Orchard electric crawler chassis and navigation method thereof

Abstract

The invention discloses an orchard electric crawler chassis and a navigation method thereof. The electric crawler chassis for the orchard comprises a walking chassis, a power battery, a control box and a related sensor. The navigation method comprises the following five steps: acquiring orchard point cloud through a three-dimensional laser radar mounted at the top of an orchard electric crawler chassis shell, extracting a region of interest by using straight-through filtering, performing down-sampling filtering on voxel point cloud, performing fitting removal on ground point cloud, performing statistical filtering denoising and performing two-dimensional projection on the point cloud. Fitting fruit tree row straight lines on two sides based on a random sampling consensus algorithm and generating a navigation center line; the expected speed is converted into the track rotating speed through a path tracking control algorithm, and the chassis is driven to walk along the navigation path. According to the invention, the navigation precision, the real-time performance and the terrain trafficability in the orchard environment are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural machinery automation technology, specifically to an electric tracked chassis suitable for orchard operations in complex terrain and its navigation method, particularly including an electrically driven power battery, a multi-sensor fusion navigation architecture, and a point cloud real-time processing and path planning algorithm for orchard row scenarios. Background Technology

[0002] Traditional orchard machinery relies heavily on manual operation or simple path planning, resulting in low efficiency, poor terrain adaptability, and insufficient precision. Existing orchard mobile platforms mostly employ wheeled structures, making them prone to slipping and getting stuck in muddy or uneven terrain. While tracked chassis can improve mobility, traditional mechanical transmission schemes suffer from low efficiency, high noise, and poor control precision, failing to meet the requirements of precision agriculture. Current orchard navigation solutions primarily rely on global navigation satellite systems (GPS) or visual sensors. The former suffers from high signal loss rates when fruit trees obstruct the view, while the latter is significantly affected by changes in lighting conditions. Three-dimensional lidar point cloud processing algorithms generally suffer from the following problems: ground weeds and low obstacles cause feature extraction errors; traditional Hough transform linear fitting is sensitive to noise. Therefore, there is an urgent need for an orchard operation platform that combines high mobility with strong anti-interference navigation capabilities. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a highly capable electric tracked chassis for orchards, solves the terrain adaptability problem, and proposes a navigation method that combines point cloud hierarchical dimensionality reduction processing with multimodal path planning to improve the robustness of orchard navigation.

[0004] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0005] An electric tracked chassis for orchards includes a frame and reducers bolted to the left and right sides of the frame in the forward direction of travel. A motor is mounted directly behind the reducers and connected to their shafts, with the motor housing bolted to the reducer housing. A set of track wheels is mounted on each of the two reducers away from the frame, each track wheel including a drive wheel, a road wheel, a tension wheel, a trailing wheel, and tracks. A power battery is fixed via a slot and bolted to the center of the rear side of the frame. A control box is mounted in the center of the frame. The control box houses an industrial control computer, a programmable logic controller (PLC), an inertial navigation sensor, and a motor driver. The industrial control computer processes data acquired by a three-dimensional lidar sensor. The system collects point cloud data and generates a navigation path, calculates the desired linear velocity and angular velocity, converts them into left and right track speeds, and transmits them to the programmable logic controller (PLC). The PLC converts the speed commands issued by the industrial control computer into CAN messages and sends them to the motor driver. The inertial navigation sensor measures the three-axis acceleration and three-axis angular velocity of the chassis to correct the data errors of the 3D LiDAR. The 3D LiDAR is fixed on the upper surface of the 3D LiDAR mounting bracket, which is installed at the top front of the orchard electric track chassis shell. The upper surface of the 3D LiDAR mounting bracket is inclined at an angle θ0 to the ground.

[0006] In the above scheme, the drive wheel is front-driven, and six pairs of load-bearing wheels are evenly arranged along the length of the frame. The towing wheel is arranged on the upper rear side of the frame and cooperates with the tensioning mechanism.

[0007] In the above scheme, the motor is a DC brushless permanent magnet motor.

[0008] In the above scheme, the track is made of high-strength wear-resistant polyurethane composite material, and the drive wheel is rigidly connected to the output end of the reducer.

[0009] A navigation method for an electric tracked chassis in an orchard:

[0010] Step (1): Collect raw point cloud data between rows of orchard in real time using 3D LiDAR and transmit it to an industrial control computer;

[0011] Step (2): Transform the point cloud data in the three-dimensional lidar coordinate system to the world coordinate system, including determining the radar tilt angle θ0, establishing the radar coordinate system and the world coordinate system, and performing point cloud coordinate transformation through the coordinate transformation matrix;

[0012] Step (3): Preprocess the converted point cloud data, including pass-through filtering to extract the region of interest, voxel point cloud downsampling filtering, ground point cloud fitting and removal, statistical filtering for noise reduction, and point cloud two-dimensional projection;

[0013] Step (4): Project onto Two-dimensional point cloud on a plane The midpoint line of the axis divides the points into left and right sides. The random sampling consensus algorithm is used to fit straight lines to the fruit tree point clouds on the left and right sides respectively; the midline of the two fitted lines is calculated as the global navigation path.

[0014] Step (5): The path tracking control algorithm is executed by the industrial control computer to convert the calculated expected linear velocity and angular velocity into the linear velocity of the left and right tracks. Then, the linear velocity of the tracks is converted into the motor speed. The speed is sent to the motor driver in the form of CAN message to control the motor to drive the chassis to move along the navigation path.

[0015] Furthermore, in step (3), the threshold parameters for the X1 axis direction of the pass-through filtering extraction region of interest are set to (2d, 22d), the threshold parameters for the Y1 axis direction are (-2D, 2D), and the threshold parameters for the Z1 axis direction are (-0.5H, 2.5H), where the parameter variables d=0.2m, D=1m, and H=1m; and the three-dimensional lidar point cloud data that is far away from the fruit tree row channel and too close or far away are cropped.

[0016] Furthermore, the voxel size of the voxel point cloud downsampling filter in step (3) is set to 0.04m×0.04m×0.04m.

[0017] Furthermore, the ground point cloud removal in step (3) employs a ground plane fitting algorithm, including:

[0018] The first step is to initialize the seed point set: according to the formula Find the lowest point in the point cloud dataset. The parameter k refers to the point cloud dataset after voxel point cloud downsampling. Along The positive direction divides space into an average number of sub-planes; The point cloud within a certain subplane is defined according to The lowest height point after sorting along the axis, where n=1, 2, 3 k; the average height of the lowest elevation points of all subplanes Consider it the lowest point in the point cloud dataset; set the threshold parameter. Select point cloud datasets with heights within the range of With threshold The points in the middle of the sum are used as the seed point set. ;

[0019] The second step, plane fitting: calculating the seed point set. The covariance matrix, substituted into the seed point set The planar model is obtained by taking the mean.

[0020] The third step is point cloud classification: calculating the point cloud dataset. Distance from each point within the plane to the orthogonal projection of the plane ,like If a point is marked as ground, it is otherwise considered a non-ground point.

[0021] The fourth step is iterative optimization: the current ground point cloud is used as the seed point set input for the next iteration and the iteration continues until the ground point cloud can be well segmented.

[0022] Furthermore, the point cloud preprocessing in step (3) reduces the amount of data by more than 90%, including a 30% reduction in data volume through pass-through filtering to extract regions of interest, an 80% reduction in data volume through voxel point cloud downsampling, a 10% reduction in data volume through ground removal, and a 9% reduction in data volume through statistical filtering.

[0023] Furthermore, step (4) is implemented as follows:

[0024] The obtained two-dimensional point cloud dataset is obtained by The midpoint line divides the points into two clusters on the left and right sides. The left side shows a two-dimensional point cloud dataset. The right side is a two-dimensional point cloud dataset; the random sampling consensus algorithm is used to fit the projected point clouds of the fruit tree rows on both sides to a straight line, and the best straight line that best represents the spatial distribution of the point cloud of the fruit tree rows on that side is fitted respectively.

[0025] The point cloud lines on the left and right sides were fitted using the random sampling consistency algorithm. The expression is shown in (3). The navigation path is the midline of the two fitted lines, and the final navigation line is obtained. The trajectory expression is shown in (4).

[0026] (3)

[0027] (4)

[0028] In formula (3) and They represent navigation lines respectively. of Coordinates, navigation lines of Coordinate values and They represent navigation lines respectively. The slope, and They represent navigation lines respectively. of Coordinates, navigation lines of Coordinate values and They represent navigation lines respectively. The intercept coefficient.

[0029] Furthermore, the conversion of the desired linear velocity and angular velocity into the linear velocities of the left and right tracks in step 5 is implemented as follows:

[0030] (5)

[0031] The desired linear velocity; The desired angular velocity; and These are the linear velocities of the left and right tracks, respectively. This represents the virtual wheel spacing.

[0032] Beneficial effects of the present invention

[0033] Compared with existing technologies, the technical effects achieved by this invention are as follows:

[0034] 1. Strong terrain adaptability: The polyurethane track + front drive wheel + multiple load wheels design improves passability by 35% and is suitable for hilly orchards with a slope of ≤15°.

[0035] 2. High navigation accuracy: Tightly coupled positioning lateral deviation ≤0.15m, point cloud processing time reduced by 60%, and end-of-line turn success rate >95%;

[0036] 3. Strong real-time performance and robustness: The amount of preprocessed point cloud is reduced by more than 90%, and the obstacle avoidance response time of the local path planning algorithm is less than 200ms, which is suitable for dynamic orchard environments. Attached Figure Description

[0037] Figure 1 This diagram illustrates the overall structure of an orchard electric tracked chassis provided by an embodiment of the present invention.

[0038] Figure 2 This diagram illustrates the structural layout of an orchard electric tracked chassis provided in an embodiment of the present invention.

[0039] Figure 3 This is a flowchart illustrating a navigation control method for an orchard electric tracked chassis provided in an embodiment of the present invention;

[0040] Figure 4 This diagram illustrates the navigation line fitting of a navigation control method for an orchard electric tracked chassis provided in an embodiment of the present invention.

[0041] In the diagram: 1-frame, 2-drive wheel, 3-road wheel, 4-tensioning wheel, 5-track roller, 6-track, 7-reducer, 8-motor, 9-tensioning mechanism, 10-power battery, 11-industrial control computer, 12-programmable logic controller, 13-motor driver, 14-inertial navigation sensor, 15-3D LiDAR, 16-3D LiDAR mounting bracket, 17-shell. Detailed Implementation

[0042] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings and embodiments. It should be noted that the described embodiments are only a part of the embodiments of the present invention, and not all possible implementations. Other embodiments that can be obtained by those skilled in the art without creative effort based on the principles of the present invention are all within the protection scope of the present invention.

[0043] This application embodiment discloses an electric tracked chassis for orchards and its navigation method. The electric tracked chassis for orchards includes a walking chassis, a power battery 10, a control box, and related sensors.

[0044] like Figure 1 , 2 As shown, the chassis includes a frame 1, two sets of tracked wheel systems (drive wheels 2, road wheels 3, tension wheels 4, trailing wheels 5, and tracks 6), a reducer 7, and a motor 8. The two sets of tracked wheel systems are located on both sides of the frame 1. The drive wheels 2 are positioned at the front side of the frame 1 (front-drive). Six pairs of road wheels 3 are evenly distributed along the length of the frame 1. The trailing wheels 5 are positioned at the upper rear side of the frame 1 and cooperate with the tensioning mechanism 9. The tracks 6 are made of high-strength, wear-resistant polyurethane composite material to adapt to the needs of orchard operations in different terrains. The power input end of the drive wheels 2 is connected to the output end of the reducer 7, and the input end of the reducer 7 is connected to the power output end of the corresponding motor 8. The motor 8 is a DC brushless permanent magnet motor, which features high efficiency, energy saving, reliability, and high power transmission efficiency. This design gives the chassis excellent terrain adaptability and passability, making it particularly suitable for muddy and uneven orchard operating environments.

[0045] The power battery 10 is the sole power source for the orchard electric tracked chassis. During operation, the electrical energy required for the motor 8 to run is directly provided by the power battery 10. Under the control of the motor driver 13, the motor 8 outputs torque, which is amplified by the reducer 7 and then transmitted to the drive wheel 2, thereby driving the entire track wheel system to rotate and propelling the orchard electric tracked chassis forward.

[0046] Control box and related sensors: The control box includes an industrial control computer 11 and a programmable logic controller 12. The industrial control computer 11 processes 3D point cloud data to generate navigation paths and calculates the rotational speeds of the two motors, and transmits the information to the programmable logic controller 12. The programmable logic controller 12 is responsible for converting the target rotational speed of the motor 8 into a CAN message and sending it to the motor driver 13. The motor driver 13 drives the two motors 8 according to the rotational speed CAN message sent by the programmable logic controller 12. The inertial navigation sensor 14 is used to measure the three-axis acceleration and three-axis angular velocity of the orchard electric tracked chassis, providing high-frequency attitude and displacement information and transmitting this information to the industrial control computer 11 to correct the positional error of the data collected by the 3D LiDAR 15. The 3D LiDAR mounting bracket 16 is installed on the top front of the orchard electric tracked chassis shell 17. The 3D LiDAR 15 is fixed to the 3D LiDAR mounting bracket 16 with bolts and is used to acquire high-precision 3D point cloud data of the orchard environment.

[0047] As another implementation scenario of this embodiment, based on the above content, a navigation method for an electric tracked chassis in an orchard is proposed. Figure 3 This is a flowchart illustrating a navigation control method for an electric tracked chassis in an orchard, as provided in this embodiment. The specific process of this method is as follows:

[0048] Step (1), raw point cloud data acquisition: First, the three-dimensional lidar 15 mounted on the top front of the orchard electric track chassis shell 17 collects the orchard point cloud data between rows and transmits it to the host computer to obtain continuous three-dimensional lidar point cloud frame data.

[0049] Step (2), point cloud coordinate transformation: First, determine the tilt angle of the 3D lidar 15. Then, through point cloud coordinate transformation, the point cloud data in the three-dimensional lidar coordinate system is transformed into point cloud data in the world coordinate system;

[0050] Step (2.1), determine the tilt angle of the 3D lidar 15: the tilt angle of the 3D lidar is the radar coordinate system The angle between the axis and the direction perpendicular to the ground; in this embodiment ;

[0051] Step (2.2), establish the radar coordinate system: The three-dimensional lidar coordinate system is a coordinate system relative to the three-dimensional lidar 15 itself; the origin O of this three-dimensional lidar coordinate system is the transmission center of the three-dimensional lidar 15; the three-dimensional lidar coordinate system The axis is perpendicular to the scanning plane of the 3D LiDAR 15, and its direction is vertically upward; 3D LiDAR coordinate system The axis remains parallel to the scanning plane of the 3D LiDAR 15, and The axis points to the starting position of the scanning circle of the 3D LiDAR 15, i.e., the 0° direction; 3D LiDAR coordinate system The axis is determined by the right-hand screw rule;

[0052] Step (2.3), establish the world coordinate system: set the origin of the world coordinate system. The point where the origin O of the 3D lidar coordinate system is perpendicularly downwards to the ground is the intersection point; the horizontal component of the operating velocity vector of the orchard electric tracked chassis is the world coordinate system. The direction of the axis; world coordinate system The axis is a vertical axis pointing upwards from the ground; world coordinate system. The axis is determined by the left-handed spiral rule;

[0053] Step (2.4), Point Cloud Coordinate Transformation: Traverse each point in the point cloud data and use a coordinate transformation matrix to transform the point cloud data from the 3D LiDAR coordinate system to the world coordinate system. The point cloud coordinate transformation formula is as follows:

[0054]

[0055] In the formula, The tilt angle of the 3D lidar. ; , , These are the point cloud coordinates in the 3D lidar coordinate system. x, y, and z are the point cloud coordinates in the world coordinate system. .

[0056] Step (3), point cloud processing: Before performing point cloud navigation line fitting and extraction, the original point cloud needs to be processed. There are many interfering point clouds in the middle channel of the original fruit tree point cloud row. It is necessary to trim the indistinct area of ​​the middle channel and remove the ground point cloud interference. After filtering the outlier point cloud, the point cloud that retains the characteristics of the fruit tree row channel and has good real-time performance is obtained, laying the foundation for subsequent navigation line fitting.

[0057] Step (3.1), through-pass filtering to extract the region of interest: The near-end cloud data is dense but too cluttered, which is not conducive to subsequent line fitting. It is necessary to extract a suitable region of interest from the dataset. Using the through-pass filtering algorithm, set... The threshold parameters in the axial direction are (2d, 22d). The threshold parameter in the axial direction is (-2D, 2D). The threshold parameters in the axial direction are set to (-0.5H, 2.5H), and the parameter variables d, D, and H are respectively... axis axis The reference unit in the axial direction is modified according to different orchard environments; in this embodiment, d=0.2m, D=1m, H=1m, that is... The actual distance along the axis ranges from 0.4m to 4.4m. The actual distance along the axis is from -2m to 2m. The actual distance along the axis is from -0.5m to 2.5m; 3D LiDAR point cloud data that is far from the fruit tree row passage and too close or far away are cropped.

[0058] Step (3.2), voxel point cloud downsampling filtering: In order to improve the real-time performance of subsequent point cloud processing, voxel mesh downsampling processing is performed on the point cloud in the region of interest. The voxel size is set to (0.04, 0.04, 0.04), which is a cube with a side length of 0.04m. This can greatly reduce the number of points while preserving the original point cloud features.

[0059] Step (3.3), Ground fitting removal: Ground point cloud is segmented and removed using a ground plane fitting algorithm;

[0060] Step 1: Initialize the seed point set: The point cloud dataset after voxel point cloud downsampling is arranged along... The positive direction divides the space into k equal sub-planes, and the point cloud sub-planes are numbered sequentially as follows: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 ... The ground plane fitting algorithm is performed on k sub-planes, and the point cloud in each sub-plane is divided according to... Sorting along the axis, then selecting the lowest elevation point in the subplane. (n=1, 2, 3) k), a frame of point cloud data has k lowest elevation points, the average elevation of the k lowest elevation points is calculated according to formula (1). , This is considered the lowest point within the point cloud dataset;

[0061] (1)

[0062] Set threshold parameters Select point cloud datasets with heights within the range of With threshold The points in the middle of the sum are used as the seed point set. In this embodiment, .

[0063] Step 2: Plane Fitting: Calculate the Seed Point Set The covariance matrix is ​​obtained by using singular value decomposition to find the eigenvectors corresponding to the minimum singular values ​​as the plane normal vector n(a,b,c), which is then substituted into the seed point set. The mean is used to obtain the initial planar model:

[0064] (2)

[0065] The third step is point cloud classification: This involves calculating the coordinates of each point in the point cloud dataset. , , Distance to the orthogonal projection of the plane ,like If a point is marked as ground, it is considered a non-ground point; otherwise, it is considered a non-ground point. This completes one segmentation step using the ground plane fitting algorithm.

[0066] The fourth step is iterative optimization, where the current ground point cloud is used as the seed point set input for the next iteration and then... iteration Repeat the process 3 to 5 times until the ground point cloud can be well segmented.

[0067] Step (3.4), statistical filtering for noise reduction: To reduce the impact of noise, statistical filtering is used to process the input point cloud. Statistical analysis is performed on the neighborhood of each point. After statistical filtering, the influence of outliers is effectively removed, making the point cloud of the fruit tree rows on both sides more concentrated, which can effectively reduce the difficulty of extracting the fitted straight line of the fruit tree rows on both sides.

[0068] Step (3.5), Two-dimensional projection of point cloud: Before fitting the straight line of the fruit tree row, the point cloud needs to be projected into a two-dimensional point cloud. The planar model is established using formula (2), where a=b=d=0 and c=1, that is, to The plane is projected into a two-dimensional point cloud image, forming a two-dimensional point cloud dataset, which prepares for subsequent line fitting based on the two-dimensional image.

[0069] Step (3.6), obtain the navigation trajectory using the random sampling consistency algorithm: the obtained two-dimensional point cloud dataset is then processed by... The midpoint line divides the points into two clusters on the left and right sides. The left side shows a two-dimensional point cloud dataset. The dataset is a two-dimensional point cloud on the right. The random sampling consensus algorithm is used to fit the projected point clouds of the fruit tree rows on both sides to a straight line, and the best straight line that best represents the spatial distribution of the point cloud of the fruit tree rows on that side is fitted.

[0070] like Figure 4 As shown, the point cloud lines on the left and right sides are fitted using the random sampling consensus algorithm. The expression is shown in (3). The navigation path is the midline of the two fitted lines, and the final navigation line is obtained. The trajectory expression is shown in (4).

[0071] (3)

[0072] (4)

[0073] In formula (3) and These represent the left navigation lines respectively. of Coordinates, right navigation line of Coordinate values and These represent the left navigation lines respectively. Right navigation line The slope, and These represent the left navigation lines respectively. of Coordinates, right navigation line of Coordinate values and These represent the left navigation lines respectively. Right navigation line The intercept coefficient.

[0074] Step (4) fit the navigation line Treating it as a global path, a local path planning algorithm is added to the system of the industrial control computer 11 to enable the path to be found even when the straight line is briefly lost. Navigation can continue even during this time. The industrial control computer 11 calculates the desired linear velocity using the path tracking control algorithm. and expected angular velocity Equation (5) is used to convert the linear velocity of the left and right tracks. and ,

[0075] (5)

[0076] The desired linear velocity is expressed in m / s. The desired angular velocity is expressed in rad / s. and These are the linear velocities of the left and right tracks, respectively, in m / s. This refers to the virtual wheel spacing, measured in meters (m). .

[0077] The obtained linear velocities of the left and right tracks and After conversion, the target speed of the two motors 8 can be obtained. The target speed command is encapsulated into a CAN message and sent to the motor driver 13 via the CAN bus. The motor driver 13 parses the CAN command and drives the two motors 8 to run, realizing the motion control of the orchard electric tracked chassis.

[0078] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent methods or modifications that do not depart from the technology of the present invention should be included within the scope of protection of the present invention.

Claims

1. An electric tracked chassis for orchards, characterized in that, include: The vehicle frame (1), two sets of track wheel systems, reducer (7), motor (8), tensioning mechanism (9), power battery (10), control box, and sensors; The reducer (2) is installed on the left and right sides of the frame (1) in the forward driving direction; The motor (8) is installed directly behind the reducer (7) and connected to it by a shaft. The housing of the motor (8) is fixed to the housing of the reducer (7). The two sets of track wheel systems are respectively installed on both sides of the reducer (7) away from the vehicle frame. Each set of track wheel systems includes a drive wheel (2), a road wheel (3), a tension wheel (4), a trailing wheel (5), and a track (6). Among them, the drive wheel (2) is arranged in front of the side of the vehicle frame (1) and adopts front drive. Six pairs of road wheels (3) are evenly arranged along the length of the vehicle frame (1). The trailing wheel (5) is arranged in the upper rear position of the side of the vehicle frame (1) and cooperates with the tension wheel (4) and the tensioning mechanism (9). The power input end of the drive wheel (2) is connected to the output end of the reducer (7), and the input end of the reducer (7) is connected to the power output end of the corresponding motor (8). The power battery (10) is fixed by a slot and installed in the middle of the rear side of the frame (1); The sensors include: an inertial navigation sensor (14) and a three-dimensional lidar (15); the three-dimensional lidar (15) is used to collect surrounding point cloud data, and the inertial navigation sensor (14) is used to measure the three-axis acceleration and three-axis angular velocity of the chassis to correct the data error of the three-dimensional lidar (15); The control box is installed in the middle of the chassis. The control box contains an industrial control computer (11), a programmable logic controller (12), and a motor driver (13). The industrial control computer (11) is used to process the point cloud data collected by the three-dimensional laser radar (15) and generate a navigation path. It calculates the desired linear velocity and desired angular velocity and converts them into track speed information, which is then transmitted to the programmable logic controller (12). The programmable logic controller (12) is used to convert the speed information command issued by the industrial control computer (11) into a CAN message and send it to the motor driver (13). The motor driver (13) outputs the corresponding torque, which is amplified by the reducer (7) and transmitted to the drive wheel (2), thereby driving the entire track wheel system to rotate and driving the orchard electric track chassis to move forward along the navigation path.

2. The orchard electric tracked chassis according to claim 1, characterized in that, The three-dimensional lidar (15) is fixed on the three-dimensional lidar mounting bracket (16), which is installed on the top front of the orchard electric track chassis shell (17). The upper surface of the three-dimensional lidar mounting bracket (16) is inclined at an angle θ0 to the ground.

3. The orchard electric tracked chassis according to claim 1, characterized in that, The motor (8) is a DC brushless permanent magnet motor.

4. The orchard electric tracked chassis according to claim 1, characterized in that, The track (6) is made of high-strength wear-resistant polyurethane composite material.

5. A navigation method for an orchard electric tracked chassis according to any one of claims 1-4, characterized in that, Includes the following steps: Step 1: Collect raw point cloud data between rows of orchard in real time using a three-dimensional lidar (15) and transmit it to an industrial control computer; Step 2: The industrial control computer transforms the point cloud data in the 3D lidar coordinate system to the world coordinate system, including determining the radar tilt angle θ0, establishing the radar coordinate system and the world coordinate system, and performing point cloud coordinate transformation through the coordinate transformation matrix; Step 3: Preprocess the converted point cloud data, including pass-through filtering to extract the region of interest, voxel point cloud downsampling filtering, ground point cloud fitting and removal, statistical filtering for noise reduction, and point cloud 2D projection. Step 4: Project onto Two-dimensional point cloud on a plane The midpoint line of the axis is divided into left and right point clouds; the random sampling consensus algorithm is used to fit straight lines to the left and right fruit tree point clouds respectively; the midline of the two fitted lines is calculated as the global navigation path; Step 5: The industrial control computer uses the path tracking control algorithm to calculate the desired linear velocity and angular velocity, and converts them into the linear velocity of the left and right tracks, then into the target speed of the motors on both sides. The speed of the left and right track motors is then sent to the motor driver (13) via CAN message through the programmable logic controller (12), thereby controlling the motor to drive the chassis to move along the navigation path.

6. The navigation method according to claim 5, characterized in that, Step 2 is implemented as follows: Step 2.1, determine the tilt angle of the three-dimensional lidar (15): define the tilt angle of the three-dimensional lidar as the radar coordinate system. The angle between the axis and the direction perpendicular to the ground is specifically... ; Step 2.2, establish the three-dimensional lidar coordinate system: the origin O of this three-dimensional lidar coordinate system is the transmission center of the three-dimensional lidar (15); the three-dimensional lidar coordinate system The axis is perpendicular to the scanning plane of the 3D lidar (15), and its direction is vertically upward; 3D lidar coordinate system The axis is kept parallel to the scanning plane of the three-dimensional lidar (15), and The axis points to the starting position of the 3D lidar (15) scanning circle, i.e., the 0° direction; the 3D lidar coordinate system The axis is determined by the right-hand screw rule; Step 2.3, Establish the world coordinate system: Set the origin of the world coordinate system. The point where the origin O of the 3D lidar coordinate system is perpendicularly downwards to the ground is the intersection point; the horizontal component of the operating velocity vector of the orchard electric tracked chassis is the world coordinate system. The direction of the axis; world coordinate system The axis is a vertical axis pointing upwards from the ground; world coordinate system. The axis is determined by the left-handed spiral rule; Step 2.4, Point Cloud Coordinate Transformation: Traverse each point in the point cloud data and use a coordinate transformation matrix to transform the point cloud data from the 3D LiDAR coordinate system to the world coordinate system. The point cloud coordinate transformation formula is as follows: In the formula, The tilt angle of the 3D lidar; , , x, y, z are the point cloud coordinates in the 3D lidar coordinate system; x, y, z are the point cloud coordinates in the world coordinate system.

7. The navigation method according to claim 6, characterized in that, Step 3 is implemented as follows: Step 3.1, through filtering to extract the region of interest: Since near-endpoint cloud data is dense but too cluttered, which is not conducive to subsequent line fitting, it is necessary to extract a suitable region of interest from the dataset. Specifically: set... The threshold parameters in the axial direction are (2d, 22d). The threshold parameter in the axial direction is (-2D, 2D). The threshold parameters in the axial direction are set to (-0.5H, 2.5H). The values ​​of the parameter variables d, D, and H are modified according to different orchard environments. Three-dimensional LiDAR point cloud data that are far away from the fruit tree row passage and too close or far away are cropped. Step 3.2, voxel downsampling filtering: To improve the real-time performance of subsequent point cloud processing, voxel mesh downsampling processing is performed on the point cloud in the region of interest. The voxel size is set to (0.04, 0.04, 0.04), which is a cube with a side length of 0.04m. This can greatly reduce the number of points while preserving the original point cloud features. Step 3.3, Ground point cloud fitting and removal: Ground point cloud is segmented and removed using a ground plane fitting algorithm; Step 3.4, Statistical filtering for noise reduction: To reduce the impact of noise, statistical filtering is used to process the input point cloud. Statistical analysis is performed on the neighborhood of each point. After statistical filtering, the influence of outliers is effectively removed, making the point cloud of the fruit tree rows on both sides more concentrated and reducing the difficulty of extracting the fitted straight line of the fruit tree rows on both sides. Step 3.5, Two-dimensional projection of point cloud: Before fitting the straight line of the fruit tree row, the point cloud needs to be projected into a two-dimensional point cloud. The planar model is established using formula (2), where a=b=d=0, c=1, that is, to The plane is projected into a two-dimensional point cloud image, forming a two-dimensional point cloud dataset, which prepares for subsequent line fitting based on the two-dimensional image.

8. The navigation method according to claim 7, characterized in that, The specific implementation of step 3.3 is as follows: The first step is to initialize the seed point set: The point cloud dataset after voxel point cloud downsampling is then processed along... The positive direction divides the space into k equal sub-planes, and the point cloud sub-planes are numbered sequentially as follows: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 ... k, arranging the point cloud in each subplane according to Sorting along the axis, then selecting the lowest elevation point in the subplane. n=1, 2, 3 k, a frame of point cloud data has k lowest elevation points, the average elevation of the k lowest elevation points is calculated according to formula (1). , This is considered the lowest point within the point cloud dataset; （1） Set threshold parameters Select point cloud datasets with heights within the range of With threshold The points in the middle of the sum are used as the seed point set. ; The second step, plane fitting: calculating the seed point set. The covariance matrix is ​​obtained by using singular value decomposition to find the eigenvectors corresponding to the minimum singular values ​​as the plane normal vector n(a,b,c), which is then substituted into the seed point set. The mean is used to obtain the initial planar model: （2） The third step is point cloud classification: This involves calculating the distance from the orthogonal projection of each point in the point cloud dataset to the plane. ,like If a point is marked as ground, it is considered a non-ground point. This completes one ground plane fitting algorithm segmentation. The fourth step is iterative optimization: At this point, the current ground point cloud is taken as the seed point set input for the next iteration and iterated until the ground point cloud can be segmented.

9. The navigation method according to claim 7 or 8, characterized in that, Step 4 is implemented as follows: The obtained two-dimensional point cloud dataset is obtained by The midpoint line divides the points into two clusters on the left and right sides. The left side shows a two-dimensional point cloud dataset. The right side is a two-dimensional point cloud dataset; the random sampling consensus algorithm is used to fit the projected point clouds of the fruit tree rows on both sides to a straight line, and the best straight line that best represents the spatial distribution of the point cloud of the fruit tree rows on that side is fitted respectively. The point cloud lines on the left and right sides were fitted using the random sampling consistency algorithm. The expression is shown in (3). The navigation path is the midline of the two fitted lines, and the final navigation line is obtained. The trajectory expression is shown in (4). （3） （4） In formula (3) and These represent the left navigation lines respectively. of Coordinates, right navigation line of Coordinate values and These represent the left navigation lines respectively. The slope, and These represent the left navigation lines respectively. of Coordinates, right navigation line of Coordinate values and These represent the left navigation lines respectively. The intercept coefficient.

10. The navigation method according to claim 9, characterized in that, The conversion of the desired linear velocity and angular velocity into the linear velocities of the left and right tracks in step 5 is implemented as follows: （5） The desired linear velocity; The desired angular velocity; and These are the linear velocities of the left and right tracks, respectively. This represents the virtual wheel spacing.

Images