3; Secondly, static obstacles are labeled: static obstacles are identified using image segmentation algorithms, and their size and position are calculated using geometric feature extraction formulas. in, , , These are the length, width, and height of the obstacle, respectively. The coordinates of the obstacle boundary. The coordinates of the obstacle's center; Then, the dynamic data is fused: the area of ​​human activity is captured by an infrared thermal imaging sensor, and the data of the dark environment is supplemented by a depth camera. The fusion formula is as follows: in, For dynamic obstacle fusion data, The fusion weight is (0.3-0.7). For infrared thermal imaging data, For depth camera data; Last update of the spatiotemporal network: The embedded processor fuses static map and dynamic obstacle layers in real time to construct a spatiotemporal mesh model, and the update frequency formula is: in, For the update cycle, For update frequency.

3. The path planning method for library drone inventory according to claim 1, characterized in that: The specific steps of S2 are as follows: S2-1: Pre-modeling and sub-region division: Based on the spatiotemporal grid model generated in S1, 5m×5m surveying sub-regions are divided according to the library bookshelf arrangement and regional functions. The formula for the area of ​​each sub-region is: in, , These are the length and width of the sub-region, respectively. , All are set to 5m; S2-2: Dynamic Priority Calculation: Input task urgency (U) and region importance (I) parameters, and calculate the dynamic priority of sub-regions using a weighted summation formula: in, =0.6 (weight for high task urgency). The weighting is based on the importance of the area (0.4 for the core collection area and 0.2 for the general reading area). S2-3: Subregion Visit Sequence Optimization: Using a genetic algorithm, with a population size of 50 and 100 iterations, the goal is to minimize visit time. The fitness function formula is: in, For sub-region To sub-region distance, Select variables for the path (1 indicates that the path passes through, 0 indicates that the path does not pass through). For sub-region The time for inventory checks; S2-4: Local Flight Path Design: Generate local flight paths within a sub-region using a "Z" shaped scanning mode. The recursive formula for flight path coordinates is: in, The scanning step size is 0.5m. To maintain a fixed flight altitude; S2-5: Safety and Smoothness Optimization: A 3D safety corridor is constructed based on static obstacle data, and the path is smoothed using a B-spline curve algorithm. The curve formula is: in, For the coordinates of the control points, For k-th order B-spline basis functions, ∈[0,1]; Set the maximum flight speed =2m / s, the velocity constraint formula is: 。 4. The path planning method for library drone inventory according to claim 1, characterized in that: The specific steps of S3 are as follows: S3-1 Data Processing: Interference suppression is performed on the sensor data using a Kalman filter algorithm. The filtering formula is as follows: in, This is the filtered state estimate. Here is the state transition matrix. To control the input matrix, To control the input, Let covariance matrix be the variance matrix. For process noise covariance, For Kalman gain, For the observation matrix, For the observed values, To observe the noise covariance, The identity matrix is ​​used; cross-validation is performed using ultrasonic and infrared sensors, and the validation formula is as follows: in, To verify the distance to the obstacle, This refers to the ultrasonic detection distance. For infrared detection distance, The allowable error is 0.2m. S3-2 Hierarchical Decision Making: Dividing decision-making levels using a distance-based formula: Long distance: Cruise mode; Mid-range: In fuzzy steering mode, the steering angle formula is: in, The distance coefficient is 6° / m. =5m; Close up: Emergency hovering triggers the A* algorithm to replan the local detour path. The path cost function formula is: in, From the starting point to the node The actual cost, For nodes The estimated cost to reach the destination; S3-3 Intelligent Environmental Linkage: Accesses the library management system via a Wi-Fi 6 Mesh network to receive real-time pedestrian flow data. The pedestrian density calculation formula is as follows: in, For human traffic density, For the number of people in the region, The area is the region; if If so, it is designated as a temporary obstacle avoidance zone.

5. The path planning method for library drone inventory according to claim 1, characterized in that: The specific steps of S4 are as follows: S4-1 Hybrid Control Architecture: The basic control adopts a dual-closed-loop cascaded PID controller, and the outer loop angle control formula is: The inner loop angular rate control formula is: in, , These are the expected commands for angle and angular rate, respectively. , These are the actual angle and angular rate, respectively. , , These are the proportional, integral, and differential coefficients, respectively; an attention-based PPO reinforcement learning agent is introduced, with the reward function formula as follows: in, As a reward for stable posture, For speed tracking rewards, Penalty for collision; S4-2 Model Deployment: The trained reinforcement learning model is embedded into the STM32F407 as a fixed-point library, working in conjunction with the PID controller. The parameter fusion formula is as follows: in, For the final control parameters, , These are the weights of the PID and PPO parameters, respectively. S4-3 Multitasking Scheduling: The system is divided into three core tasks: data acquisition, algorithm calculation, and motor control, with an anti-disturbance response time. The response time constraint formula is: in, For data collection time, The algorithm's computation time, To control the execution time of instructions.

6. The path planning method for library drone inventory according to claim 1, characterized in that: The specific steps of S5 are as follows: S5-1 Breakpoint Triggering and State Snapshot Storage: Breakpoint triggering conditions include both active and passive triggering. When triggered, the UAV performs the "hovering stabilization - state snapshot - anchor point calibration" action; the current centimeter-level coordinates are obtained through the UWB positioning module, and the surrounding static obstacle features are extracted as spatial anchor points through the spatiotemporal grid model. The four-dimensional spatiotemporal anchor point calibration formula is: in,( () represents three-dimensional coordinates. For trigger time, , , These are pitch angle, roll angle, and yaw angle, respectively; in the operation progress snapshot, the formula for calculating sub-area scan coverage is: in, For scan coverage, The scanned area. This represents the total area of ​​the sub-region; S5-2 Go-Ahead Preparation and Precise Backtracking: When the UAV restarts or resumes operations, after updating the environmental data, precise backtracking is achieved through a coordinate deviation correction formula. in, For coordinate deviation, For anchor point coordinates, The current coordinates, For correction factor, To backtrack to the target coordinates; the backtracking error is required to be | |≤5cm; S5-3 Dynamic Priority Replanning and Go-Around Execution: Based on the unfinished tasks, after recalculating the dynamic priority, the optimized formula for the go-around path length is as follows: in, This represents the total length of the go-around path. The distance between path nodes; the flight speed is adjusted based on the remaining battery power, and the adaptive speed formula is: in, Remaining battery power Fully charged For the speed of resuming flights; S5-4 Seamless Operation Status Transition: After the drone reaches the starting position of the unfinished task, the operation parameters are restored, and the data merging formula is as follows: in, For complete operation data, For snapshot data, Added data for resumption of flights; S5-5 Safety Redundancy Design: The formula for determining remaining power is: in, The amount of electricity needed to complete the remaining tasks. This represents the maximum flight distance on a full charge; if Then the return route will be automatically planned.

7. A library inventory drone adapted to the path planning method as described in any one of claims 1-6, characterized in that: The device includes a drone body, which houses a core control area using an STM32F4 MCU. The drone body has a Wi-Fi 6 Mesh antenna at the bottom, a LiDAR at the top, and an infrared thermal imaging sensor and a single / dual-lens camera at the bottom. The drone body also incorporates an ORB-SLAM3 vision module and a UWB ultra-wideband positioning module.

8. The library inventory drone according to claim 7, characterized in that, The four extended arms of the UAV are each equipped with a four-way obstacle avoidance module, which is a four-channel ultrasonic sensor; the core control area is also equipped with a UWB ultra-wideband positioning module, a low-noise motor and a flexible circuit board, and the operating noise of the UAV is controlled within 50dB.

9. The library inventory drone according to claim 7, characterized in that, The UAV body is equipped with an attitude control module. The core control of the attitude control module uses an STM32F407 microcontroller and is equipped with the FreeRTOS operating system. The sensor combination of the attitude control module is connected to the MPU-6050, HMC5883L, MS5611 and UWB ultra-wideband positioning module through the I2C / SPI interface.

10. The library inventory drone according to claim 7, characterized in that: The attitude control module uses a low-noise motor as its actuator, which supports precise PWM speed regulation. A depth camera is also installed inside the main body of the drone.

Images