A walking control method and control system for a pineapple harvesting robot

Abstract

The application discloses a walking control method and control system of a pineapple picking robot, and relates to the technical field of agricultural equipment control. The method comprises the following steps: obtaining plant spatial distribution data of a pineapple field to be picked, determining plant positions and regional densities, and planning movable regions based on the densities. Target pineapple feature information is obtained, and it is determined whether the target pineapple meets picking standards. If the target pineapple meets the picking standards, the target pineapple is marked as a picking target. A picking field is determined according to a maximum picking radius of the robot, and it is determined whether the target is in the range. If the target is not in the range, the spatial positions of the target and the robot are obtained, an elevation grid map is generated by laser radar scanning, terrain features are represented by a multi-dimensional feature vector, and terrain undulation is defined. An undulation judgment threshold is set, and mechanical suspension stroke and track width are dynamically adjusted according to a comparison result, so that the picking target can be smoothly reached for picking. The method improves picking efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of agricultural equipment control technology, specifically to a walking control method and control system for a pineapple harvesting robot. Background Technology

[0002] In modern agricultural automation, mechanized harvesting has become a key technology for increasing crop yields and reducing labor costs. Pineapples, as a tropical fruit, grow in complex environments with dense plant distribution and varied terrain, posing numerous challenges to the harvesting process. Traditional manual harvesting is not only inefficient but also struggles to achieve accurate identification and standardized harvesting practices.

[0003] To improve the efficiency and quality of pineapple harvesting, researchers have gradually applied advanced technologies such as visual recognition and LiDAR to pineapple harvesting robots. These technologies help robots identify target pineapples and perform accurate path planning. However, existing technologies still have many shortcomings in handling pineapple harvesting. For example, traditional path planning methods often fail to effectively consider the undulating characteristics of the terrain, causing the robot to be unable to adapt to changing ground conditions, thus affecting harvesting efficiency and safety. Therefore, how to utilize the spatial distribution data of pineapple plants, combined with terrain undulation characteristics, to achieve efficient and stable walking control of the robot has become an urgent technical problem to be solved.

[0004] In the prior art, CN108575283A discloses a control method and control system for a pineapple harvesting robot. Specifically, the robot autonomously plans its path, identifies the pineapple using image recognition and obtains its harvesting coordinates, analyzes these coordinates, and adjusts the horizontal and vertical positions of the cutting module to move it to the stem of the pineapple to be harvested, facilitating the harvesting and cutting action. After completing the harvesting task, the robot returns to its docking point. This method fully utilizes the unique growth pattern of the pineapple to automate harvesting, offering advantages such as accurate target positioning and simple, convenient harvesting. However, while this control method can autonomously plan its path, it may not effectively handle complex terrain and spatial variations. In complex terrain conditions, the control system may not adequately consider ground undulations and obstacles, leading to problems such as the robot getting stuck or unable to move smoothly during actual movement. Therefore, the stability and effectiveness of the control system are reduced.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a walking control method and control system for a pineapple harvesting robot, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A walking control method for a pineapple harvesting robot, comprising the following steps:

[0009] S1: Obtain spatial distribution data of pineapple plants in the pineapple field to be collected, determine the spatial distribution location of pineapple plants in the pineapple field to be collected based on the spatial distribution data of pineapple plants, determine the spatial distribution density of each area in the entire pineapple field to be collected, and plan the movable area using the density-oriented method.

[0010] S2: Obtain the feature information of the target pineapple, and determine whether the current target pineapple meets the picking identification criteria based on the feature information of the target pineapple. If it meets the criteria, mark the target pineapple as the picking target and execute step S3. Otherwise, change the target pineapple and return to execute step S2.

[0011] S3: Obtain the maximum picking radius of the pineapple picking robot picking device, determine the picking area range based on the maximum picking radius, and detect whether the current picking target is within the picking area range. If so, pick the current target; otherwise, proceed to step S4.

[0012] S4: Obtain the spatial location of the current picking target and the location of the pineapple picking robot. Based on the spatial location information and the movable area, plan the target walking path. Scan the terrain data of the target walking path of the pineapple picking robot with LiDAR to generate an elevation grid map. Use multi-dimensional feature vectors to characterize the terrain feature data of the target walking path area and define the terrain undulation of the target walking path based on the multi-dimensional feature vectors.

[0013] S5: Set a terrain undulation judgment threshold. Compare the terrain undulation of the target walking path area with the terrain undulation judgment threshold. If it exceeds the terrain undulation judgment threshold, dynamically adjust the suspension stroke and track width of the pineapple picking robot until it walks to the current picking target location to pick the pineapple.

[0014] Furthermore, spatial distribution data of pineapple plants in the pineapple field to be collected is obtained. Specifically, spatial distribution data refers to point cloud data of pineapple plants. The logic for determining the spatial distribution location of pineapple plants in the pineapple field to be collected based on the spatial distribution data of pineapple plants is as follows: three-dimensional point cloud data of pineapple plants is obtained through laser scanning. A spatial coordinate system is established in the pineapple field to be collected, with the geometric center of the pineapple field to be collected as the origin, the east direction as the positive x-axis, the north direction as the positive y-axis, and the z-axis direction determined by the right-hand rule. In the established coordinate system, the coordinates of the pineapple plant point cloud data are recorded. The coordinates of the pineapple plant point cloud data are filtered, segmented, and classified to identify the spatial distribution location of pineapple plants.

[0015] The spatial distribution density of each area within the entire pineapple field to be collected is determined, and the logic for using density-guided path planning to plan movable areas is as follows: Based on the distribution density of pineapple plant point cloud data, clustering algorithms are used to cluster the processed point cloud data, identify and mark the clusters of pineapple plants, and select the spatially continuous areas within the pineapple field to be collected that are not marked as pineapple plant clusters as selectable areas.

[0016] The selectable area is divided into several grid sub-regions. Hydrological feature data on the grid sub-regions are detected. Based on the hydrological feature data, waterlogged areas in the selectable area are identified. This process is repeated until all grid sub-regions are traversed. Grid sub-regions identified as waterlogged areas are marked in the selectable area. The remaining unmarked grid sub-regions are then considered movable areas. The hydrological feature data includes the surface water area and the maximum water depth. The specific logic for identifying waterlogged areas is as follows: a waterlogging judgment threshold is set, and the hydrological feature data is compared with the waterlogging judgment threshold. Grid sub-regions that exceed the waterlogging judgment threshold are marked as waterlogged areas.

[0017] Further, the characteristic information of the target pineapple is obtained, including the pineapple outline roundness, average surface roughness, pineapple outline area, and the proportion of yellow-green area of ​​the peel;

[0018] The specific method for obtaining the feature information of the target pineapple is as follows: collect several pineapple plant images with known feature information as sample images, mark the pineapple fruits in the sample images with the minimum bounding rectangle to form artificial markers, and annotate the known feature information at the corresponding markers to generate feature data images. Map the feature data images one by one with the original sample images to form a training sample dataset.

[0019] Based on the data in the training sample dataset, a neural network model is established. The original sample images in the training sample dataset are used as input, and the corresponding feature data images are used as labels to train the neural network model. The input of the trained neural network model is the image of the pineapple plant to be detected, and the output is an image that identifies the pineapple target in the image and represents the feature information.

[0020] Furthermore, based on the characteristic information of the target pineapple, it is determined whether the current target pineapple meets the picking and identification criteria. The specific formula used to determine the constraints of the picking and identification criteria is as follows:

[0021]

[0022] In the formula, R U R represents the roundness of the pineapple outline of the target pineapple. yz Ra is the outline roundness threshold, and Ra is the average surface roughness of the target pineapple. yz S is the average roughness threshold. E Let S be the area of ​​the pineapple outline of the target pineapple. yz BL represents the threshold area of ​​the pineapple outline, where BL is the percentage of yellow-green area on the target pineapple's peel. min BL is the minimum percentage threshold. max The maximum percentage threshold;

[0023] For a target pineapple that meets the constraints, it is recorded as the harvest target.

[0024] Furthermore, the maximum harvesting radius of the pineapple harvesting robot is obtained, specifically referring to the maximum horizontal and vertical movement distance of the robot. A rectangular harvesting area is formed based on the maximum horizontal and vertical harvesting radii. Based on this area, it is determined whether the current harvesting target is within the harvesting area. The specific logic for this detection is as follows:

[0025] The robot detects the connection between the root and stem of the fruit to be picked, and generates a set of detection points in the established spatial coordinate system. It then determines whether the area formed by the detection point cloud of the current picking target is entirely within the picking area. If so, the current picking target is picked. If the picking area does not fully cover the area, the robot is controlled to move.

[0026] Furthermore, a topographic grid map is generated by scanning the terrain data of the target walking path ahead of the pineapple picking robot using LiDAR. The terrain data includes: maximum elevation difference, average slope per unit area, and average surface roughness. The pineapple picking robot scans the target walking path within 5 meters ahead in real time to acquire terrain data. The formula used to calculate the average slope per unit area is:

[0027]

[0028] In the formula, θ mean θ is the unit average slope. m This represents the slope of the center point of the m-th elevation grid cell in the target walking path within 5m in front of the pineapple picking robot, M is the total number of elevation grid cells in the target walking path within 5m in front of the pineapple picking robot, and m is the index of the elevation grid cell in the target walking path within 5m in front of the pineapple picking robot, m∈[1,2,…,M];

[0029]

[0030] In the formula, h(i,j) m Let (i,j) be the elevation of the center point (i,j) within the m-th elevation grid cell along the target walking path within 5m in front of the pineapple harvesting robot. This represents the elevation gradient along the x-direction. This represents the elevation gradient along the y-direction;

[0031] The terrain relief of the target walking path is defined based on a multi-dimensional feature vector, and the formula used to calculate the terrain relief is as follows:

[0032]

[0033] In the formula, ECI represents the terrain undulation of the target walking path within 5m in front of the pineapple picking robot, and H... max Ra is the maximum elevation difference along the target walking path within 5 meters in front of the pineapple picking robot. d Let ω1, ω2, and ω3 be the average surface roughness of the target walking path within 5m in front of the pineapple picking robot, and let ω1, ω2, and ω3 be the weighting coefficients of the unit average slope, maximum elevation difference, and average surface roughness, respectively, where ω2 ≥ ω1 > ω3, and ω1, ω2, and ω3 are all greater than 0.

[0034] Furthermore, the specific logic underlying the dynamic adjustment of the suspension travel and track width of the pineapple harvesting robot is as follows: The suspension travel and track width are adjusted based on the terrain undulation. The formula for dynamically correcting the suspension travel based on the terrain undulation is as follows:

[0035]

[0036] In the formula, SR represents the suspension stroke of the pineapple harvesting robot after dynamic adjustment, SR0 is the initial value of the suspension stroke, and α is the sensitivity coefficient of the suspension stroke to changes in terrain.

[0037] The formula used to dynamically adjust the track width based on terrain undulation is as follows:

[0038] SK = SK0 * [1 + β * ECI 2 / 3 ]

[0039] In the formula, SK is the track width of the pineapple picking robot after dynamic adjustment, SK0 is the initial value of the track width, and β is the sensitivity coefficient of the track width to terrain changes.

[0040] This invention also provides a walking control system for a pineapple harvesting robot, wherein the walking control system is used to execute the aforementioned walking control method for a pineapple harvesting robot, comprising:

[0041] The mobile path planning module is used to acquire spatial distribution data of pineapple plants in the pineapple field to be collected, determine the spatial distribution location of pineapple plants in the pineapple field to be collected based on the spatial distribution data of pineapple plants, determine the spatial distribution density of each area in the entire pineapple field to be collected, and plan the movable area using the density-oriented method.

[0042] The picking target determination module is used to obtain the feature information of the target pineapple, and determine whether the current target pineapple meets the picking identification criteria based on the feature information of the target pineapple. If it meets the criteria, the target pineapple is marked as the picking target and step S3 is executed; otherwise, the target pineapple is replaced and the process returns to step S2.

[0043] The picking range capture module is used to obtain the maximum picking radius of the pineapple picking robot picking device, determine the picking area range based on the maximum picking radius, detect whether the current picking target is within the picking area range, if so, pick the current target; otherwise, execute step S4.

[0044] The path terrain detection module is used to obtain the spatial location of the current picking target and the location of the pineapple picking robot. Based on the spatial location information and the movable area, the target walking path is planned. The terrain data of the target walking path of the pineapple picking robot is scanned by LiDAR to generate an elevation grid map. Multidimensional feature vectors are used to characterize the terrain feature data of the target walking path area, and the terrain undulation of the target walking path is defined based on the multidimensional feature vectors.

[0045] The walking dynamic adjustment module is used to set the terrain undulation judgment threshold. It compares the terrain undulation of the target walking path area with the terrain undulation judgment threshold. If it exceeds the terrain undulation judgment threshold, it dynamically adjusts the suspension stroke and track width of the pineapple picking robot until it walks to the current picking target position and picks the pineapple.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] This method first obtains spatial distribution data of pineapple plants, determines walkable paths based on the pineapple distribution data, and eliminates abnormal road sections by using ground hydrological characteristic parameters, thereby optimizing the walking path, reducing the occurrence of abnormal situations, and thus significantly improving harvesting efficiency.

[0048] Secondly, by generating an elevation grid map through real-time scanning of terrain data and analyzing terrain undulation based on multi-dimensional feature vectors, a rapid response to terrain changes can be achieved. When the robot enters areas with significant terrain undulations, the suspension travel and track width are dynamically adjusted, improving the robot's stability and maneuverability in complex terrain. This effectively reduces the risk of equipment damage caused by terrain changes and improves harvesting safety. It enables the robot to operate autonomously in a wide range of terrain conditions, improving work efficiency and harvesting quality. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0050] Figure 2 A scatter plot showing the spatial distribution of pineapple plants;

[0051] Figure 3 A simulated bar chart of the spatial distribution of pineapple plants;

[0052] Figure 4 This is a schematic diagram showing the adjustment range of suspension travel and track width;

[0053] Figure 5 This is a curve fitting the terrain undulation versus suspension travel.

[0054] Figure 6 This is a curve fitting the terrain relief versus track width.

[0055] Figure 7 This represents the decrease in the average vibration acceleration after adjustment.

[0056] Figure 8 This is a schematic diagram of the overall system structure of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0058] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0059] Example:

[0060] Please see Figures 1-7 The present invention provides a technical solution:

[0061] A walking control method for a pineapple harvesting robot, comprising the following steps:

[0062] Step 1: Obtain spatial distribution data of pineapple plants in the pineapple field to be collected. Based on the spatial distribution data of pineapple plants, determine the spatial distribution location of pineapple plants in the pineapple field to be collected, determine the spatial distribution density of each area in the entire pineapple field to be collected, and use the density-guided method to plan the movable area.

[0063] The spatial distribution data of pineapple plants in the pineapple field to be collected is obtained. Specifically, the spatial distribution data refers to the point cloud data of pineapple plants. The logic for determining the spatial distribution location of pineapple plants in the pineapple field based on the spatial distribution data of pineapple plants is as follows: three-dimensional point cloud data of pineapple plants are obtained by laser scanning. A spatial coordinate system is established in the pineapple field to be collected, with the geometric center of the pineapple field as the origin, the positive x-axis as due east, the positive y-axis as due north, and the z-axis direction determined by the right-hand rule. In the established coordinate system, the coordinates of the pineapple plant point cloud data are recorded. The coordinates of the pineapple plant point cloud data are filtered, segmented, and classified to identify the spatial distribution location of pineapple plants.

[0064] The spatial distribution density of each area within the entire pineapple field to be collected is determined, and the logic for using density-guided path planning to plan movable areas is as follows: Based on the distribution density of pineapple plant point cloud data, clustering algorithms are used to cluster the processed point cloud data, identify and mark the clusters of pineapple plants, and select the spatially continuous areas within the pineapple field to be collected that are not marked as pineapple plant clusters as selectable areas.

[0065] First, point cloud data of the pineapple field is collected using LiDAR, drones, or other sensors. The collected point cloud data contains the three-dimensional coordinate information of each point. Before clustering, the point cloud data needs to be preprocessed for noise removal: filters, such as statistical outlier removal methods, are used to remove noise and unnecessary points. If the point cloud data volume is large, methods such as Voxel Grid filters can be used for downsampling to reduce the amount of computation. A suitable clustering algorithm is selected to process the point cloud data. Commonly used clustering algorithms include: the DBSCAN algorithm, which selects an appropriate neighborhood radius and a minimum number of points. The advantage of DBSCAN is that it can identify clusters of arbitrary shapes and handle noise. Implementation steps: For each point, find the number of points in its neighborhood. If the number of points in the neighborhood is greater than or equal to the minimum number of points, mark it as a core point and form a cluster. Recursively add core points in the neighborhood to the cluster. Mark points not included in any cluster as noise. Alternatively, the MeanShift algorithm can be used, which calculates the mean position of the data points and continuously updates the cluster centers until convergence is reached. It is suitable for cases where the cluster shape is irregular.

[0066] From the clustering results, spatially continuous regions that are not marked as pineapple plant clusters are identified. Through adjacency matrix or spatial graph analysis, it is checked whether the unmarked points form continuous regions. Regions that meet the conditions are marked as selectable regions to facilitate subsequent path planning.

[0067] The selectable area is divided into several grid sub-regions. Hydrological feature data on the grid sub-regions are detected. Based on the hydrological feature data, waterlogged areas in the selectable area are identified until all grid sub-regions are traversed. Grid sub-regions identified as waterlogged areas are marked in the selectable area. The remaining unmarked grid sub-regions are used as movable areas. The hydrological feature data includes the surface water area and the maximum water depth. The specific logic for identifying waterlogged areas is as follows: a waterlogging judgment threshold is set, the hydrological feature data is compared with the waterlogging judgment threshold, and grid sub-regions that exceed the waterlogging judgment threshold are marked as waterlogged areas.

[0068] The area of ​​water accumulation on the road surface can be obtained using image data. Image processing techniques, such as threshold segmentation, can be used to identify waterlogged areas and convert the image into a grayscale image. Adaptive thresholding or the Otsu's method is then applied for binarization to distinguish between waterlogged and non-waterlogged areas. Connectivity analysis is used to locate the waterlogged areas and calculate their area.

[0069] The maximum water depth in the waterlogged area can be directly read from the real-time data of the water level sensor.

[0070] A waterlogging detection threshold is set, and hydrological characteristic data is compared with the threshold. Grid sub-regions exceeding the threshold are marked as waterlogged areas. The specific logic for determining waterlogged areas is as follows: an impassable coefficient is calculated based on the hydrological characteristic data, and this coefficient is compared with the waterlogging detection threshold. The waterlogged area is determined based on the comparison result. The formula used to calculate the impassable coefficient is as follows:

[0071]

[0072] In the formula, SCI is the inaccessibility factor, and MS is the inaccessibility factor. m For the surface area of ​​the road and the water area, L max This represents the maximum water depth.

[0073] It should be noted that the higher the impassable coefficient (SCI) value, the more water is accumulated on the road surface in the current area, the greater the depth of the water, and the less suitable it is for the pineapple picking robot to move around.

[0074] Among them, the road surface area and water area MS m This reflects the extent of water accumulation and directly affects the traffic capacity of the area. A larger water accumulation area means a larger potentially affected area and a higher risk to traffic. Therefore, the area of ​​water accumulation on the road (MS) is considered a significant indicator. m Proportional to the inaccessibility coefficient SCI, through the square root function The impact on water surface area is non-linear, meaning that the growth rate of SCI slows down as the water surface area increases. This is because, in reality, after the area increases to a certain extent, the marginal effect of the impact decreases, that is, the impact of further increases in area on traffic capacity is relatively reduced. The square root can prevent the SCI value from increasing abnormally when the water surface area is extremely large, thus maintaining the rationality of the judgment.

[0075] Maximum water depth L max The depth of the water was quantified; the greater the depth, the more significant the impact on passage. Deep water may prevent the pineapple harvesting robot from passing and even pose a threat to its safety. This was determined using the logarithmic function ln(1+L). maxIt is also a non-linear function that can effectively reflect the impact of depth on traffic capacity. When the water depth is small, the impact is relatively small, but when the depth increases to a certain extent, the effect will be significantly enhanced.

[0076] The specific logic used to determine the specific waterlogged area based on the comparison results is as follows:

[0077] When 0 ≤ SCI < 1.0 * yz, the grid sub-region is considered passable;

[0078] When SCI≥1.0*yz, the grid sub-region is determined to be a waterlogged area, indicating that the grid sub-region is impassable;

[0079] Where yz is the threshold for judging water accumulation. The specific threshold can be determined based on the track support height of the pineapple harvesting robot combined with expert experience.

[0080] Step 2: Obtain the feature information of the target pineapple. Based on the feature information of the target pineapple, determine whether the current target pineapple meets the picking identification criteria. If it does, mark the target pineapple as the picking target and execute step S3. Otherwise, change the target pineapple and return to execute step S2.

[0081] The characteristic information of the target pineapple is obtained, including the pineapple's outline roundness, average surface roughness, pineapple outline area, and the proportion of yellow-green area in the peel.

[0082] The specific method for obtaining the feature information of the target pineapple is as follows: collect several pineapple plant images with known feature information as sample images, mark the pineapple fruits in the sample images with the minimum bounding rectangle to form artificial markers, and annotate the known feature information at the corresponding markers to generate feature data images. Map the feature data images one by one with the original sample images to form a training sample dataset.

[0083] Based on the data in the training sample dataset, a neural network model is established. The original sample images in the training sample dataset are used as input, and the corresponding feature data images are used as labels to train the neural network model. The input of the trained neural network model is the image of the pineapple plant to be detected, and the output is an image that identifies the pineapple target in the image and represents the feature information.

[0084] The neural network model is specifically a Long Short-Term Memory (LSTM) network model. An activation function and optimization algorithm are selected, with the Tanh function chosen as the activation function and Adam as the optimization algorithm for the LSTM model. The formula for the Tanh function is:

[0085]

[0086] In the formula, f(x) represents the Tanh function, and the independent variable r represents the weighted sum of the neuron's inputs, that is, the result of the weighted sum of the inputs received by the neuron from the previous layer.

[0087] Simultaneously, the hyperparameters of the LSTM model are set, including: number of network layers, number of iterations, learning rate, batch size, number of training iterations, number of batches, and number of hidden layer neurons.

[0088] The network is set to a 3-layer structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training iterations is set to 100, the batch size is set to 256, and the number of hidden layer neurons is set to 32.

[0089] Based on the characteristic information of the target pineapple, it is determined whether the current target pineapple meets the picking and identification criteria. The specific formula used to determine the constraints of the picking and identification criteria is as follows:

[0090]

[0091] In the formula, R U R represents the roundness of the pineapple outline of the target pineapple. yz Ra is the outline roundness threshold, and Ra is the average surface roughness of the target pineapple. yz S is the average roughness threshold. E Let S be the area of ​​the pineapple outline of the target pineapple. yz BL represents the threshold area of ​​the pineapple outline, where BL is the percentage of yellow-green area on the target pineapple's peel. min BL is the minimum percentage threshold. max The maximum percentage threshold;

[0092] For target pineapples that meet the constraints, they are recorded as harvest targets. The specific thresholds are set based on expert experience and the pineapple variety.

[0093] Step 3: Obtain the maximum picking radius of the pineapple picking robot's picking device, determine the picking area range based on the maximum picking radius, and detect whether the current picking target is within the picking area range. If so, pick the current picking target; otherwise, proceed to step S4.

[0094] The maximum harvesting radius of the pineapple harvesting robot is obtained, specifically referring to the maximum horizontal and vertical movement distance of the robot. A rectangular harvesting area is formed based on the maximum horizontal and vertical harvesting radii. Within this area, it is determined whether the current harvesting target is within the harvesting area. The specific logic for this determination is as follows:

[0095] The system detects the connection between the root and stem of the fruit to be picked, and generates a set of detection points in the established spatial coordinate system. It then determines whether the area formed by the detection point cloud of the current picking target is entirely within the picking area. If so, the current picking target is picked; otherwise, the pineapple picking robot is controlled to move.

[0096] When detecting the current picking target, it is necessary to pay attention to the coordinates of the connection between the pineapple's root and stem. The specific steps are as follows: Obtain the coordinates of the picking target: Obtain the coordinates of the current pineapple's root using LiDAR, set the density and distribution of the point cloud, for example, use a spherical or cubic method to generate a point cloud around the target, ensuring that the generated point cloud covers all parts of the pineapple, obtain the boundary coordinates of the generated point cloud set, determine its minimum and maximum range in the spatial coordinate system, if all points in the point cloud are within the defined picking area, then it can be determined that the current picking target is within the picking area.

[0097] Step 4: Obtain the spatial location of the current picking target and the location of the pineapple picking robot. Based on the spatial location information and the movable area, plan the target walking path. Scan the terrain data of the target walking path of the pineapple picking robot with LiDAR to generate an elevation grid map. Use multi-dimensional feature vectors to characterize the terrain feature data of the target walking path area and define the terrain undulation of the target walking path based on the multi-dimensional feature vectors.

[0098] The process involves using LiDAR, depth cameras, or other sensors to obtain the spatial coordinates of the pineapple being harvested. Specific steps include: scanning the surrounding environment with sensors to identify the pineapple's location and extracting the coordinates of its geometric center; and then using a global positioning system such as GPS or a local positioning system such as LiDAR SLAM to obtain the pineapple harvesting robot's spatial position.

[0099] The specific method for planning the target walking path based on spatial location information and movable areas includes: using the center spatial coordinates of the pineapple picking robot as the starting coordinates and the geometric center of the current picking target as the target spatial coordinates, extracting its planar coordinates, and using a path planning algorithm based on the planar coordinates to plan the path within the selectable area to obtain the target walking path. Specifically, the path planning algorithm can be either the A* algorithm (suitable for pathfinding on a known map, highly efficient and able to avoid obstacles) or the Dijkstra algorithm (suitable for finding the shortest path, but relatively less efficient). Combining the current picking target location and the robot's location, the optimal path for the robot to reach the picking target is calculated, specifically using the shortest path as the optimal path. The selected path planning algorithm is then used to generate a path from the starting point to the destination, ensuring that all areas traversed by the path are movable.

[0100] The terrain data along the target walking path of the pineapple harvesting robot is scanned by LiDAR to generate an elevation grid map. The terrain data includes: maximum elevation difference, average unit slope, and average surface roughness. The pineapple harvesting robot scans the target walking path within 5 meters in real time to acquire terrain data. The formula used to calculate the average unit slope is:

[0101]

[0102] In the formula, θ mean θ is the unit average slope. m This represents the slope of the center point of the m-th elevation grid cell in the target walking path within 5m in front of the pineapple picking robot, M is the total number of elevation grid cells in the target walking path within 5m in front of the pineapple picking robot, and m is the index of the elevation grid cell in the target walking path within 5m in front of the pineapple picking robot, m∈[1,2,…,M];

[0103]

[0104] In the formula, h(i,j) m Let (i,j) be the elevation of the center point (i,j) within the m-th elevation grid cell along the target walking path within 5m in front of the pineapple harvesting robot. This represents the elevation gradient along the x-direction. This represents the elevation gradient along the y-direction;

[0105] The terrain relief of the target walking path is defined based on a multi-dimensional feature vector, and the formula used to calculate the terrain relief is as follows:

[0106]

[0107] In the formula, ECI represents the terrain undulation of the target walking path within 5m in front of the pineapple picking robot, and H... max Ra is the maximum elevation difference along the target walking path within 5 meters in front of the pineapple picking robot. d Let θ0 be the average surface roughness of the target walking path within 5m in front of the pineapple picking robot, and θ0 be the reference slope, which is usually zero and represents the slope under flat ground. ω1, ω2 and ω3 are the weighting coefficients of unit average slope, maximum elevation difference and average surface roughness, respectively, where ω2≥ω1>ω3, and ω1, ω2 and ω3 are all greater than 0.

[0108] It should be noted that the terrain undulation index (ECI) reflects the impact of terrain on robot movement. The higher the value, the more complex the terrain, and the more necessary it is to adjust the robot's parameters to ensure smooth movement.

[0109] Where the unit average slope θ meanθ represents the average slope angle of the ground. The larger the value, the steeper the slope, and the more difficult it is for the robot to move. This affects the stability and movement efficiency of the robot during walking. Therefore, the average slope θ per unit area is important. mean Proportional to the terrain relief (ECI), it reflects the impact of ground tilt on movement by calculating the difference between the average slope and the baseline slope. The denominator contains 1+θ. mean The term is used to normalize the slope value to avoid the influence of extreme values ​​caused by excessively large slopes.

[0110] Maximum elevation difference H max This represents the height difference between the highest and lowest points in the terrain. A larger value indicates a greater vertical difference between the high and low points, indicating a more uneven and undulating terrain. This significantly increases the difficulty of movement for robots on undulating terrain, especially for robots that need to overcome significant elevation changes during movement. Therefore, it is directly proportional to the terrain relief (ECI). The maximum elevation difference is expressed as the square root of the elevation difference. It is used to reflect the impact of elevation undulations on terrain complexity. Compared to a linear relationship, the square root form helps to reduce the impact when there are large elevation differences.

[0111] Average surface roughness Ra d Describing the smoothness of the ground surface, as the undulation of the terrain increases and the surface becomes rougher, the challenges faced by the robot during movement will significantly increase, potentially leading to instability, slower speed, and higher energy consumption. Therefore, both factors need to be considered in path planning and operational strategies to ensure the robot's movement efficiency and safety. The natural logarithm of the average surface roughness is ln(1+Ra). d This is used to handle the effect of roughness on motion, allowing small roughness changes to have a relatively small effect, while larger roughness changes will have a significantly increased effect.

[0112] Maximum elevation difference H max It is a key indicator of terrain undulation because it directly affects the robot's mobility. A larger elevation difference means the robot needs to overcome greater changes in gravitational potential energy while moving, thus having the greatest impact. Average slope θ mean It is also important, but generally, the slope has a smaller impact on the robot's driving stability than the elevation difference, and the average surface roughness Ra d The impact on the robot is relatively small, mainly affecting friction. Although roughness affects the efficiency and stability of movement, its impact is generally lower than that of slope and elevation difference. Therefore, we set ω2≥ω1>ω3, and ω1, ω2, and ω s All are greater than 0. At the same time, ω1, ω2 and ω3 are generally less than 1.

[0113] The specific method for acquiring terrain data is as follows: Within the scanning area, the robot continuously collects ground height information using LiDAR or a stereo camera, generating 3D point cloud data of the terrain. The acquired point cloud data is processed to extract the ground point cloud. From the processed point cloud, the highest and lowest points are extracted to calculate the maximum elevation difference. The height variations of the ground point cloud are analyzed to calculate the average roughness. Specifically, the height difference between each point and its neighboring points is calculated, and the absolute values ​​of these height differences are taken to calculate their average value as the roughness index.

[0114] Step 5: Set the terrain undulation judgment threshold. Compare the terrain undulation of the target walking path area with the terrain undulation judgment threshold. If it exceeds the terrain undulation judgment threshold, dynamically adjust the suspension stroke and track width of the pineapple picking robot until it walks to the current picking target location to pick the pineapple.

[0115] The system sets a terrain undulation judgment threshold. The terrain undulation of the target walking path area is compared with the terrain undulation judgment threshold. If the terrain undulation of the current detection section is less than or equal to the terrain undulation judgment threshold, the suspension travel and track width of the pineapple picking robot will not be adjusted, indicating that the picking work can be completed using the current suspension travel and track width.

[0116] The suspension travel and track width are adjusted based on the terrain undulation. The formula used for dynamically correcting the suspension travel based on the terrain undulation is as follows:

[0117]

[0118] In the formula, SR represents the suspension stroke of the pineapple harvesting robot after dynamic adjustment, SR0 is the initial value of the suspension stroke, and α is the sensitivity coefficient of the suspension stroke to changes in terrain.

[0119] It's important to note that the primary function of the suspension system in a robot is to absorb the impacts and vibrations from uneven ground, maintaining the robot's stability and comfort. The length of the suspension travel directly affects the robot's traversal capability and terrain adaptability. In areas with significant undulations, the robot experiences greater impacts and vibrations while navigating. Increasing the suspension travel helps the robot absorb these impacts more effectively, reducing the risk of damage to the robot's structure and improving its stability in complex terrain.

[0120] pass This reflects the impact of terrain undulation on suspension travel. As ECI increases, the complexity and undulation of the terrain also increase, thus requiring greater suspension travel to accommodate these changes. Using the square root form makes the relationship between the increase in suspension travel and terrain undulation non-linear, reflecting that the increase in suspension travel demand is more significant under higher undulation conditions. This non-linear relationship is more realistic because in more complex terrain, the additional suspension travel contributes more significantly to stability.

[0121] α is an adjustable parameter that represents the sensitivity of the suspension system to changes in terrain. By adjusting α, the suspension travel can be flexibly controlled according to different terrain characteristics and robot design requirements. Generally, α is taken between 0.01 and 0.05.

[0122] The formula used to dynamically adjust the track width based on terrain undulation is as follows:

[0123] SK = SK0 * [1 + β * ECI 2 / 3 ]

[0124] In the formula, SK is the track width of the pineapple picking robot after dynamic adjustment, SK0 is the initial value of the track width, and β is the sensitivity coefficient of the track width to terrain changes.

[0125] It's important to note that on terrain with significant undulations, robots may experience more lateral tilting and vertical vibrations while navigating. Increasing track width effectively increases the contact area with the ground, thereby improving robot stability and reducing the risk of tipping over. On uneven terrain, increased track width significantly improves traction, preventing slippage on soft or uneven surfaces. Wide tracks better adapt to terrain variations, ensuring smooth robot movement under various conditions. Therefore, the adjustment range for track width is directly proportional to the degree of terrain undulation. (Using ECI) 2 / 3 This is used to represent the relationship between the increase in track width and the required terrain undulation, ensuring that the increase in track width is more significant under higher undulation.

[0126] β is an adjustable parameter that represents the sensitivity of track width to terrain changes. By adjusting β, the track width can be flexibly controlled according to different terrain characteristics and robot design requirements. Typically, β is taken between 0.01 and 0.1.

[0127] Table 1 shows some statistics on suspension travel and track width after adjustment based on terrain undulation.

[0128] Table 1: Adjustment data for suspension travel and track width

[0129]

[0130] The data shows that as the terrain undulation changes, the adjusted suspension travel and track width are also adjusted accordingly to optimize the equipment's performance under different terrain conditions. For example, from region 1 to region 5, although the initial suspension travel and track width remain unchanged, the adjusted suspension travel and track width increase with the increase in terrain undulation. This phenomenon indicates that in areas with higher terrain undulation, the equipment requires greater suspension travel and track width to maintain stability and adaptability. Region 5 has an undulation of 3.00, and the adjusted track width is 0.26, showing that the equipment performance is effectively improved under this condition, better adapting to complex terrain. Regarding vibration acceleration, the reduction in vibration acceleration shows different trends with changes in terrain undulation. Region 3 has an undulation of 2.50, and the average reduction in vibration acceleration after adjustment reaches 0.15, indicating that the design optimization in this region effectively reduces vibration and improves operational smoothness.

[0131] Please see Figure 8 The present invention also provides a walking control system for a pineapple harvesting robot, wherein the walking control system for the pineapple harvesting robot is used to execute the aforementioned walking control method for the pineapple harvesting robot, comprising:

[0132] The mobile path planning module is used to acquire spatial distribution data of pineapple plants in the pineapple field to be collected, determine the spatial distribution location of pineapple plants in the pineapple field to be collected based on the spatial distribution data of pineapple plants, determine the spatial distribution density of each area in the entire pineapple field to be collected, and plan the movable area using the density-oriented method.

[0133] The picking target determination module is used to obtain the feature information of the target pineapple, and determine whether the current target pineapple meets the picking identification criteria based on the feature information of the target pineapple. If it meets the criteria, the target pineapple is marked as the picking target and step S3 is executed; otherwise, the target pineapple is replaced and the process returns to step S2.

[0134] The picking range capture module is used to obtain the maximum picking radius of the pineapple picking robot picking device, determine the picking area range based on the maximum picking radius, detect whether the current picking target is within the picking area range, if so, pick the current target; otherwise, execute step S4.

[0135] The path terrain detection module is used to obtain the spatial location of the current picking target and the location of the pineapple picking robot. Based on the spatial location information and the movable area, the target walking path is planned. The terrain data of the target walking path of the pineapple picking robot is scanned by LiDAR to generate an elevation grid map. Multidimensional feature vectors are used to characterize the terrain feature data of the target walking path area, and the terrain undulation of the target walking path is defined based on the multidimensional feature vectors.

[0136] The walking dynamic adjustment module is used to set the terrain undulation judgment threshold. It compares the terrain undulation of the target walking path area with the terrain undulation judgment threshold. If it exceeds the terrain undulation judgment threshold, it dynamically adjusts the suspension stroke and track width of the pineapple picking robot until it walks to the current picking target position and picks the pineapple.

[0137] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0138] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0139] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0140] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for controlling the walking of a pineapple picking robot, characterized by, The specific steps include: S1: Obtain spatial distribution data of pineapple plants in the pineapple field to be collected, determine the spatial distribution positions of the pineapple plants in the pineapple field to be collected based on the spatial distribution data of the pineapple plants, determine the spatial distribution density of each region in the entire pineapple field to be collected, and plan a movable region using a density-guided method; S2: Obtain feature information of a target pineapple, determine whether the current target pineapple meets the picking recognition standard based on the feature information of the target pineapple, mark the target pineapple as a picking target and perform step S3 if the target pineapple meets the picking recognition standard, otherwise, replace the target pineapple and return to perform step S2; S3: Obtain the maximum picking radius of a pineapple picking robot picking device, determine the picking field range based on the maximum picking radius, detect whether the current picking target is within the picking field range, and pick the current picking target if the current picking target is within the picking field range, otherwise, perform step S4; S4: Obtain the spatial position of the current picking target and the field where the pineapple picking robot is located, plan a target walking path based on the spatial position information and the movable region, scan the terrain data of the target walking path of the pineapple picking robot by a laser radar, generate an elevation grid map, use a multi-dimensional feature vector to represent the terrain feature data of the target walking path region, and define the terrain undulation of the target walking path based on the multi-dimensional feature vector; S5: Set a terrain undulation judgment threshold, compare the terrain undulation of the target walking path region with the terrain undulation judgment threshold, dynamically adjust the suspension stroke and track width of the pineapple picking robot if the terrain undulation exceeds the terrain undulation judgment threshold, and pick the current picking target after the pineapple picking robot walks to the current picking target position.

2. The walking control method of the pineapple picking robot according to claim 1, characterized in that: The spatial distribution data of pineapple plants in the pineapple field to be collected, wherein the spatial distribution data specifically refers to pineapple plant point cloud data, the logic for determining the spatial distribution positions of the pineapple plants in the pineapple field to be collected based on the spatial distribution data of the pineapple plants is as follows: three-dimensional point cloud data of the pineapple plants is obtained through laser scanning, a spatial coordinate system is established in the pineapple field to be collected, the geometric center of the pineapple field to be collected is taken as the origin, the positive east direction is taken as the positive direction of the x-axis, the positive north direction is taken as the positive direction of the y-axis, and the z-axis direction is determined through the right-hand rule, in the established coordinate system, pineapple plant point cloud data coordinates are recorded, and the pineapple plant point cloud data coordinates are filtered, segmented, and classified to identify the spatial distribution positions of the pineapple plants; The logic for planning a movable region using a density-guided path is as follows: based on the distribution density of the pineapple plant point cloud data, the processed point cloud data is clustered through a clustering algorithm, the clusters of the pineapple plants are identified and marked, and the regions in the pineapple field to be collected that are not marked as pineapple plant clusters and are spatially continuous are taken as selectable regions. The optional area is divided into a plurality of grid sub-areas, hydrological feature data on the grid sub-areas is detected, a waterlogging area in the optional area is identified according to the hydrological feature data, until all the grid sub-areas are traversed, the grid sub-areas identified as the waterlogging area are marked in the optional area, and the remaining grid sub-areas not marked are taken as movable areas, wherein the hydrological feature data includes a road water area and a maximum waterlogging depth, and the logic for judging the waterlogging area specifically comprises: setting a waterlogging judgment threshold, comparing the hydrological feature data with the waterlogging judgment threshold, and marking the grid sub-areas exceeding the waterlogging judgment threshold as the waterlogging area.

3. The walking control method of the pineapple picking robot according to claim 2, characterized in that: Obtain feature information of a target pineapple, the feature information of the target pineapple including pineapple profile circularity, surface average roughness, pineapple profile area, and peel yellow-green area proportion; The specific method for obtaining the feature information of the target pineapple comprises: collecting a plurality of pineapple plant images with known feature information as sample images, marking pineapple fruits in the sample images by a minimum circumscribed rectangle to form artificial marks, labeling the known feature information at the corresponding marks to generate feature data images, and mapping the feature data images and the original sample images one by one to form a training sample data set; Based on the data in the training sample data set, a neural network model is established, the original sample images in the training sample data set are taken as inputs, and the corresponding feature data images are taken as labels, the neural network model is trained, and the trained neural network model inputs a pineapple plant image to be detected and outputs an image in which a pineapple target is recognized and characterized by feature information.

4. The walking control method of the pineapple picking robot according to claim 3, characterized in that: Determine whether the current target pineapple meets the picking recognition standard based on the feature information of the target pineapple, and the formula for determining the constraint condition of the picking recognition standard is: In the formula, R U is the circularity of the pineapple profile of the target pineapple, R yz is the circularity threshold, Ra is the surface average roughness of the target pineapple, Ra yz is the average roughness threshold, S E is the pineapple profile area of the target pineapple, S yz is the pineapple profile area threshold, BL is the ratio of the yellow-green area of the peel of the target pineapple, BL min is the minimum ratio threshold, BL max is the maximum ratio threshold. For the target pineapple meeting the constraint condition, mark it as a picking target.

5. The walking control method of a pineapple picking robot according to claim 4, characterized in that: Obtain the maximum picking radius of the pineapple picking robot picking device, wherein the maximum picking radius specifically refers to the maximum movement distance of the pineapple picking robot picking device in the horizontal direction and the vertical direction, form a rectangular picking field range according to the maximum picking radius in the horizontal direction and the vertical direction, and determine whether the current picking target is within the picking field range based on the picking field range determined based on the maximum picking radius, wherein the logic for determining and judging specifically comprises: Detect the coordinates of the connection between the fruit root of the current picking target and the stem in the established spatial coordinate system to generate a detection point cloud set, determine whether the field formed by the detection point cloud set of the current picking target is entirely within the picking field range, and if so, pick the current picking target, and if not, control the pineapple picking robot to move.

6. The walking control method of a pineapple picking robot according to claim 5, characterized in that: Scan the terrain data of the target walking path in front of the pineapple picking robot by laser radar to generate an elevation grid map, and the terrain data includes: maximum elevation difference, unit average slope, and ground average roughness, wherein the pineapple picking robot scans the target walking path within 5m in front of the pineapple picking robot in real time to obtain the terrain data, and the formula for calculating the unit average slope is: where θ mean is the unit average slope, θ m represents the cell slope of the center point in the mth elevation grid cell in the target walking path within 5 m in front of the pineapple picking robot, M is the total number of elevation grid cells in the target walking path within 5 m in front of the pineapple picking robot, m is the index of the elevation grid cell in the target walking path within 5 m in front of the pineapple picking robot, m ∈ [1, 2, …, M]; where h(i,j) m is the elevation of the center point (i,j) in the mth elevation grid cell in the goal walking path within 5 m in front of the pineapple picking robot, denotes the elevation gradient in the x direction, denotes the elevation gradient in the y direction; The terrain undulation of the destination walking path is defined based on a multi-dimensional feature vector, wherein a formula for calculating the terrain undulation is: In the formula, ECI is the terrain roughness of the walking path of interest within 5 m in front of the pineapple picking robot, H max is the maximum elevation difference of the walking path of interest within 5 m in front of the pineapple picking robot, Ra d is the average ground roughness of the walking path of interest within 5 m in front of the pineapple picking robot, ω1, ω2 and ω3 are weight coefficients of the unit average slope, the maximum elevation difference and the average ground roughness respectively, wherein ω2≥ω1>ω3, and ω1, ω2 and ω3 are all greater than 0.

7. The walking control method of a pineapple picking robot according to claim 6, characterized in that: The logic for dynamically adjusting the suspension stroke and track width of the pineapple picking robot is based on the terrain undulation, wherein the formula for dynamically correcting the suspension stroke based on the terrain undulation is: In the formula, SR represents the dynamically adjusted suspension stroke of the pineapple picking robot, SR0 is the initial value of the suspension stroke, and a is the sensitivity coefficient of the suspension stroke to terrain changes; In the formula, SK is the dynamically adjusted track width of the pineapple picking robot, SK0 is the initial value of the track width, and β is the sensitivity coefficient of the track width to terrain changes. SK = SK0* [1 + β*ECI 2 / 3 ] The walking control system of the pineapple picking robot is used to execute the walking control method of the pineapple picking robot according to any one of claims 1-7, comprising:

8. A walking control system of a pineapple picking robot, characterized by: A moving path planning module is configured to obtain pineapple plant spatial distribution data of a pineapple field to be harvested, determine the spatial distribution positions of pineapple plants in the pineapple field to be harvested based on the pineapple plant spatial distribution data, determine the spatial distribution density of each region in the entire pineapple field to be harvested, and plan a movable region using a density-guided method; A picking target determination module is configured to obtain feature information of a target pineapple, determine whether the current target pineapple meets the picking recognition standard based on the feature information of the target pineapple, mark the target pineapple as a picking target and execute step S3 if it meets the picking recognition standard, or replace the target pineapple and return to execute step S2 if it does not meet the picking recognition standard; A picking range capture module is configured to obtain the maximum picking radius of a picking device of the pineapple picking robot, determine the picking field range based on the maximum picking radius, and detect whether the current picking target is within the picking field range, and pick the current picking target if it is within the picking field range, or execute step S4 if it is not within the picking field range; A path terrain detection module is configured to obtain the spatial position of the current picking target and the pineapple field where the pineapple picking robot is located, plan a destination walking path based on the spatial position information and the movable region, scan the terrain data of the destination walking path of the pineapple picking robot through a laser radar, generate an elevation grid map, use a multi-dimensional feature vector to represent the terrain feature data of the destination walking path region, and define the terrain undulation of the destination walking path based on the multi-dimensional feature vector; A walking dynamic adjustment module is configured to set a terrain undulation judgment threshold, compare the terrain undulation of the destination walking path region with the terrain undulation judgment threshold, and dynamically adjust the suspension stroke and track width of the pineapple picking robot if the terrain undulation exceeds the terrain undulation judgment threshold, until the walking reaches the position of the current picking target for picking. ​

Images