Collaborative operation method for unmanned aerial vehicle and unmanned tractor

Abstract

The invention provides an unmanned aerial vehicle and unmanned tractor collaborative operation method, which comprises the following steps: step 1, scanning and constructing a two-dimensional grid map of a field environment through a laser radar installed on an unmanned aerial vehicle, and sending the two-dimensional grid map to an unmanned tractor; step 2, the unmanned tractor carries out field full-coverage path planning based on the two-dimensional grid map and by considering tractor steering constraints; step 3, performing a path tracking task by the unmanned tractor based on field full-coverage path planning; and 4, when the path tracking task is carried out, information obtained by real-time scanning of a laser radar carried by the robot is fused, obstacle detection is carried out, and the path is dynamically updated to be a collision-free path according to the obstacle.

Description

TECHNICAL FIELD

[0001] The present application relates to a method for unmanned equipment cooperative operation, in particular to a method for unmanned aerial vehicle and unmanned tractor cooperative operation. BACKGROUND

[0002] The information provided in this section is merely background information related to the present disclosure and can not necessarily be prior art.

[0003] At present, intelligent agricultural equipment technologies represented by unmanned aerial vehicles (UAV) and unmanned tractors (UT) are developing rapidly and showing great potential in precision agriculture. However, there are significant complementary limitations in the working capacity of unmanned aerial vehicles and unmanned tractors:

[0004] Unmanned aerial vehicles have wide field of view and flexible operation in the air, but have short endurance time and insufficient working intensity; unmanned tractors have strong working load and long working endurance, but have limited field of view.

[0005] Existing application modes mainly focus on independent operation of unmanned aerial vehicles or unmanned tractors, or only perform shallow and non-real-time information interaction (such as unmanned aerial vehicles planning tractor paths after preliminary mapping). This mode cannot fully exert the collaborative advantages of the two, and it is difficult to meet the higher requirements of modern agriculture on working efficiency, precision, real-time performance and adaptability. Therefore, how to break through the technical limitations of single equipment and build an efficient, real-time and intelligent unmanned aerial vehicle-unmanned tractor cooperative operation system to realize the deep integration of air-ground information, dynamic collaborative allocation and closed-loop execution of tasks is a key technical problem to improve the level of agricultural automation and intelligence. There is still a lack of mature, efficient and low-cost collaborative solutions in the prior art.

[0006] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0007] The present application aims to solve the technical problems of the prior art and provides a method for unmanned aerial vehicle and unmanned tractor cooperative operation.

[0008] In order to solve the above technical problems, the present application discloses a method for unmanned aerial vehicle and unmanned tractor cooperative operation, comprising the following steps:

[0009] Step 1: scanning and constructing a two-dimensional grid map of the field environment by a laser radar installed on the unmanned aerial vehicle, and sending it to the unmanned tractor;

[0010] Step 2, the unmanned tractor plans a full-coverage path in the field based on the two-dimensional grid map, considering the tractor steering constraints;

[0011] Step 3, the unmanned tractor performs a path tracking task based on the full-coverage path planning in the field;

[0012] Step 4, while performing the path tracking task, the information obtained by the laser radar carried by the unmanned tractor is fused in real time to detect obstacles, and the path is dynamically updated to a collision-free path according to the obstacles.

[0013] Further, the two-dimensional grid map of the field environment constructed in step 1 includes the following steps:

[0014] Step 1-1, pre-process the point cloud obtained by the laser radar scanning on the unmanned aerial vehicle, that is, perform motion distortion compensation and point cloud filtering to remove motion distortion and noise; wherein,

[0015] Motion distortion compensation corrects the point cloud by interpolation method using inertial measurement unit or odometer data to eliminate distortion caused by sensor motion;

[0016] Point cloud filtering reduces noise and removes invalid points by statistical outlier filtering;

[0017] Step 1-2, construct a three-dimensional point cloud map based on the LOAM method;

[0018] Step 1-3, project the three-dimensional point cloud map to a two-dimensional plane to generate a two-dimensional grid map.

[0019] Further, the two-dimensional grid map generated in step 1-3 includes:

[0020] Height filtering method is adopted, that is, a height threshold range z min ≤z≤z max is set, and only valid point clouds within the height threshold range are retained;

[0021] Set the map resolution to r, divide the field environment into an MxN grid array, and associate each grid with a numerical value, that is, a grid state, indicating the probability of the grid being occupied, and update the grid state using a logit model, which is represented as follows:

[0022]

[0023] Wherein, l(m i ) is the logit value of grid m i , that is, the grid state, p(m i |z t ) is the occupancy probability under the observation z t , that is, the ranging reading of the laser radar, and l0 is the prior value.

[0024] Further, the field full coverage path planning in step 2 comprises the following steps:

[0025] Step 2-1, initialize the grid state, start and end points, open list OpenList, closed list CloseList and coverage matrix Cover_Matrix;

[0026] Step 2-2, perform path search and four-neighborhood state evaluation, specifically as follows:

[0027] Step 2-2-1, take out the node N with the minimum total cost estimate value from the open list OpenList, move it into the closed list CloseList, and mark the position corresponding to the node N in the coverage matrix Cover_Matrix as 0, indicating that it has been covered;

[0028] Step 2-2-2, expand the four-neighborhood nodes P in the up, down, left and right directions of the node N, and for each neighborhood node P, perform the following operations in turn:

[0029] If the neighborhood node P is an obstacle or in the closed list CloseList, skip this neighborhood node P, otherwise set the cost from the node N to this neighborhood node P as 1, and calculate the temporary actual cost g temp , as follows:

[0030] g temp =g(N)+cost

[0031] If the neighborhood node P is not in the open list OpenList, add this neighborhood node P to the open list OpenList, and calculate the total cost estimate f(P) of the neighborhood node P to the end point G, specifically as follows:

[0032] g(P)=g temp

[0033] h(P)=Manhattan(P,G)

[0034] f(P)=g(P)+h(P)

[0035] And set the parent node of this neighborhood node P as the node N, wherein Manhattan(P,G) represents the Manhattan distance between the neighborhood node P and the end point G, g(P) is the actual cost from the start point S to the neighborhood node P, and h(P) is the heuristic estimated cost of the neighborhood node P to the end point G;

[0036] If the neighborhood node P is already in the open list OpenList and satisfies:

[0037] g temp < g(P)

[0038] Then the following updates are made:

[0039] g(P) = g temp

[0040] f(P) = g temp + h(P)

[0041] and set the parent of the neighbor node P to be the node N;

[0042] Step 2-2-3, if the open list OpenList is not empty, return to execute step 2-2-1.

[0043] Step 2-3, dead zone processing and restart, as follows:

[0044] Step 2-3-1, if the open list OpenList is empty and there is an element with a value of 1 in the cover matrix Cover_Matrix, select the node closest to the last node in the closed list CloseList from the nodes corresponding to the element with a value of 1 as a new starting point S new , and execute step 2-3-2, otherwise execute step 2-4;

[0045] Step 2-3-2, empty the open list OpenList and the closed list CloseList;

[0046] Step 2-3-3, add the new starting point S new to the open list OpenList, set the total cost estimate f(S new ), the actual cost g(S new ), and the heuristic estimate cost h(S new ) corresponding to the new starting point S new , as follows:

[0047] g(S new ) = 0

[0048] h(S new ) = Manhattan(S new , G)

[0049] f(S new ) = h(S new )

[0050] Step 2-3-3, return to execute step 2-2;

[0051] Step 2-4, path generation, as follows:

[0052] From the end point G, i.e. the last node in the CloseList, backtracking according to the parent node of the current node until the start point S, to generate a path point sequence;

[0053] Step 2-5, optimizing the path generated in step 2-4.

[0054] Further, the initialization of the grid state, the start point and the end point, the OpenList, the CloseList and the Cover_Matrix in step 2-1 are as follows:

[0055] Step 2-1-1, assigning a grid state to each grid in the grid array, marking all workable grids as 1 and marking obstacle grids as -1;

[0056] Step 2-1-2, setting the start point S and the end point G in the grid array;

[0057] Step 2-1-3, initializing the OpenList, as follows:

[0058] Adding the start point S to the OpenList;

[0059] Step 2-1-4, calculating the total cost estimate f(S) of the current node, i.e. the start point S to the end point G, as follows:

[0060] f(S) = g(S) + h(S)

[0061] Wherein, g(S) is the actual cost from the start point S to the current node, i.e. the start point S, which is the initial value, as follows:

[0062] g(S) = 0

[0063] h(S) is the heuristic estimated cost of the current node, i.e. the start point S to the end point H, as follows:

[0064] h(S) = Manhattan(S, G) = |x S -x G |+|y S -y G |

[0065] Wherein, Manhattan(S, G) is the Manhattan distance of the current node, i.e. the start point S to the end point G, (x S ,y S ) represents the coordinates of the start point S, and (x G ,y G ) represents the coordinates of the end point H.

[0066] Step 2-1-5, initialize the close list CloseList as an empty list;

[0067] Step 2-1-6, initialize the cover matrix Cover_Matrix, as follows:

[0068] Each element in the cover matrix Cover_Matrix corresponds to a grid array, and the workable grid is marked as 1, the obstacle is marked as -1, and the starting point S position is marked as 0.

[0069] Further, the path generated in step 2-4 is optimized in step 2-5, that is, the relationship between the minimum turning radius r of the unmanned tractor and the size of the working width w is determined, and the turning path in the generated path is re-planned, as follows:

[0070] When R = w / 2, a semicircular turn is adopted;

[0071] When R < w / 2, an arch-shaped turn is adopted;

[0072] When R > w / 2, a pear-shaped turn is adopted.

[0073] Further, the path tracking task in step 3 includes:

[0074] Step 3-1, calculate the preview distance L d , as follows:

[0075] L d = k v ·v + L min

[0076] Where k v is the speed coefficient; v is the current speed of the unmanned tractor; L min is the minimum preview distance;

[0077] Step 3-2, target point selection, as follows:

[0078] At a forward distance of L d on the path, select the target point p t (x t ,y t ) that satisfies the following conditions:

[0079] ‖p c -p t ‖2≥L d

[0080] Where p c (x c ,y c ) is the current position of the unmanned tractor;

[0081] Step 3-3, calculate the steering curvature κ, represented as follows:

[0082]

[0083] where θ c is the current heading angle of the unmanned tractor;

[0084] Step 3-4, calculate the desired front wheel steering angle δ of the unmanned tractor, represented as follows:

[0085] δ = arctan(κ·L w )

[0086] where L w is the tractor wheelbase;

[0087] Step 3-5, the unmanned tractor travels forward according to the calculated desired front wheel steering angle δ.

[0088] Further, the dynamic updating of the path to a collision-free path according to the obstacle in step 4 includes:

[0089] Step 4-1, obstacle judgment and path dynamic updating trigger, as follows:

[0090] Let the obstacle point cloud set scanned by the laser radar carried by the unmanned tractor at the current time be O, represented as follows:

[0091] O = {(x o ,y o )}

[0092] Set the judgment condition for whether the obstacle point cloud enters the forward safety area, represented as follows:

[0093]

[0094] where represents the existence, (x(t),y(t)) represents the current position of the unmanned tractor, R safe is the preset safety radius;

[0095] If the judgment condition is met, the path dynamic updating is triggered, step 4-2 is executed, otherwise step 4-1 is re-executed and the obstacle point cloud set is updated;

[0096] Step 4-2, speed space constraint modeling, as follows:

[0097] Let the linear speed and angular speed of the unmanned tractor at the current time be v and ω, the maximum linear acceleration be a min , and the maximum angular acceleration be b maxIf the control period is Δt, then the velocity search range for the next moment is defined as follows:

[0098] v∈[v min ,v max ]∩[va max ·Δt,v+a max ·Δt]

[0099] ω∈[ω min ,ω max ]∩[ω-b max ·Δt,v+b max ·Δt]

[0100] Get dynamic window The feasible set;

[0101] Step 4-3, for each candidate velocity pair Future prediction time domain T p Within, a kinematic model is used for trajectory simulation, as shown below:

[0102] x(t+Δt)=x(t)+v i cos(θ(t))·Δt

[0103] y(t+Δt)=y(t)+v i sin(θ(t))·Δt

[0104] θ(t+Δt)=θ(t)+ω i ·Δt

[0105] Where θ(t) represents the heading angle of the unmanned tractor at the current moment;

[0106] Based on the trajectory simulation results, the predicted trajectory τ is obtained. i (t), represented as follows:

[0107] τ i (t)={x(t),y(t),θ(t)}

[0108] A target function is set to evaluate each predicted trajectory τ. i (t), objective function G(v) i ,ω i ), which is represented as follows:

[0109]

[0110] Where, θ goal θ is the direction angle of the target point. end For the predicted trajectory τ i (t)'s end orientation angle; (x) o ,yo ) represents the obstacle point; The current point cloud set of obstacles;

[0111] Step 4-4, select the optimal speed pair (v) * ,ω * ), which is represented as follows:

[0112]

[0113] Steps 4-5: Use the optimal speed pair (v) * ,ω * The path is dynamically updated to a collision-free path by using the predicted trajectory and its corresponding trajectory.

[0114] Furthermore, the method also includes:

[0115] During steps 1 to 4, the drone docks and charges with the unmanned tractor according to preset conditions.

[0116] Furthermore, in step 5, the drone docks and charges with the unmanned tractor according to preset conditions, including:

[0117] Step 5-1: When the drone's battery level is lower than a preset value, interrupt the current task and send a charging request to the unmanned tractor.

[0118] Step 5-2: After receiving the charging request, the unmanned tractor stops in place, adjusts the magnetic charging platform to the working position, and broadcasts its own GIS coordinates.

[0119] Step 5-3: Based on the GIS coordinates broadcast by the unmanned tractor, the drone plans its landing path and lands on the magnetic charging platform for wireless charging.

[0120] Beneficial effects:

[0121] 1. This invention utilizes a two-dimensional grid map of the field environment acquired by a drone. Based on this map, the unmanned tractor takes into account the tractor's steering constraints and applies the A* algorithm to achieve full-coverage path planning for the field, significantly improving path feasibility and operational stability.

[0122] 2. The unmanned tractor uses the planned full-coverage path and Pure Pursuit for path tracking. During operation, it can combine its onboard LiDAR for real-time dynamic obstacle detection and apply the DWA algorithm to dynamically update the collision-free path, which significantly improves the safety of unmanned operation.

[0123] 3. A retractable magnetic wireless charging platform was designed at the rear of the unmanned tractor. Combined with the drone's autonomous return mechanism, dynamic wireless charging was achieved during operation, improving the overall operating time and the level of intelligent energy management of the system. Attached Figure Description

[0124] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0125] Figure 1 This is a schematic diagram of the air-ground collaborative and complementary operation process in this invention.

[0126] Figure 2 A schematic diagram of the full-coverage path planning process based on the A* algorithm, considering steering constraints for driverless tractors.

[0127] Figure 3 This is a schematic diagram of the air-ground collaborative energy management process in this invention. Detailed Implementation

[0128] This invention provides a method for collaborative operation between unmanned aerial vehicles (UAVs) and unmanned tractors. By enabling grid map generation based on UAV LiDAR, full-coverage path planning, path tracking, and real-time dynamic obstacle avoidance for the unmanned tractor, it achieves complementary air-ground collaborative operations. The core objective is to construct an efficient, real-time, and intelligent air-ground collaborative operation system that deeply integrates the complementary advantages of UAVs and unmanned tractors, achieving air-ground information sharing and collaborative decision-making, and significantly improving operational efficiency, accuracy, and intelligence in complex farmland environments.

[0129] Specifically, the present invention aims to solve the following key technical problems:

[0130] 1) Air-ground collaborative and complementary operation: The drone's lidar generates a grid map and sends it to the unmanned tractor; the unmanned tractor performs full-coverage path planning based on the map and then performs path tracking, while the lidar is equipped to achieve real-time dynamic obstacle avoidance, thereby achieving full-coverage operation in the field.

[0131] 2) Air-ground collaborative energy management: To address the pain point of poor drone endurance, a retractable magnetic charging platform is modified at the rear of the unmanned tractor to support wireless charging of drones.

[0132] To achieve complementary air-ground operations, this invention provides a method for collaborative operation between a drone and an unmanned tractor, enabling such operations. Figure 1As shown, the main steps include: Step 1: Constructing a two-dimensional grid map of the field environment using a LiDAR installed on the drone; Step 2: Sending the grid map to the unmanned tractor via the ROS distributed communication network. Based on the grid map and considering the tractor's steering constraints, the unmanned tractor uses the A* algorithm to plan a path covering the entire field; Step 3: The unmanned tractor uses the PurePursuit algorithm to complete the path tracking task based on the planned path, while simultaneously integrating the LiDAR information onboard in real time to perform real-time dynamic obstacle detection, and applying the DWA (Dynamic Window Approach) algorithm to dynamically update the collision-free path, thereby achieving full coverage operation of the field.

[0133] The specific technical solution of the present invention is as follows:

[0134] Step 1: Construct a 2D grid map of the field environment using a LiDAR device installed on the drone. Constructing a 2D grid map based on LiDAR mainly includes the following steps:

[0135] Step 1: Preprocessing of the laser point cloud. The raw point cloud data acquired by lidar typically contains motion distortion and noise, requiring preprocessing to improve subsequent matching accuracy. This primarily involves motion distortion compensation and point cloud filtering. Motion distortion compensation utilizes inertial measurement unit (IMU) or odometry data, employing interpolation methods (such as linear interpolation) to correct the point cloud and eliminate distortion caused by sensor motion. Point cloud filtering uses statistical outlier filtering to reduce noise and remove invalid points (such as those that are too close or too far away).

[0136] Step 2: Construct a 3D point cloud map based on the LOAM (Lidar Odometry and Mapping) algorithm. The specific steps are as follows:

[0137] a. Feature Extraction: Extract edge points and planar points from the point cloud. Edge points correspond to sharp features in the scene, and planar points correspond to continuous surfaces. b. Inter-Frame Matching: Calculate the pose transformation matrix between adjacent frames using the iterative nearest-neighbor algorithm. Its mathematical representation is:

[0138]

[0139] Where, p i For the current frame point cloud, q i c. Global optimization: By combining pose graph optimization with loop closure detection to correct accumulated errors, a globally consistent 3D point cloud map is generated.

[0140] Step 3: Perform 3D point cloud projection and generate a 2D raster map. To construct the 2D raster map, the 3D point cloud needs to be projected onto a 2D plane using a height filtering method, i.e., setting a height threshold z. min ≤z≤z ma Only valid point clouds near the ground are retained. The map resolution is set to r (e.g., 0.05 m / grid), and the environment is divided into an M×N grid array. Each grid is typically associated with a value between 0 and 1, representing the probability that the grid is occupied; 0 indicates the grid is definitely unoccupied (no obstacles); 1 indicates the grid is definitely occupied (an obstacle is present). The mapping algorithm updates the probability of all grid occupancy in real time based on the LiDAR ranging readings using the following formula. In the formula, z... t This refers to the ranging reading of the lidar. The grid state is updated using a logarithmic probability model, with the following formula:

[0141]

[0142] Wherein, l(m) i ) is the grid m i The logarithmic probability value, p(m) i |z t ) for observing z t The occupancy probability is given by l0, where l0 is the prior value. The occupancy probability increases for the grid cell containing the laser endpoint. The idle probability increases for the grid cells through which the line connecting the laser origin and endpoint passes.

[0143] Step 2: The raster map is sent to the unmanned tractor via the ROS distributed communication network. Based on the raster map and considering the tractor's steering constraints, the unmanned tractor uses the A* algorithm to perform full-coverage path planning for the field. Specifically, as follows... Figure 2 As shown:

[0144] Step 1: Initialize the grid state, start and end points, open list (OpenList), closed list (CloseList), and cover matrix (Cover_Matrix):

[0145] Assign a status value to each grid cell: all workable grid cells are marked as 1, and obstacles are marked as -1.

[0146] Set the starting point S (plane coordinates) and the ending point G (plane coordinates).

[0147] Initialize the open list OpenList;

[0148] Add the starting point S to the OpenList;

[0149] Let g(S) = 0 represent the actual cost from the starting point S to the current node S;

[0150] Using h(S) = Manhattan(S,G) = |x S -x G |+|y S -y G | represents the Manhattan distance from the current node S to the destination G, which serves as a heuristic estimate of the cost from S to G;

[0151] Let f(S) = g(S) + h(S) represent the total cost estimate from the starting point S to the ending point G.

[0152] Initialize and set a close list CloseList(empty list).

[0153] Initialize the Cover_Matrix (all workable grids are marked as 1, obstacles are marked as -1, and the starting point S is set to 0. Because this invention performs full-coverage path planning, the tractor is required to pass through all workable grids in the grid map. Therefore, a coverage matrix is ​​defined here to help determine whether each workable grid has been covered. It can be understood as copying the initialized grid map, and then changing the workable grid from 1 to 0 every time the tractor passes through a workable grid, indicating that it has been covered).

[0154] Step 2: Path search and four-neighbor state evaluation. Repeat the following steps until the open list OpenList is empty:

[0155] Take the node N with the smallest total valuation function value from the OpenList, move it into the CloseList, and mark the position of N in the Cover_Matrix as 0 (covered).

[0156] Expanding the 4 neighboring nodes P of node N (in the four directions: up, down, left, and right): For each neighboring node P:

[0157] If the neighboring node P is an obstacle (state -1) or in the CloseList, skip it; otherwise, calculate the cost from node n to neighboring node P (set to 1; in a grid map, the cost of each step is 1, so the cost value is 1); calculate the temporary actual cost:

[0158] g temp = g(N) + cost;

[0159] If the neighboring node P is not in the open list OpenList, then add it to the open list OpenList and set: g(P) = g temp, h(P) = Manhattan(P, G), f(P) = g(P) + h(P), set the parent node of the neighborhood node P as node N; Manhattan(P, G) represents the Manhattan distance between the neighborhood node P and the end point G;

[0160] If the neighborhood node P is already in the open list OpenList and g temp < g(P), then update g(P) = g temp , f(P) = g temp + h(P), set the parent node of the neighborhood node P as node N.

[0161] Even if node N is the end point G at this time, continue the search until full coverage, that is, there are no uncovered non - obstacle points in the coverage matrix Cover_Matrix, that is, points with a value of 1.

[0162] Step 3: Dead - zone handling and restart:

[0163] If the open list OpenList is empty but there are still uncovered non - obstacle points (i.e., points with a value of 1) in the coverage matrix Cover_Matrix: Select the uncovered point closest to the last node (i.e., the last visited point) in the current closed list CloseList as the new starting point S new ; Clear the open list OpenList and the closed list CloseList; Add S new to the open list OpenList, set g(S new ) = 0, h(S new ) = Manhattan(S new , G), f(S new ) = h(S new ); Return to Step 2.

[0164] If the open list OpenList is empty and there are no uncovered non - obstacle points in the coverage matrix Cover_Matrix, then go to Step 4.

[0165] Step 4: Path generation: Starting from the end point G, backtrack according to the parent node until the starting point S to generate a sequence of path points (stored in order).

[0166] Step 5: Path optimization: After completing the preliminary full - coverage path planning, to avoid large offsets during the turning process of the driverless tractor between rows and improve path continuity, judge the size relationship between the minimum turning radius R of the tractor and the working width w, and perform secondary planning on the turning paths in the planned path. Specifically:

[0167] When R = w / 2, semi-circular turning and line wrapping are adopted; when R < w / 2, arched turning and line wrapping are adopted; when R > w / 2, pear-shaped turning and line wrapping are adopted; each turning type can be represented as a standard geometric model and inserted into the original path through interpolation to achieve path continuity and turning feasibility.

[0168] Step 3: The driverless tractor uses the Pure Pursuit algorithm based on the planned path to complete the path tracking task. At the same time, it fuses the lidar information carried by itself in real time for real-time dynamic obstacle detection and applies the DWA algorithm to dynamically update the collision-free path. Specifically:

[0169] Use the Pure Pursuit algorithm to complete the path tracking task. Specifically:

[0170] Step 1: Preview distance L d Adaptive calculation:

[0171] L d = k v ·v + L min

[0172] Among them, k v is the speed coefficient; v is the current tractor speed; L min is the minimum forward viewing distance.

[0173] Step 2: Target point selection: Select the target point p d (x t , y t ) on the path at a distance of L t that satisfies

[0174] ‖p c − p t ‖2 ≥ L d

[0175] Among them, p c (x [[ID=�1]] c , y c ) is the current position of the tractor.

[0176] Step 3: Steering curvature κ calculation:

[0177]

[0178] Among them, θ c is the current heading angle of the tractor.

[0179] Step 4: Tractor expected front wheel angle calculation: <0,000530>δ = arctan(κ · L w )

[0181] Among them, L w This refers to the tractor's wheelbase.

[0182] When the lidar on the tractor detects a dynamic obstacle, it triggers DWA (Depth-Warping Analysis) to replan the local path and update it to a collision-free path, ensuring operational safety. Specifically:

[0183] Step 1: Obstacle Judgment and Triggering Mechanism: The lidar on the unmanned tractor scans the surrounding environment at a certain frequency, forming a point cloud set of obstacles at the current moment. And calculate whether the obstacle has entered the forward safe zone:

[0184]

[0185] in, Let (x(t), y(t)) represent the location of the unmanned tractor, and let (x(t), y(t)) represent the location of the unmanned tractor when an obstacle enters the set safety radius R. safe When this happens, DWA local obstacle avoidance planning is immediately triggered.

[0186] Step 2: Velocity Space Constraint Modeling: Let the current linear velocity and angular velocity of the tractor be v and ω respectively, and the maximum linear acceleration be a. min The maximum angular acceleration is b max If the control period is Δt, then the velocity search range for the next moment is defined as follows:

[0187] v∈[v min ,v max ]∩[va max ·Δt,v+a max ·Δt]

[0188] ω∈[ω min ,ω max ]∩[ω-b max ·Δt,v+b max ·Δt]

[0189] This leads to a dynamic window. The feasible set.

[0190] Step 3: For each candidate velocity pair (v) i ,ω i )∈W, in the future prediction time domain T p Within, the trajectory is simulated using the following kinematic model:

[0191] x(t+Δt)=x(t)+v i cos(θ(t))·Δt

[0192] y(t+Δt)=y(t)+v i sin(θ(t))·Δt

[0193] θ(t+Δt)=θ(t)+ω i ·Δt

[0194] Generate predicted trajectory τ i (t)={x(t),y(t),θ(t)}, where; each predicted trajectory is evaluated according to the following objective function:

[0195]

[0196] Where, θ goal θ is the direction angle of the target point. end Let x be the angle at the end of the trajectory; o ,y o ) represents the obstacle point; This is the current set of obstacle point clouds.

[0197] Step 4: Optimal Speed ​​Pair Selection and Execution: Select the optimal solution from all candidate speeds.

[0198]

[0199] The speed pair is used as the current control output to update the tractor's execution command, thereby achieving local trajectory correction and dynamic obstacle avoidance.

[0200] Furthermore, considering the vastness of field operations and the limited range of drones, to avoid drones having to travel long distances to and from charging stations during collaborative operations, unmanned tractors can wirelessly charge drones nearby. A retractable magnetic charging platform can be modified at the rear of the unmanned tractor, enabling collaborative energy management between air and ground. Figure 3 As shown, it includes the following steps:

[0201] Step 1: When the drone's battery level is less than 20%, interrupt the mission and send a charging request to the ROS network;

[0202] Step 2: After responding to the request, the unmanned tractor remains stationary, extends its magnetic charging platform to the working position, and broadcasts its own GIS coordinates.

[0203] Step 3: The drone plans its landing path based on the received coordinates, achieves precise landing, magnetic docking, and completes wireless charging.

[0204] Example:

[0205] A specific embodiment of the UAV-unmanned tractor collaborative operation system of the present invention is as follows:

[0206] First, a drone (FanciSwarm open-source drone) equipped with a lidar generates 3D point cloud data containing field information in real time using the LOAM algorithm. After height filtering, the data is projected into a 2D grid map. The obstacle occupancy probability is dynamically updated using a logarithmic probability model.

[0207] After receiving 2D grid map data, the unmanned tractor (ET504 electric tractor) converts latitude and longitude coordinates into planar coordinates using the central meridian 120°E as a reference and the Gaussian projection formula. For irregularly shaped fields, the system uses a convex hull algorithm to expand them into regular rectangular areas. During the path planning phase, the field is divided into grids matching the tractor's width. The optimal turning type is dynamically selected based on the tractor's minimum turning radius and working range, including semi-circular turns, arched turns, or pear-shaped turns. A smooth turning path is generated using a cubic spline interpolation algorithm. If the path planning process gets stuck in a local dead zone, the system automatically calls the A* escape algorithm to replan the path.

[0208] During the path tracking phase, the system employs a pure tracking algorithm for precise path following. The pre-aiming distance is dynamically adjusted based on the tractor's current speed, and the steering curvature, after precise calculation, outputs front wheel steering angle commands via a PID controller. Simultaneously, the tractor-mounted LiDAR continuously scans the surrounding environment at a frequency of 10Hz. When an obstacle is detected entering a preset safety radius, the DWA algorithm is immediately triggered for dynamic obstacle avoidance. The speed search space is strictly constrained, and the trajectory evaluation function comprehensively considers multiple factors such as target orientation, obstacle avoidance distance, and speed priority to ensure the system's real-time response capability.

[0209] To address the issue of insufficient drone endurance, the system employs an intelligent air-ground collaborative energy management solution. When the drone's battery monitoring module detects that the remaining battery level is below 20%, it sends a charging request via the ROS communication network, including key information such as current coordinates and remaining battery power. Upon receiving the request, the unmanned tractor deploys its tail-integrated retractable charging platform via an electric push rod, and the magnetic docking interface is activated in preparation for docking. The drone plans its return path based on the A* algorithm, switches to visual guidance mode upon approaching the charging platform, achieves precise positioning by recognizing the ArUco code on the platform, and automatically activates the magnetic docking device to complete docking when it is 0.2 meters away from the platform. Subsequently, the wireless charging module performs rapid charging.

[0210] In its specific implementation, this application provides a computer storage medium and a corresponding data processing unit. The computer storage medium is capable of storing a computer program, which, when executed by the data processing unit, can run the invention's content regarding a method for collaborative operation of a drone and an unmanned tractor, as well as some or all of the steps in various embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0211] Those skilled in the art will clearly understand that the technical solutions in the embodiments of the present invention can be implemented using computer programs and their corresponding general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of computer programs, i.e., software products. These computer program software products can be stored in a storage medium and include several instructions to cause a device containing a data processing unit (which may be a personal computer, server, microcontroller, MCU, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0212] This invention provides a concept and method for collaborative operation of unmanned aerial vehicles (UAVs) and unmanned tractors. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for collaborative operation of a drone and an unmanned tractor, characterized in that, Includes the following steps: Step 1: Scan and construct a two-dimensional grid map of the field environment using a lidar mounted on the drone, and send it to the unmanned tractor; Step 2: Based on a two-dimensional grid map, the unmanned tractor performs full-coverage path planning for the field, taking into account the tractor's steering constraints. Step 3: The unmanned tractor performs path tracking based on full field coverage path planning; Step 4: While performing path tracking, the system integrates information obtained from real-time scanning by its onboard LiDAR to detect obstacles and dynamically update the path to a collision-free path based on the obstacles detected.

2. The method for collaborative operation of a drone and an unmanned tractor according to claim 1, characterized in that, The construction of a two-dimensional raster map of the field environment described in step 1 includes the following steps: Step 1-1 involves preprocessing the point cloud obtained from the lidar scan on the UAV, specifically performing motion distortion compensation and point cloud filtering to remove motion distortion and noise; among these steps... Motion distortion compensation uses inertial measurement unit or odometer data to correct point clouds through interpolation methods, eliminating distortion caused by sensor motion. Point cloud filtering employs statistical outlier filtering to reduce noise and remove invalid points; Steps 1-2: Construct a 3D point cloud map based on the LOAM method; Steps 1-3 project the 3D point cloud map onto a 2D plane to generate a 2D raster map.

3. The method for collaborative operation of a drone and an unmanned tractor according to claim 2, characterized in that, The generation of the two-dimensional raster map described in steps 1-3 includes: The height filtering method is used, that is, a height threshold range z is set. min ≤z≤z max Only retain valid point clouds within the height threshold range; Set the map resolution to r, and divide the field environment into an M×N grid array. Each grid is associated with a value, i.e., the grid state, which represents the probability that the grid is occupied. The grid state is updated using a log-odds model, as shown below: Wherein, l(m) i ) is the grid m i The logarithmic probability value, i.e., the grid state, p(m) i |z t ) for observing z t That is, the occupancy probability under the ranging reading of the lidar, where l0 is the prior value.

4. The method for collaborative operation of a drone and an unmanned tractor according to claim 3, characterized in that, Step 2, which describes the full-coverage path planning for farmland, includes the following steps: Step 2-1: Initialize the grid state, start and end points, open list (OpenList), closed list (CloseList), and cover matrix (Cover_Matrix); Step 2-2 involves path search and four-neighbor state evaluation, as detailed below: Step 2-2-1: Take the node N with the smallest total cost estimate from the OpenList, move it to the CloseList, and mark the position of the corresponding node N in the Cover_Matrix as 0, indicating that it has been covered; Step 2-2-2: Expand the four neighboring nodes P in the four directions (up, down, left, and right) of node N. For each neighboring node P, perform the following operations in sequence: If the neighboring node P is an obstacle or is in the CloseList, then skip the neighboring node P; otherwise, set the cost from node N to the neighboring node P to 1, and calculate the temporary actual cost g. temp , means as follows: g temp =g(N)+cost If the neighboring node P is not in the open list, then add the neighboring node P to the open list and calculate the total cost estimate f(P) of the current node to reach the destination G from the neighboring node P, as follows: g(P)=g temp h(P) = Manhattan(P,G) f(P) = g(P) + h(P) And set the parent node of the neighboring node P to node N, where Manhattan(P,G) represents the Manhattan distance between the neighboring node P and the destination G, g(P) is the actual cost from the starting point S to the current node, i.e. the neighboring node P, and h(P) is the heuristic estimated cost from the current node, i.e. the neighboring node P, to the destination G. If the neighboring node P is already in the open list OpenList and satisfies: g temp <g(p) Then perform the following updates: g(p)=g temp f(P)=g temp +h(P) And set the parent node of the neighboring node P to node N; Step 2-2-3, if the open list OpenList is not empty, then return to execute Step 2-2-1; Step 2-3, dead zone handling and restart, specifically as follows: Step 2-3-1: If the open list OpenList is empty and the cover matrix Cover_Matrix contains an element with a value of 1, then select the node closest to the last node in the close list CloseList from the node corresponding to the element with a value of 1 as the new starting point S. new If the condition is met, proceed to step 2-3-2; otherwise, proceed to step 2-4. Step 2-3-2, clear the open list OpenList and the closed list CloseList; Step 2-3-3, set the new starting point S new Add to OpenList and set a new starting point S. new The corresponding total cost estimate f(S) new ), actual cost g(S) new ) and heuristic cost estimation h(S) new ), which is represented as follows: g(S new )=0 h(S new )=Manhattan(S new ,G) f(S new )=h(S new ) Step 2-3-3, return to execute Step 2-2; Step 2-4, path generation, specifically as follows: Starting from the end point G, which is the last node in the closed list CloseList, trace back according to the parent node of the current node until the start point S to generate a sequence of path points; Step 2-5, optimize the path generated in Step 2-4.

5. A method for collaborative operation of a drone and an unmanned tractor according to claim 4, characterized in that, The initialization of the grid state, start and end points, open list OpenList, closed list CloseList, and coverage matrix Cover_Matrix described in Step 2-1 is specifically as follows: Step 2-1-1, assign grid states to each grid in the grid array, mark all operable grids as 1, and mark obstacle grids as -1; Step 2-1-2, set the start point S and the end point G in the grid array; Step 2-1-3, initialize the open list OpenList, specifically as follows: Add the start point S to the open list OpenList; Step 2-1-4, calculate the total cost estimate f(S) from the current node, i.e., the start point S, to the end point G, which is expressed as follows: f(S) = g(S) + h(S) where g(S) is the actual cost from the start point S to the current node, i.e., the start point S, and is the initial value at this time, which is expressed as follows: g(S) = 0 h(S) is the heuristic estimated cost from the current node, i.e., the start point S, to the end point H, which is expressed as follows: h(S)=Manhattan(S,G)=|x S -x G |+|y S -y G | Where Manhattan(S,G) is the Manhattan distance from the current node (starting point S) to the ending point G, (x S ,y S (x) represents the coordinates of the starting point S. G ,y G () represents the coordinates of the endpoint G; Step 2-1-5, initialize the closed list CloseList as an empty list; Step 2-1-6, initialize the coverage matrix Cover_Matrix, specifically as follows: Each element in the coverage matrix Cover_Matrix corresponds one-to-one with the grid array, mark the operable grids as 1, the obstacles as -1, and the position of the start point S as 0.

6. A method for collaborative operation of a drone and an unmanned tractor according to claim 5, characterized in that, The optimization of the path generated in Step 2-4 described in Step 2-5, that is, judge the size relationship between the minimum turning radius R and the working width w of the driverless tractor, and perform quadratic programming on the turning paths in the generated path, specifically as follows: When R = w / 2, use a semi-circular turn; When R < w / 2, use an arched turn; When R > w / 2, use a pear-shaped turn.

7. A method for collaborative operation of a drone and an unmanned tractor according to claim 6, characterized in that, The path tracking task described in Step 3 includes: Step 3-1, calculate the aiming distance L d , means as follows: L d =k v ·v+L min Where, k v L is the speed coefficient; v is the current speed of the driverless tractor; L min This is the minimum forward sight distance; Step 3-2, target point selection, specifically as follows: The forward distance along the path is L. d Select target point p. t (x t ,y t ), satisfying the following conditions: ‖p c -p t ‖2≥L d Where, p c (x c ,y c () represents the current location of the driverless tractor; Step 3-3, calculate the steering curvature κ, which is expressed as follows: Where, θ c The current heading angle of the unmanned tractor; Step 3-4, calculate the expected front wheel angle δ of the driverless tractor, which is expressed as follows: δ=arctan(κ·L w ) Among them, L w This refers to the tractor's wheelbase. Step 3-5, the driverless tractor moves forward according to the calculated expected front wheel angle δ.

8. A method for collaborative operation of a drone and an unmanned tractor according to claim 7, characterized in that, The dynamic update of the path to a collision-free path according to the obstacles described in Step 4 includes: Step 4-1, obstacle detection and path dynamic update triggering, is as follows: Let the set of obstacle point clouds obtained by the lidar on the unmanned tractor at the current moment be . It is expressed as follows: The criteria for determining whether an obstacle point cloud enters the forward safe zone are set as follows: in, Let R represent the position of the driverless tractor at the current moment, where (x(t), y(t)) represents the position of the driverless tractor. safe The preset safety radius; If the judgment condition is met, the path dynamic update is triggered and step 4-2 is executed; otherwise, step 4-1 is executed again and the obstacle point cloud set is updated. Step 4-2, velocity space constraint modeling, as follows: Let the linear velocity and angular velocity of the unmanned tractor at the current moment be v and ω, respectively, and the maximum linear acceleration be a. min The maximum angular acceleration is b max If the control period is Δt, then the velocity search range for the next moment is defined as follows: v∈[v min ,v max ]∩[v-a max ·Δt,v+a max ·Δt] ω∈[ω min ,oh max ]∩[ω-b max ·Δt,v+b max ·Δt] Get dynamic window The feasible set; Step 4-3, for each candidate velocity pair Future prediction time domain T p Within, a kinematic model is used for trajectory simulation, as shown below: x(t+Δt)=x(t)+v i cos(θ(t))·Δt y(t+Δt)=y(t)+v i sin(θ(t))·Δt θ(t+Δt)=θ(t)+ω i ·Δt Where θ(t) represents the heading angle of the unmanned tractor at the current moment; Based on the trajectory simulation results, the predicted trajectory τ is obtained. i (t), represented as follows: τ i (t)={x(t),y(t),θ(t)} A target function is set to evaluate each predicted trajectory τ. i (t), objective function G(v) i ,ω i ), which is represented as follows: Where, θ goal θ is the direction angle of the target point. end For the predicted trajectory τ i (t)'s end orientation angle; (x) o ,y o ) represents the obstacle point; The current point cloud set of obstacles; Step 4-4, select the optimal speed pair (v) * ,ω * ), which is represented as follows: Steps 4-5: Use the optimal speed pair (v) * ,ω * The path is dynamically updated to a collision-free path by using the predicted trajectory and its corresponding trajectory.

9. A method for collaborative operation of a drone and an unmanned tractor according to claim 8, characterized in that, The method further includes: During steps 1 to 4, the drone docks and charges with the unmanned tractor according to preset conditions.

10. A method for collaborative operation of a drone and an unmanned tractor according to claim 9, characterized in that, Step 5 involves the drone docking and charging with the unmanned tractor according to preset conditions, including: Step 5-1: When the drone's battery level is lower than a preset value, interrupt the current task and send a charging request to the unmanned tractor. Step 5-2: After receiving the charging request, the unmanned tractor stops in place, adjusts the magnetic charging platform to the working position, and broadcasts its own GIS coordinates. Step 5-3: Based on the GIS coordinates broadcast by the unmanned tractor, the drone plans its landing path and lands on the magnetic charging platform for wireless charging.

Images