A control method of a picking robot

Abstract

The application discloses a control method of a picking robot, and particularly relates to the technical field of intelligent control; the method is characterized in that three-dimensional point cloud data of all fruits in a fruit cluster area is collected in real time, and a space connection topology structure of fruit stalks is established; a finite element mechanics analysis model is established based on the space connection topology structure of the fruit stalks, a stress transmission path after a picking action is applied to the fruit stalks is analyzed, and a position deviation of adjacent un-picked fruits is predicted; a multi-target path optimization algorithm is used to generate an initial path for continuous picking of the picking robot, and a space position of a next fruit picking action is dynamically corrected according to fruit stalk stress transmission path data; and actual position deviation data of the picking action is used to identify abnormal change characteristics of the position and posture of the fruits, so that control parameters and path planning of the next picking action are adjusted in real time. The method improves the accuracy and stability of picking path planning, and improves the overall efficiency and reliability of the picking operation of the robot.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and more specifically, to a control method for a harvesting robot. Background Technology

[0002] With the rapid development of agricultural automation technology, harvesting robots are widely used in fruit harvesting operations in orchards.

[0003] When existing harvesting robots continuously harvest fruit, the torque applied to the fruit stalk by the previous harvesting action can easily be transmitted to adjacent fruit stalks through the fruit stalk connection structure, resulting in fruit stalk interference. This causes adjacent unharvested fruits to shift and twist, causing the actual fruit position in the next harvesting action to deviate from the expected initial position. This leads to problems such as grasping failure or fruit damage when the harvesting robot is working continuously, affecting the stability and efficiency of the robot's harvesting operation. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a control method for a harvesting robot to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A control method for a harvesting robot includes the following steps:

[0007] S1: Real-time acquisition of 3D point cloud data of all fruits in the fruit cluster area, analysis of the spatial position relationship of fruits and the connection structure of fruit stalks, and output of spatial connection structure data of all fruit stalks in the fruit cluster.

[0008] S2: Based on the spatial connection structure data of all fruit stalks in the fruit cluster, analyze the stress transmission path after the picking action is applied to the fruit stalk of the target fruit, predict the positional offset of adjacent unpicked fruits, and output the stress transmission path data of the fruit stalk.

[0009] S3: Based on the spatial connection structure data of all fruit stalks in the fruit cluster, plan the initial path sequence for the picking robot to continuously pick fruits, and output the robot's initial picking path;

[0010] S4: Based on the stress transmission path data of the fruit stalk and the robot's initial picking path, dynamically correct the spatial position of the next fruit picking action and output the corrected robot picking path.

[0011] S5: Based on the actual position deviation data after each picking action of the robot, identify the abnormal change characteristics of the fruit position and posture, and output the abnormal change characteristic data of the fruit position and posture.

[0012] S6: Based on the abnormal change characteristic data and the corrected robot picking path, adjust the control parameters and path planning of the robot's next picking action in real time.

[0013] In a preferred embodiment, S1 specifically refers to:

[0014] Real-time acquisition of 3D point cloud data of all fruits in the fruit cluster area, followed by preprocessing to obtain preprocessed 3D point cloud data;

[0015] Based on the preprocessed 3D point cloud data, the spatial position and geometric shape of the fruit within the fruit cluster area are identified, and the spatial positional relationship between the fruits and the characteristics of the fruit stem connection structure within the fruit cluster area are analyzed.

[0016] Based on the spatial positional relationship between fruits and the characteristics of the fruit stalk connection structure within the fruit cluster area, a spatial connection topology structure between all fruit stalks within the fruit cluster area is established, and the spatial connection structure data of all fruit stalks within the fruit cluster area is output.

[0017] In a preferred embodiment, S2 specifically refers to:

[0018] Based on the spatial connection structure data of all fruit stalks within the fruit cluster area, a finite element mechanical analysis model of the fruit stalk connection structure within the fruit cluster area is established.

[0019] The transmission path of torque and stress generated by the clamping mechanism of the picking robot acting on the fruit stem when picking the target fruit is analyzed based on the finite element mechanical analysis model.

[0020] Based on the transmission path of torque and stress, predict the positional shift of adjacent unpicked fruits due to stress transmission through the fruit stalk connection structure.

[0021] Based on the positional offset of adjacent unpicked fruits, output the stress transmission path data of the fruit stalk.

[0022] In a preferred embodiment, S3 specifically refers to:

[0023] Based on the spatial connection structure data of all fruit stalks within the fruit cluster area, a multi-objective optimization function is constructed, which includes the robotic arm movement distance, the amount of change in robotic arm posture, and the risk index of position offset.

[0024] Under the constraints of the robotic arm joint motion range, fruit space collision avoidance, and continuous picking sequence, a multi-objective path optimization algorithm is used to generate several candidate continuous picking path sequences.

[0025] Calculate the comprehensive evaluation value of the multi-objective optimization function for each candidate continuous harvesting path sequence;

[0026] The candidate continuous picking path sequence with the best comprehensive evaluation value is determined as the robot's initial picking path.

[0027] In a preferred embodiment, S4 specifically refers to:

[0028] For the next target picking action in the robot's initial picking path, extract the spatial prediction offset of the corresponding target fruit from the fruit stalk stress transmission path data;

[0029] Calculate the target grasping pose of the next target fruit at the time of picking based on the spatial prediction offset, and generate the path correction amount;

[0030] The initial picking path of the robot is incrementally updated using path corrections to form an updated path node sequence;

[0031] The updated path node sequence is checked for continuity and reachability to generate a corrected robot picking path that satisfies the kinematic constraints of the robotic arm.

[0032] In a preferred embodiment, S5 specifically refers to:

[0033] Acquire the actual grasping pose data after the current target fruit is picked and the target grasping pose data recorded for the current target fruit in the corrected robot picking path;

[0034] Calculate the three-dimensional position deviation vector and attitude angle deviation vector between the actual grasping pose data and the target grasping pose data to form the current position deviation data;

[0035] The difference operation is performed between the current position deviation data and the spatial prediction offset of the current target fruit in the fruit stalk stress transfer path data to obtain the fruit position error residual vector.

[0036] Threshold discrimination and cluster analysis are performed on the residual vector of fruit position error to extract the abnormal change features of fruit position and posture;

[0037] By combining the abnormal change characteristics with the current position deviation data, abnormal change characteristic data of fruit position and posture are generated.

[0038] In a preferred embodiment, S6 specifically refers to:

[0039] Based on the abnormal change characteristic data, calculate the control parameter correction amount for the next target picking action;

[0040] The weights of the position offset risk index in the multi-objective optimization function are updated based on the control parameter correction.

[0041] The updated multi-objective optimization function is used to dynamically replan the corrected robot harvesting path to obtain the adjusted robot harvesting path.

[0042] Based on the control parameter correction and the adjusted robot picking path, the control parameters and path planning for the robot's next picking action are adjusted in real time.

[0043] The technical effects and advantages of the control method for a harvesting robot of the present invention are as follows:

[0044] By acquiring real-time 3D point cloud data of all fruits in the fruit cluster area and constructing the spatial connection topology of the fruit stalks, the spatial layout and mechanical relationship between fruits are accurately reflected. Based on the spatial connection structure of all fruit stalks in the fruit cluster, the positional offset of adjacent unpicked fruits is predicted, which can pre-compensate for micro-displacements caused by torque transmission during continuous picking. Multi-objective path planning is performed by combining the spatial connection structure data of all fruit stalks in the fruit cluster to generate the optimal initial picking sequence that takes into account the risks of movement distance, posture changes, and positional offset. The picking path is dynamically corrected using the predicted offset, which can reduce the grasping failure rate caused by causal interference effect. By extracting abnormal features from the actual grasping deviation, continuous or directional offsets can be identified. Finally, based on the abnormal change feature data and the corrected robot picking path, adaptive adjustment of clamping force, approach speed, and picking sequence is achieved. This improves the stability and success rate of continuous picking operations, reduces fruit damage, and improves robot picking efficiency and fruit integrity rate. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of a control method for a harvesting robot according to the present invention;

[0046] Figure 2 This is a data table of control parameters and path planning for the robot's fruit-picking action. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0048] Example

[0049] Figure 1 The present invention provides a control method for a harvesting robot, which includes the following steps:

[0050] S1: Real-time acquisition of 3D point cloud data of all fruits in the fruit cluster area, analysis of the spatial position relationship of fruits and the connection structure of fruit stalks, and output of spatial connection structure data of all fruit stalks in the fruit cluster.

[0051] S2: Based on the spatial connection structure data of all fruit stalks in the fruit cluster, analyze the stress transmission path after the picking action is applied to the fruit stalk of the target fruit, predict the positional offset of adjacent unpicked fruits, and output the stress transmission path data of the fruit stalk.

[0052] S3: Based on the spatial connection structure data of all fruit stalks in the fruit cluster, plan the initial path sequence for the picking robot to continuously pick fruits, and output the robot's initial picking path;

[0053] S4: Based on the stress transmission path data of the fruit stalk and the robot's initial picking path, dynamically correct the spatial position of the next fruit picking action and output the corrected robot picking path.

[0054] S5: Based on the actual position deviation data after each picking action of the robot, identify the abnormal change characteristics of the fruit position and posture, and output the abnormal change characteristic data of the fruit position and posture.

[0055] S6: Based on the abnormal change characteristic data and the corrected robot picking path, adjust the control parameters and path planning of the robot's next picking action in real time.

[0056] S1: Real-time acquisition of 3D point cloud data of all fruits in the fruit cluster area, analysis of the spatial positional relationship of the fruits and the connection structure of the fruit stalks, and output of the spatial connection structure data of all fruit stalks in the fruit cluster, including:

[0057] Real-time acquisition of 3D point cloud data of all fruits in the fruit cluster area, followed by preprocessing to obtain preprocessed 3D point cloud data;

[0058] The fruit cluster region refers to a spatial area containing multiple fruits, fruit stalks, and adjacent branches. 3D point cloud data refers to a discrete set of points with 3D coordinates generated by a depth camera or laser rangefinder. Each point contains Cartesian coordinates X, Y, and Z, in meters or millimeters, and includes fields such as surface normal vector, curvature, intensity, and timestamp. Real-time acquisition refers to continuously acquiring 3D point cloud data at fixed time intervals, which can be 33 milliseconds, 50 milliseconds, or 100 milliseconds, depending on the robot arm's movement speed and the target scene's lighting conditions. The sensor is fixed above the robot arm's end effector or near the robot arm's base. The installation angle is calibrated using a calibration plate, and the calibration results provide the rotation matrix and translation vector between the sensor coordinate system and the robot arm's base coordinate system for coordinate unification. The preprocessing workflow includes time synchronization, outlier removal, noise filtering, voxel downsampling, region of interest clipping, and normal vector estimation. Time synchronization synchronizes the point cloud frames within the same control cycle with the robot arm encoder's attitude data to ensure consistent coordinate transformation. Outlier removal employs statistical methods, using the K-nearest neighbor distance distribution of each point as a basis, and setting a standard deviation threshold (e.g., K is set to 20-50, and the standard deviation threshold to 1.0-2.0) to remove points with abnormal distance distributions. Noise filtering uses radius filtering, with a radius threshold of 10-30 mm and a minimum neighbor number threshold of 5-10 to remove isolated points. Voxel downsampling uses a voxel raster method, with voxel side lengths of 2-10 mm to obtain a point cloud with uniform density, reducing computational load. Region of interest (ROI) clipping is based on fruit cluster bounding boxes, obtained through coarse segmentation, with a 10%-20% redundant boundary left in the clipping range to preserve the point cloud at the connection between the fruit stalk and branch. Normal vector estimation is based on local neighborhood principal component analysis, with the neighborhood radius set to 2-3 times the voxel side length, outputting a unit normal vector for each point. The preprocessed 3D point cloud data is output in a structured data format. The data structure includes a point coordinate matrix, a normal vector matrix, a curvature vector, a timestamp sequence, and a coordinate system identifier. The file format can be binary point cloud format or a custom project cache format. For example, using an apple cluster as the object, a depth frame with a resolution of 1280×720 is acquired at a distance of one meter from the target. The voxel downsampling voxel side length is set to 5 mm, and the outlier removal parameters are set to K=30 and the standard deviation threshold = 1.5, resulting in preprocessed 3D point cloud data with approximately 150,000 to 200,000 points per frame. The data format, parameters, and coordinate definitions remain unchanged.

[0059] Based on the preprocessed 3D point cloud data, the spatial position and geometric shape of the fruit within the fruit cluster area are identified, and the spatial positional relationship between the fruits and the characteristics of the fruit stem connection structure within the fruit cluster area are analyzed.

[0060] Identifying the spatial location of fruits refers to determining the geometric centroid position and reference point position of each fruit. The identification process includes point cloud segmentation, target clustering, and geometric fitting. Point cloud segmentation uses a geometric continuity-based clustering method to perform Euclidean clustering on the preprocessed 3D point cloud data. The clustering distance threshold is set to 10-20 mm based on the voxel side length, the minimum number of cluster points is set to 500-1000, and the maximum number of cluster points is set to within 200,000, in order to separate individual fruits from branches. After target clustering is completed, geometric shape determination is performed on each candidate cluster. The geometric shape is determined by spherical or ellipsoidal fitting, and the fitting method is parameter fitting in the least squares sense. The output is the sphere radius or ellipsoid major and minor axis parameters, the fitting residual, and the bounding box parameters aligned with the circumscribed axis. The spatial location is determined by the fitted centroid, and the geometric shape is determined by the fitted parameters. To ensure continuity with the pedicel analysis, the pedicel connection region needs to be identified in the fruit cluster point cloud. The fruit stalk connection structure features refer to the geometric and topological properties of the slender cylindrical or conical structure connecting the fruit and the branch in the point cloud. These include the stalk axis direction vector, the coordinates of the bottom endpoint of the stalk, the local curvature distribution of the contact ring between the stalk and the fruit, and the range of the proximal diameter of the stalk. The stalk axis direction vector is obtained by cylindrical fitting of a point set with a radius of 20 to 40 mm in the neighborhood below or to the side of the fruit. The cylindrical fitting radius ranges from 3 to 10 mm, and the minimum axial length threshold is 10 to 30 mm. The coordinates of the bottom endpoint of the stalk are obtained by extending the cylindrical axis in the negative direction to the nearest intersection with the branch point cloud. The spatial relationship between fruits is generated through two structures: the first structure is the Euclidean distance matrix between the centroids of the fruits, where the matrix element d(i,j) represents the distance between the i-th fruit and the j-th fruit; the second structure is a set of relative orientation vectors of the fruits, where vector v(i,j) points from the centroid of the i-th fruit to the centroid of the j-th fruit, and after normalization, it is used to constrain the approach direction in path planning. For example, in a scenario containing three adjacent apples, Euclidean clustering yields three fruit clusters and one set of branch clusters; the fitted radius of the sphere in each fruit cluster is approximately 45 to 55 millimeters; the fitted radius of the cylindrical stem is approximately 3 to 5 millimeters, with the axial direction forming an angle of 30 to 60 degrees with the direction of gravity; in the distance matrix, d(1,2) = 120 millimeters, d(2,3) = 95 millimeters, and d(1,3) = 210 millimeters; the set of relative orientation vectors is used to generate a candidate set of approach directions for grasping. Through the above identification and analysis, four types of data are obtained: fruit spatial location, fruit geometric shape, spatial relationship between fruits, and stem connection structure features. A unified coordinate system identifier and timestamp are recorded in the data structure to ensure consistency with the input of finite element mechanics analysis and multi-objective path optimization.

[0061] Based on the spatial positional relationship between fruits and the characteristics of the fruit stalk connection structure within the fruit cluster area, a spatial connection topology structure between all fruit stalks within the fruit cluster area is established, and the spatial connection structure data of all fruit stalks within the fruit cluster area is output.

[0062] Spatial connectivity topology refers to a graph structure reflecting the connections between fruit stalks and between fruit stalks and branches. The graph structure includes a set of nodes and a set of edges. The node set contains two types of nodes: the first type is fruit stalk nodes, each recording the stalk axis direction vector, the coordinates of the bottom endpoint of the stalk, the near-end diameter of the stalk, the estimated length of the stalk, and the index of the fruit it belongs to; the second type is branch connection nodes, each recording the approximate direction of the branch centerline, the local radius of the branch, and its position in a unified coordinate system. The edge set represents two types of connectivity relationships: the connectivity between fruit stalk nodes and branch connection nodes represents the connection between fruit stalks and branches; the connectivity between multiple fruit stalk nodes under the same branch connection node represents the association of fruit stalk groups on the same branch. The connectivity relationship construction rules are based on two conditions: spatial proximity and axial geometric consistency. Edges are established when both spatial distance thresholds and axial angle thresholds are met. The spatial distance threshold is set to the nearest distance from the bottom endpoint of the fruit stalk to the branch point cloud being less than 15 mm, and the axial angle threshold is set to the angle between the fruit stalk axis and the local direction of the branch being less than 30 degrees. The above rules ensure that the graph structure is consistent with the actual structure, avoiding erroneous connections. The spatial connection topology is stored simultaneously in two forms: an adjacency list and an adjacency matrix. The adjacency list records the index of each node's connected nodes and edge attributes. The edge attributes include the connection type identifier and the calculation time. The element A(p,q) of the adjacency matrix A takes a value of zero or one, with a value of one indicating that node p and node q are connected. In the engineering implementation, the spatial connection structure data is encapsulated using a unified data frame. The data frame fields include a node list, an edge list, an adjacency matrix, a timestamp, and a coordinate system identifier. For example, in a fruit cluster consisting of three apples and a branch, the graph structure contains three fruit stalk nodes and one branch connecting node. The edge list contains three edges, connecting the three fruit stalk nodes and the branch connecting node respectively. The three rows and three columns of the adjacency matrix corresponding to the branch connecting node have a value of one, and the remaining positions have a value of zero. To ensure data consistency, each fruit stalk node in the spatial connection structure data stores an index corresponding to the spatial position and geometric shape of the fruit in step S1, ensuring consistency in the mapping from geometric identification to topology construction.

[0063] S2: Based on the spatial connection structure data of all fruit stalks in the fruit cluster, analyze the stress transmission path after the picking action is applied to the fruit stalk of the target fruit, predict the positional offset of adjacent unpicked fruits, and output the stress transmission path data of the fruit stalk, including:

[0064] Based on the spatial connection structure data of all fruit stalks within the fruit cluster area, a finite element mechanical analysis model of the fruit stalk connection structure within the fruit cluster area is established.

[0065] The spatial connection structure data of all fruit stalks within the fruit cluster area includes a node list, edge list, adjacency matrix, timestamp, and coordinate system identifier. The fruit stalk nodes in the node list contain the stalk axis direction vector, stalk bottom endpoint coordinates, stalk near-end diameter, estimated stalk length, and the index of the corresponding fruit. Branch connection nodes record the branch centerline direction, local radius, and position. The finite element mechanical analysis model is a digital simulation model built based on the spatial connection structure data. The finite element mechanical analysis model construction method includes model geometry construction, material property setting, mesh generation, and boundary condition definition. The model geometry construction is based on the spatial connection topology of the fruit stalks. A three-dimensional geometric solid model of the fruit stalk connection structure is established based on the stalk axis direction vector and the stalk bottom endpoint coordinates, using a cylindrical or conical modeling method. The bottom position of the cylinder or cone coincides with the stalk bottom endpoint coordinates, the axial length is the estimated stalk length, the bottom radius is half the stalk near-end diameter or the radius obtained from point cloud fitting, and the top radius is set according to the fruit contact ring diameter. Material properties are set based on the elasticity and rigidity characteristics of actual fruit stalks, which are typically made of plant fiber. Young's modulus is set to 50 MPa to 500 MPa, Poisson's ratio to 0.2 to 0.4, and density to 800 to 1200 kg / m³. Meshing is performed using common finite element analysis methods, with hexahedral or tetrahedral elements and element sizes ranging from 0.5 mm to 2 mm. Boundary conditions describe the fixed constraints and loading conditions of the fruit stalk connection structure. Branch connection nodes are under fixed constraints to simulate the support and fixation conditions of the actual fruit stalk and branch connection. The ends of the fruit stalk nodes and the connection points with the fruit are designated as loading surfaces. The model geometry, material properties, mesh generation, and boundary condition definitions are all recorded in the finite element mechanical analysis model data structure. For example, in a cluster of three apples, a finite element mechanical analysis model is established with three fruit stems connected. Each fruit stem is a cylindrical structure with a diameter of 6 mm and a length of 40 mm. The Young's modulus of the material is 200 MPa, the Poisson's ratio is 0.3, the mesh element size is 1 mm, and the branch connection nodes are fixed in a spatial coordinate system.

[0066] The transmission path of torque and stress generated by the clamping mechanism of the picking robot acting on the fruit stem when picking the target fruit is analyzed based on the finite element mechanical analysis model.

[0067] This study analyzes the stress and torque transmission path characteristics within the fruit stem and its connecting structure when the gripping mechanism of a harvesting robot applies mechanical force to the fruit stem. The mechanical force applied to the end of the fruit stem by the gripping mechanism is determined by the mechanism's motion parameters. The magnitude of the gripping force is generally between 0.5 and 5 Newtons, and the direction of the force is between 10 and 60 degrees along the fruit stem axis or perpendicular to the axis, depending on the robotic arm's gripping posture. A finite element method (FEM) model applies the gripping force to the loading surface at the end of the fruit stem. Through static or dynamic simulation analysis, the stress field distribution and torque transmission path of the nodes and elements within the fruit stem and connecting structure are obtained. The stress field data includes the magnitude and direction of stress in each element, and the torque transmission path data includes the route of stress transmission from the loading surface to the fixed node (branch connection node) unit by unit, along with the change in stress torque. The transmission path can be represented graphically or as a data list. The data list records the distribution of stress and torque from the loading surface to the branch connection node as they change with spatial nodes. For example, when a harvesting robot applies a tensile force of 2 Newtons to the stalk of a target fruit, with the direction of the force at a 30-degree angle to the axial direction of the stalk, finite element analysis reveals that the maximum stress occurs at the junction of the stalk and the branch, which is approximately 3 MPa. The torque is transmitted from the top of the stalk node by node to the bottom, and the stress gradually decreases along the length of the stalk.

[0068] Based on the transmission path of torque and stress, predict the positional shift of adjacent unpicked fruits due to stress transmission through the fruit stalk connection structure.

[0069] Based on torque and stress transmission path data, this study analyzes and predicts the spatial positional shift of adjacent unpicked fruits when stress is transmitted from the fruit stalk to the stalk of the adjacent fruit. The stress transmission to the adjacent fruit stalk causes minute deformations, including changes in the length and tilt of the stalk axis, resulting in changes in the spatial position and orientation of the connected fruits. The methods for predicting positional shifts include calculating the change in the position of the stalk tip using linear interpolation based on simulation data, and calculating the distance and direction of the shift in the fruit's center of gravity due to stalk tilting based on elastic deformation theory. The calculated positional shift data includes three-dimensional displacement vectors and changes in orientation angles. The three-dimensional displacement vectors include displacements in the X, Y, and Z axes (in millimeters); the changes in orientation angles include the tilt angle of the stalk axis relative to its initial state (in degrees). For example, in the case of a cluster of three apples, after applying a tensile force to the first fruit, stress transmission path analysis predicts that the second fruit will experience a spatial shift of 2 mm along the X-axis and 1 mm along the Y-axis due to the transmitted stress, and the fruit stem tilt angle will increase by 5 degrees. The predicted position shift data is recorded in the data table for use in optimizing the harvesting control strategy.

[0070] Based on the positional offset of adjacent unpicked fruits, output the stress transmission path data of the fruit stalk;

[0071] The predicted positional offset data of adjacent unpicked fruits is output through a specific data structure, namely, the fruit stem stress transmission path data, including the fruit number, the three-dimensional vector of positional offset, the change in attitude angle, the corresponding finite element analysis data record number, and the timestamp. This data is used by the picking robot to make real-time corrections to the fruit grasping position and attitude in the planning of the next picking action.

[0072] S3: Based on the spatial connection structure data of all fruit stalks in the fruit cluster, plan the initial path sequence for the harvesting robot to continuously harvest fruits, and output the robot's initial harvesting path, including:

[0073] Based on the spatial connection structure data of all fruit stalks within the fruit cluster area, a multi-objective optimization function is constructed, which includes the robotic arm movement distance, the amount of change in robotic arm posture, and the risk index of position offset.

[0074] The spatial connection structure data of all fruit stalks within the fruit cluster area includes a node list, an edge list, an adjacency matrix, timestamps, and coordinate system identifiers. The node list records the axial direction vector, bottom endpoint coordinates, near-end diameter, estimated length, and fruit index for each fruit stalk node. Branch connection nodes record the branch centerline direction, local radius, and position. The edge list records the connection relationships between nodes. The adjacency matrix records the connection relationships between nodes in matrix form, with matrix elements being either zero or one, where one indicates a connection between corresponding nodes. The multi-objective optimization function is a mathematical expression used to evaluate the continuous harvesting path of the harvesting robot. It includes multiple evaluation indicators, each assigned a corresponding weight coefficient. These indicators are fused through linear or nonlinear combinations to obtain a comprehensive evaluation value for the path. The multi-objective optimization function includes three indicators: robotic arm movement distance, robotic arm posture change, and position offset risk. The robotic arm's travel distance refers to the length of the stroke required for the end effector of the robotic arm to move from the current fruit position to the next fruit position, calculated using three-dimensional Euclidean distance, with the unit being millimeters. The robotic arm's posture change refers to the degree of posture change that occurs when the end effector completes the picking of two adjacent fruits. Posture is represented using Euler angles or quaternions, and the change is calculated as the difference in posture angles, with the unit being degrees. The positional offset risk index is obtained based on the fruit stalk stress transmission path data, reflecting the degree of positional offset risk that may occur due to the mechanical force transmission effect. A higher positional offset risk index indicates a higher probability of positional offset during fruit picking. The positional offset risk index is obtained by analyzing the positional offset vector amplitude recorded in the fruit stalk stress transmission path data and the stability of the fruit stalk connection structure. The square or exponential function of the offset amplitude is used as the positional offset risk index. For example, when the predicted positional offset amplitude is 2 millimeters, the positional offset risk index is calculated as 4 (square method) or a higher exponential form to reflect a high-risk situation. The multi-objective optimization function expression is as follows: Objective function value = Weight coefficient 1 × robotic arm movement distance + Weight coefficient 2 × robotic arm posture change + Weight coefficient 3 × position offset risk index. The weight coefficients are determined according to the robot's task requirements. For example, when robotic arm movement efficiency is the primary concern, weight coefficient 1 is set to 0.5, and weight coefficients 2 and 3 are each set to 0.25. If position offset risk control is the primary concern, then weight coefficient 3 is set to 0.6, and weight coefficients 1 and 2 are each set to 0.2. The calculation results and weight coefficients of each index are preset before each harvesting task and are uniformly recorded and retrieved in the robot control system.

[0075] Under the constraints of the robotic arm joint motion range, fruit space collision avoidance, and continuous picking sequence, a multi-objective path optimization algorithm is used to generate several candidate continuous picking path sequences.

[0076] The constraints on the joint motion range of the robotic arm refer to the limitation on the range of motion angles of each joint of the harvesting robot. These limits are determined based on the hardware parameters of the robotic arm. For example, the angle ranges of each joint in a six-axis robotic arm are ±170 degrees, ±120 degrees, ±135 degrees, ±190 degrees, ±120 degrees, and ±360 degrees. The robot cannot perform actions exceeding these ranges. The constraints on avoiding collisions in the fruit space prevent the end effector of the robotic arm from colliding with other unharvested fruits within the fruit cluster area during the harvesting process, thus avoiding damage to the robotic arm or the fruits. The constraints on the continuous harvesting sequence indicate that there exists an optimal continuous harvesting order among the fruits. That is, after the current fruit harvesting action is completed, the harvesting position of the next target fruit must consider the principles of minimizing the robotic arm's movement distance, minimizing posture changes, and minimizing the risk of positional deviation. These constraints are expressed in the form of mathematical inequalities or equations in the robot path optimization algorithm, and all constraints must be satisfied during robot motion path optimization.

[0077] The multi-objective path optimization algorithm, based on the constrained optimization problem consisting of the above three constraints and the multi-objective optimization function, gradually generates multiple feasible candidate continuous picking path sequences through heuristic search or intelligent optimization methods. Each candidate path sequence is an ordered set of several continuous fruit picking actions, and each fruit picking action records the grasping position and posture of the target fruit.

[0078] Genetic algorithms are used as multi-objective path optimization algorithms, specifically as follows:

[0079] Define the chromosome encoding method. Chromosome encoding represents the candidate continuous picking path sequence of the picking robot. That is, the gene sequence of each chromosome corresponds to a fruit picking order. The value at each gene locus represents the fruit number. For example, in a fruit cluster containing three apple fruits, the gene sequence [1,2,3] represents the picking order of fruit 1-fruit 2-fruit 3.

[0080] A population initialization strategy is determined, where the population is the initial set of candidate solutions. The population size is typically 20 to 50 chromosomes, and 30 chromosomes are selected. The initial population is generated using a random sorting method, for example, by randomly arranging the numbers of the fruits to be picked in the fruit cluster to obtain the initial candidate picking order.

[0081] Define a fitness function, which is the multi-objective optimization function, and calculate it as follows:

[0082] Fitness value = weighting coefficient 1 × robotic arm movement distance + weighting coefficient 2 × robotic arm posture change + weighting coefficient 3 × position offset risk index;

[0083] The weighting coefficients 1, 2, and 3 are set to 0.5, 0.25, and 0.25, respectively.

[0084] For each candidate path chromosome, the robot arm movement distance, robot arm posture change, and position offset risk index between two adjacent fruits are calculated, and the comprehensive fitness value is calculated based on the weight coefficient. The smaller the fitness value, the better the path.

[0085] Genetic operators are defined, including selection, crossover, and mutation operators;

[0086] The selection operator uses a roulette wheel method, selecting chromosomes with higher fitness values ​​to enter the next generation;

[0087] The crossover operator uses a partial matching crossover method. For example, it crosses two path chromosomes [1,2,3] and [3,1,2] by randomly selecting the crossover segment, exchanging the gene positions within the segment, and generating a new path sequence.

[0088] The mutation operator uses the exchange mutation method, which randomly selects two gene loci within the chromosome, for example, mutating [1,2,3] into [2,1,3] to ensure diversity.

[0089] The algorithm terminates when the number of iterations is set to 50 to 100 generations, for example, 80 generations. When the preset number of generations is reached, or when the fitness value changes less than a preset threshold, the algorithm terminates and outputs the path chromosome with the smallest fitness value, which is the optimal continuous harvesting path sequence.

[0090] Taking a fruit cluster containing three apples as an example, through the above genetic algorithm implementation process, multiple path chromosomes are initially randomly generated. After 80 generations of iterative calculation, the chromosome path sequence with the smallest comprehensive fitness value is [1,3,2]. At this time, the total movement distance of the robotic arm is 200 mm, the total posture change is 25 degrees, and the position offset risk index is 3, which is better than other paths. Thus, this chromosome sequence is determined as the initial robot picking path.

[0091] Calculate the comprehensive evaluation value of the multi-objective optimization function for each candidate continuous harvesting path sequence;

[0092] For each candidate continuous harvesting path sequence, a multi-objective optimization function is used to calculate the entire path sequence. For each continuous harvesting action within the path, the robotic arm's movement distance, attitude change, and positional offset risk index are calculated. The three index values ​​for each action are multiplied by their respective weighting coefficients and then summed to obtain the comprehensive evaluation value of the current candidate continuous harvesting path sequence. A smaller comprehensive evaluation value indicates a better path sequence. For example, for candidate path sequence 1 (fruit 1-fruit 2-fruit 3), the robotic arm's movement distance for each continuous action is calculated to be 120 mm and 95 mm, the attitude change to be 10 degrees and 8 degrees, and the positional offset risk index to be 2 and 1.5, respectively. The comprehensive evaluation value is obtained after combining these values ​​with the weighting coefficients. All candidate path sequences are calculated sequentially, and the comprehensive evaluation values ​​are recorded in the path sequence evaluation data table.

[0093] The candidate continuous harvesting path sequence with the best comprehensive evaluation value is determined as the robot's initial harvesting path;

[0094] After calculating the comprehensive evaluation value of all candidate continuous picking path sequences, the path sequence with the smallest comprehensive evaluation value is selected from the path sequence evaluation data table as the optimal path sequence. This optimal path sequence is then determined as the robot's initial picking path, serving as the initial execution sequence for the robot's actual picking actions. The robot's initial picking path records the target spatial position and posture data of the robot for each fruit picking action. This data is used for robot motion control and path correction tasks during actual picking operations, and its format is consistent with the path sequence evaluation data table.

[0095] S4: Based on the fruit stalk stress transmission path data and the robot's initial picking path, dynamically correct the spatial position of the next fruit picking action, and output the corrected robot picking path, including:

[0096] For the next target picking action in the robot's initial picking path, extract the spatial prediction offset of the corresponding target fruit from the fruit stalk stress transmission path data;

[0097] The robot's initial picking path records the target grasping pose of each fruit, including the three-dimensional coordinates of the target fruit's spatial position and the three-dimensional attitude angle data of the robotic arm's end effector. The next target picking action of the robot's initial picking path indicates the next fruit picking task to be executed after the current picking action is completed. The fruit stem stress transmission path data includes the fruit number, three-dimensional vector data of position offset, attitude angle change data, corresponding finite element analysis data number, and timestamp, recording the predicted offset of spatial position and attitude between fruit stems due to mechanical forces. The spatial predicted offset is expressed as a three-dimensional displacement vector and attitude angle offset. The three-dimensional displacement vector records the spatial position offset of the corresponding fruit due to the mechanical force applied to adjacent fruits in the previous picking action, caused by the transmission effect through the fruit stem connection structure, and is expressed as coordinate offset values ​​in the X, Y, and Z axes, with the offset amplitude in millimeters. The coordinate system definition is consistent with steps S1 and S2. The attitude angle offset is the deflection angle of the fruit's axis direction from its original initial direction due to the torque transmission from the fruit stalk. It is recorded as the deflection angle with reference to the robot arm coordinate system, in degrees. For example, if the next target picking action in the robot's initial picking path is to pick apple fruit number 2, then based on the position offset of fruit 2 recorded in the fruit stalk stress transmission path data, the three-dimensional displacement vector and attitude angle offset are extracted. For example, the three-dimensional displacement vectors are 2.0 mm offset in the X direction, 1.5 mm offset in the Y direction, and -0.5 mm offset in the Z direction; the attitude angle offset is a 3-degree clockwise deflection around the Y-axis. The extracted spatial prediction offset data is used to calculate the robot's grasping pose correction.

[0098] Calculate the target grasping pose of the next target fruit at the time of picking based on the spatial prediction offset, and generate the path correction amount;

[0099] Based on the spatial prediction offset, the target grasping pose of the next target fruit, recorded in the initial picking path, is calculated at the actual picking time to correct the robot's picking action execution path. The target grasping pose includes position and attitude. Position is obtained by vector addition of the original 3D coordinates of the target fruit and the predicted offset 3D displacement vector; that is, the original target position coordinates are added to the predicted displacements in the X, Y, and Z directions to obtain the final adjusted grasping position coordinates. Attitude is obtained by vector addition of the original 3D attitude angles of the initially planned robotic arm end effector and the predicted attitude angle offset; that is, the original attitude angles are directly added to the predicted deflection angles to generate the final grasping attitude data. The path correction is the incremental value after correcting the position and attitude respectively. The displacement correction is the 3D displacement vector data, and the attitude angle correction is the 3D angle deflection data. For example, for apple number 2, the original target grasping pose space position is (300.0, 150.0, 120.0) mm, and the attitude angle is (0, 90, 0) degrees. After adding the predicted displacement vector (2.0, 1.5, -0.5) mm and the attitude angle deflection (0, 3, 0) degrees, the adjusted target grasping position is (302.0, 151.5, 119.5) mm, and the attitude angle is adjusted to (0, 93, 0) degrees. The path correction is recorded as the displacement correction (2.0, 1.5, -0.5) mm and the attitude angle correction (0, 3, 0) degrees. The path correction is used to update the robot's initial picking path.

[0100] The initial picking path of the robot is incrementally updated using path corrections to form an updated path node sequence;

[0101] The initial picking path nodes record the spatial position of the target fruit and the robotic arm posture for each consecutive picking action of the robot. Incremental updates are used, which involve adding calculated displacement and posture angle corrections to the target fruit node position and posture data that need correction, based on the original path node data, to form corrected path node data. The update process is as follows: for the target node position coordinates in the robot's initial picking path, the displacement corrections are added item by item; for the posture angle data, the posture angle corrections are added item by item, generating updated path node sequence data. The updated path node sequence data records the actual spatial position coordinates of each updated node and the three-dimensional posture data of the robotic arm's end effector. Taking apples as an example, the original position coordinates of fruit number 2 (300.0, 150.0, 120.0) mm are updated to (302.0, 151.5, 119.5) mm, and the original posture (0, 90, 0) degrees is updated to (0, 93, 0) degrees, completing the incremental update and generating corrected node data.

[0102] The updated path node sequence is checked for continuity and reachability to generate a corrected robot picking path that satisfies the kinematic constraints of the robotic arm.

[0103] The updated path node sequence needs to undergo continuity and reachability checks to ensure that the robot arm's motion path is smooth, continuous, and achievable when performing continuous actions. Continuity check examines whether there are abrupt changes in motion of the robot arm's end effector between adjacent path nodes, ensuring that the robot arm's position and posture changes are within preset thresholds. Reachability check verifies whether the spatial position and posture of each path node are within the robot arm's workspace. Specifically, this involves verifying the robot arm's forward and inverse kinematics calculations to ensure that all joints can reach the target node's position and posture without exceeding the physical joint motion limits. Continuity check is based on the difference between the positions and postures of two adjacent nodes. The position change threshold is set to no greater than the robot arm's maximum step length in a single movement, e.g., 50 mm, and the posture change threshold is set to no more than 10 degrees. Reachability check performs kinematic calculations based on the robot arm's joint angle range, ensuring that the kinematic solution for each node exists and that the joint angles meet the motion limits, e.g., for a six-axis robot arm, the joint angles should not exceed the limit range of ±170 degrees to ±360 degrees. Once the path node sequence passes the continuity and reachability checks, the output is the corrected robot picking path that satisfies the kinematic constraints of the robotic arm. For example, after updating the nodes of an apple cluster, both the continuity and reachability checks are satisfied, resulting in the final corrected robot picking path node sequence, which is then used to perform the picking operation.

[0104] S5: Based on the actual positional deviation data after each picking action of the robot, identify abnormal changes in the fruit's position and posture, and output the abnormal change feature data of the fruit's position and posture, including:

[0105] Acquire the actual grasping pose data after the current target fruit is picked and the target grasping pose data recorded for the current target fruit in the corrected robot picking path;

[0106] The actual grasping pose data after harvesting the current target fruit refers to the actual position coordinates and attitude angles recorded by the robot arm's end effector when actually harvesting the fruit. This actual grasping pose data is measured and recorded in real time by high-precision encoders and attitude sensors installed on the end effector. The actual grasping pose data includes spatial coordinate data and attitude data of the end effector when actually grasping the fruit. The spatial coordinate data consists of the position coordinates of the end effector in a three-dimensional Cartesian coordinate system measured by the robot arm's joint encoder, including X, Y, and Z coordinates, in millimeters. The attitude data is measured by the attitude sensor and expressed in three-dimensional attitude angles, typically using Euler angles, including rotation angles around the X-axis, Y-axis, and Z-axis, in degrees. The actual grasping pose data is stored in the robot arm control system in the form of structured data frames. The data frame fields include a timestamp, coordinate system identifier, position coordinate vector, and attitude angle vector.

[0107] The target grasping pose data recorded for the current target fruit in the corrected robot picking path refers to the theoretical values ​​of the grasping target position and posture of the current fruit recorded in the node sequence of the corrected robot picking path. The target grasping pose data is calculated based on the fruit stem stress transmission path data and the initial picking path, including calculated and adjusted spatial position coordinates and posture angles. The spatial position coordinates use the same three-dimensional Cartesian coordinate system as the actual grasping pose data, with the coordinate system definition consistent with steps S1 and S2, and the data format also consistent, including X, Y, and Z coordinates. The posture data is also represented by the three-dimensional posture angles of the robotic arm's end effector, including rotation angles around the X-axis, Y-axis, and Z-axis, in degrees. The target grasping pose data is also stored in the form of structured data frames, with fields including target fruit number, timestamp, coordinate system identifier, target position coordinates, and target posture angle data, ensuring complete consistency in data format when compared with the actual grasping pose data.

[0108] For example, in an apple-picking robot scenario, when apple number 3 is successfully picked up by the robot, the high-precision position encoder installed on the end effector of the robotic arm records the actual grasping position coordinates as (502.0, 304.0, 158.0) mm, and the attitude sensor at the end effector records the attitude data as 0 degrees rotation around the X-axis, 93 degrees rotation around the Y-axis, and 1 degree rotation around the Z-axis. However, in the corrected robot picking path node sequence, for apple number 3, the calculated target grasping pose is (500.0, 305.0, 160.0) mm, with attitude angles of 0 degrees rotation around the X-axis, 90 degrees rotation around the Y-axis, and 0 degrees rotation around the Z-axis.

[0109] Calculate the three-dimensional position deviation vector and attitude angle deviation vector between the actual grasping pose data and the target grasping pose data to form the current position deviation data;

[0110] The current position deviation data includes a 3D position deviation vector and an attitude angle deviation vector. The 3D position deviation vector is obtained by calculating the difference between the actual grasping position coordinates and the target grasping position coordinates in each dimension. Specifically, it is the difference between the actual grasping position coordinates and the target grasping position coordinates, expressed in millimeters. The three dimensions correspond to the deviation values ​​of the X-axis, Y-axis, and Z-axis, respectively. The attitude angle deviation vector is obtained by calculating the difference between the actual grasping attitude angle and the target grasping attitude angle. Specifically, it is the difference between the actual attitude angle and the target attitude angle, expressed in degrees. The three dimensions are the difference in rotation angles around the X-axis, Y-axis, and Z-axis, respectively. The 3D position deviation vector and the attitude angle deviation vector are merged to form a unified data structure, which is the current position deviation data, recording the difference between the robot's actual picking behavior and the planned picking behavior. For example, for the apple-picking action numbered 3, the actual grasping pose data records the position coordinates as (502.0, 304.0, 158.0) mm, while the target grasping pose data shows the position coordinates as (500.0, 305.0, 160.0) mm. The calculated position deviation vector is (2.0, -1.0, -2.0) mm, indicating that the actual grasping position is offset by 2 mm in the X-axis direction, -1 mm in the Y-axis direction, and -2 mm in the Z-axis direction relative to the target position. Regarding the attitude angles, the actual attitude is (0 degrees, 93 degrees, 1 degree), and the target attitude is (0 degrees, 90 degrees, 0 degrees). Therefore, the attitude angle deviation vector is calculated as (0, 3, 1) degrees, indicating no deviation around the X-axis, a 3-degree deviation around the Y-axis, and a 1-degree deviation around the Z-axis.

[0111] The difference operation is performed between the current position deviation data and the spatial prediction offset of the current target fruit in the fruit stalk stress transfer path data to obtain the fruit position error residual vector.

[0112] The current position deviation data represents the actual measurement error between the current actual grasping position and posture of the fruit and the target grasping position and posture. The spatial prediction offset recorded in the fruit stalk stress transfer path data represents the theoretical offset predicted based on the mechanical transfer effect between the fruit stalks in the fruit cluster area. The spatial prediction offset includes a position prediction offset vector and an attitude angle prediction offset vector. The position prediction offset vector records the predicted offset values ​​in the X, Y, and Z axis directions, while the attitude angle prediction offset vector records the attitude angle predicted offset values ​​around the X, Y, and Z axes. The difference operation refers to calculating the vector difference between the current position deviation data and the spatial prediction offset. Subtraction is performed on the position data and attitude data dimension by dimension to analyze the deviation between the actual error and the theoretical prediction error. The final result is called the fruit position error residual vector, which includes the position error residual vector and the attitude error residual vector.

[0113] The residual vector of position error represents the amount of error remaining after subtracting the theoretically predicted position deviation from the actual measured position deviation in the X, Y, and Z axes. It reflects the position deviation caused by other interference factors that may exist during the fruit picking process, in addition to the mechanical effect of the fruit stalk. The residual vector of attitude error represents the residual angle error after subtracting the theoretically predicted attitude deviation from the actual measured attitude angle deviation in the X, Y, and Z axes. It reflects the source of additional deviation introduced by the attitude angle during the actual grasping process.

[0114] The residual vector of position error = current position deviation vector - position prediction offset vector; the residual vector of attitude error = attitude angle deviation vector - attitude angle prediction offset vector. For example, taking fruit number 3 in an apple cluster as an example, the position deviation vector in the current position deviation data is (2.0, -1.0, -2.0) mm, the attitude angle deviation vector is (0, 3, 1) degrees, the position prediction offset vector for fruit number 3 in the fruit stem stress transfer path data is (1.5, -0.5, -1.0) mm, and the attitude prediction offset vector is (0, 2, 0) degrees. Then the calculated result of the residual vector of position error is (0.5, -0.5, -1.0) mm; the calculated result of the residual vector of attitude error is (0, 1, 1) degrees. The above results are the residual vector data of fruit position error.

[0115] Threshold discrimination and cluster analysis are performed on the residual vector of fruit position error to extract the abnormal change features of fruit position and posture;

[0116] The residual vector of fruit position error reflects the portion of the actual deviation in fruit position and posture that cannot be explained by the mechanical effects of the fruit stalk, and may contain additional unknown interference factors or random disturbances. The residual vector of error needs to be analyzed in depth through threshold discrimination and cluster analysis to extract abnormal changes in fruit position and posture in order to adjust harvesting strategies.

[0117] Threshold discrimination refers to setting and comparing threshold values ​​for the amplitude of the residual error vector. The amplitude is defined as the Euclidean distance or magnitude between the residual vectors of 3D position error and attitude error. The threshold is set based on statistical analysis of historical harvesting data to determine the normal distribution range of the residual error vector amplitude. For example, the threshold for the residual position error vector is set between 0.5 mm and 1.0 mm, and the threshold for the residual attitude error vector is set between 1 degree and 3 degrees. If the amplitude of the residual error vector exceeds the set threshold, an abnormal change is considered to exist in the harvesting action, and it is recorded as abnormal change characteristic data.

[0118] Cluster analysis involves statistically analyzing the direction, magnitude, and frequency of the residual error vector across several consecutive harvesting actions. Using data clustering algorithms such as K-means clustering, it counts the frequency of anomalous errors with similar directions and magnitudes, i.e., the frequency of occurrence. The results of cluster analysis are characteristic parameters of anomalous changes in fruit position and posture, including the magnitude and direction vector of the residual error vector, and the frequency of occurrence of the anomalous change in consecutive actions. The direction vector is represented by a normalized unit vector of the residual error vector, recording the main spatial or postureal change direction of the error; the magnitude is the size of the error; and the frequency of occurrence is the recurrence frequency of similar error anomalies across several consecutive harvesting actions.

[0119] For example, in an apple picking scenario, if the apple numbered 3 is picked ten times consecutively, and it is found that the amplitude of the residual vector of position error is about 1.2 mm in eight of the times, exceeding the threshold of 1.0 mm, and the direction is concentrated in the positive X-axis direction, and the number of times is 8, then it is considered that there is an abnormal change in spatial position. The abnormal change characteristic parameters include an amplitude of 1.2 mm, a direction of positive X-axis unit vector, and a number of times of occurrence of 8.

[0120] By combining the abnormal change characteristics with the current position deviation data, abnormal change characteristic data of fruit position and posture are generated;

[0121] Anomaly change features are recorded together with current position deviation data in a unified data structure to form fruit position and posture anomaly change feature data, including: target fruit number, current position deviation data (including position deviation vector and posture deviation vector), anomaly change feature amplitude, anomaly change feature direction unit vector, anomaly change feature duration count, timestamp, and coordinate system identifier. For example, for apple number 3, in the current picking action, the current position deviation vector is (2.0, -1.0, -2.0) mm, and the posture deviation vector is (0, 3, 1) degrees. After threshold discrimination and cluster analysis, the anomaly change feature amplitude is 1.2 mm, the direction unit vector is in the positive X-axis direction, and the duration count is 8. All the above information together form the fruit position and posture anomaly change feature data frame, which is used for control parameter correction and real-time path adjustment.

[0122] S6: Based on the abnormal change characteristic data and the corrected robot picking path, adjust the control parameters and path planning of the robot's next picking action in real time, including:

[0123] Based on the abnormal change characteristic data, calculate the control parameter correction amount for the next target picking action;

[0124] Based on the abnormal change characteristic data, the control parameter correction amount for the next target picking action is calculated. That is, according to the degree and direction of the abnormal change, the execution parameters of the robot picking action are dynamically adjusted in real time to reduce the probability of abnormal changes in position and posture during the picking action, ensure the picking process is stable and reliable, and reduce the risk of damage to the fruit and the robotic arm.

[0125] The control parameter correction includes three control parameters: clamping force correction, approach speed correction, and clamping time correction. These are determined based on the amplitude, direction, and duration of abnormal changes in the characteristic data, and are generated through function mapping, proportional relationships, or table lookup.

[0126] The calculation method for the clamping force correction is based on the amplitude and frequency of abnormal change characteristics. The initial clamping force is usually set between 0.5 Newtons and 5 Newtons. If the amplitude of the abnormal change characteristic is large and the frequency is high, it indicates that the robot is causing fruit position displacement or abnormal posture due to insufficient or excessive clamping force. The clamping force should be increased or decreased according to the actual situation. The calculation method is to set a clamping force correction coefficient. If the amplitude of the abnormal change characteristic is greater than the set threshold and the direction is pointing towards the positive X-axis or a weak area connected to the fruit stem, the clamping force is increased by 5% to 15%. If the amplitude indicates that the clamping force is too large and causes the fruit to deflect, the clamping force is decreased by 5% to 15%. For example, if in the apple picking action number 3, the amplitude of the abnormal change characteristic reaches 1.2 mm multiple times and the direction of the abnormal change characteristic is pointing towards a weak area of ​​the fruit stem, then the clamping force is increased by 10% from the initial setting of 2 Newtons, that is, increased by 0.2 Newtons, and the corrected clamping force is 2.2 Newtons.

[0127] The approach speed correction is determined based on the amplitude and direction of the abnormal change characteristics. The initial approach speed is typically set between 10 mm / s and 100 mm / s. If the amplitude of the abnormal change characteristics is large and the direction points to the direction that causes a large positional deviation when the robotic arm approaches the fruit, it indicates that the robotic arm's approach speed may be too fast, causing fruit collision or abnormal fruit stem position. In this case, the approach speed is reduced by 10% to 30%. If the fruit shows a positional deviation but the direction is unrelated to the approach speed, no adjustment is needed. For example, if the abnormal change characteristics indicate that the robotic arm collided with the fruit during five consecutive picking actions, with an amplitude of 1.0 mm to 1.5 mm and a direction in the direction of the robotic arm's approach, the initial approach speed of 50 mm / s needs to be reduced by 20%, resulting in a corrected approach speed of 40 mm / s to reduce the probability of abnormal positional deviations when approaching the fruit.

[0128] The clamping time correction is calculated based on the number of occurrences of abnormal changes and the characteristics of posture deviation. The clamping time is usually set between 1 and 5 seconds. If the number of occurrences of abnormal changes is high and mainly manifests in posture deviation, it indicates that the clamping time may be too short, causing the fruit to start moving before it is stable, resulting in abnormal posture. Therefore, the clamping time should be increased by 10% to 25%. For example, if there are 7 consecutive abnormal posture deviations with an amplitude exceeding the threshold of 3 degrees, the clamping time is increased by 15% from the initial 2 seconds, resulting in a corrected clamping time of 2.3 seconds to ensure fruit stability.

[0129] When abnormal change data indicates that both the approach speed and clamping force need to be reduced simultaneously, first calculate the extent to which the clamping action time is prolonged due to the reduced approach speed, and analyze the stress response characteristics of the fruit stalk during the prolonged time, thereby making corresponding secondary corrections to the clamping force:

[0130] A coupling model is established between clamping force, approach velocity, and stress transmission in the fruit stalk, specifically as follows:

[0131] First, the stress distribution of the fruit stalk structure under different clamping forces and application times was simulated and analyzed using the finite element method. The characteristic curve of stress accumulation in the fruit stalk was obtained, and the variation law of stress with increasing application time under different clamping forces was determined. It shows that when the clamping force is small, the stress increases linearly or slowly with increasing application time, while when the clamping force is large, it shows a non-linear accelerating growth trend. For example, under the condition of a clamping force of 2 Newtons and an application time of 1 second, the maximum stress in the fruit stalk is 2 MPa. After the clamping time is extended to 1.5 seconds, the stress may reach 2.8 MPa, an increase of 40%. When the clamping force is 1.8 Newtons, the stress growth trend slows down, and the stress may only increase to 2.4 MPa, an increase of 20%. These laws are used to determine the sensitive relationship between the magnitude of the clamping force and the extension of the application time.

[0132] In actual harvesting operations, when the robot's approach speed decreases from an initial value (e.g., 50 mm / s) to a corrected value (e.g., 40 mm / s), the time required for the robotic arm's end effector to travel from the safe distance to the gripping position will increase accordingly. For example, if the initial approach distance is 100 mm and the time is 2 seconds, it will be extended to 2.5 seconds after correction. Therefore, the stabilization time window before the gripping action begins decreases, requiring an increase in the action time of the gripping action to ensure gripping stability.

[0133] Based on the above coupling effect analysis, once it is determined that the approach speed needs to be reduced (e.g., by 20%), the resulting increase in the clamping force duration (e.g., 0.5 seconds) is calculated, and combined with the fruit stalk stress characteristic curve, it is determined whether the clamping force needs to be fine-tuned.

[0134] Taking apple picking as an example, assume that the abnormal change characteristics continuously indicate an amplitude of 1.2 mm, a direction pointing in the positive X-axis, and that the robotic arm's approach speed is too fast, causing an abnormal fruit collision. The initial clamping force is 2 Newtons, and the initial approach speed is 50 mm / s.

[0135] If the approach speed is adjusted solely, decreasing it by 20% to 40 mm / s results in a 0.5-second extension of the clamping time (from 2 seconds to 2.5 seconds). Based on the above fruit stalk stress characteristic curve, with the clamping force remaining at 2 Newtons, the stress increases from 2 MPa to 2.8 MPa. This excessive increase will lead to a more serious risk of stress transmission through the fruit stalk. Therefore, a secondary adjustment to the clamping force is considered, reducing it by 10% (i.e., a reduction of 0.2 Newtons, resulting in a clamping force of 1.8 Newtons). After reducing the clamping force, the stress increase decreases from 40% to 20%, effectively mitigating the stress accumulation effect caused by the extended clamping time.

[0136] The calculated clamping force correction, approach speed correction, and clamping time correction are used to form control parameter correction data, which are uniformly recorded in the control parameter correction data frame. The data frame fields include target fruit number, clamping force correction (Newtons), approach speed correction (millimeters per second), clamping time correction (seconds), timestamp, and coordinate system identifier.

[0137] The weights of the position offset risk index in the multi-objective optimization function are updated based on the control parameter correction.

[0138] The multi-objective optimization function is a mathematical expression used for evaluating the continuous harvesting path of the harvesting robot. It includes three indicators: the movement distance of the robotic arm, the change in the posture of the robotic arm, and the position offset risk indicator. Each indicator corresponds to a certain weight coefficient, and the indicators are combined linearly or nonlinearly to form a comprehensive evaluation value of the path.

[0139] The positional displacement risk index represents the risk that harvested fruit may shift position due to the mechanical transmission effect of the fruit stalk. The initial weighting coefficient is set between 0.2 and 0.6. The initial weighting coefficient is determined based on the differences in sensitivity to positional displacement risk caused by characteristics such as fruit firmness, stalk length, stalk stiffness, and fruit cluster density in different harvesting scenarios.

[0140] Regarding fruit firmness, the lower the fruit firmness, the higher the sensitivity to the clamping force of the fruit stem. The more pronounced the stress transmission effect on the fruit stem during clamping, the greater the risk of position displacement. Therefore, the initial weighting coefficient of the position displacement risk index needs to be set in a relatively high range (e.g., 0.4 to 0.6). For example, when targeting fruits with low firmness such as grapes or strawberries, the initial weighting coefficient of the position displacement risk index can be set to 0.5 to fully reflect the position displacement risk caused by the clamping action; for fruits with higher firmness such as apples or pears, the initial weighting of the position displacement risk index can be set lower, such as between 0.2 and 0.3.

[0141] Regarding the length of the fruit stalk, longer stalks generally have greater freedom of movement, resulting in more complex stress transmission paths and larger fruit position displacement after mechanical force application. Therefore, the weighting coefficient of the position displacement risk index should be increased. For apple stalks with a length exceeding 30 mm, the weighting coefficient of the position displacement risk index can be set to 0.4 to 0.6. For fruits with shorter stalks (such as those less than 20 mm in length), the weighting coefficient of the position displacement risk index can be reduced to between 0.2 and 0.3.

[0142] Regarding the stiffness of the fruit stalk, stalks with lower stiffness are more prone to deformation during clamping, leading to positional displacement. Therefore, the weighting coefficient for such fruits needs to be appropriately increased (e.g., 0.5 to 0.6). Conversely, stalks with higher stiffness have a stronger resistance to deformation and a lower risk of positional displacement, so the weighting coefficient can be set in a lower range (e.g., 0.2 to 0.3).

[0143] Regarding fruit cluster density, higher cluster density means smaller spatial distances between adjacent fruits, making it easier for robotic arm movements to cause positional shifts and trigger a chain reaction, leading to the risk of continuous positional shifts. Therefore, the weighting coefficient for the positional shift risk index of high-density fruit clusters (such as grape or cherry tomato clusters) should be appropriately increased to 0.4 to 0.6, while for more dispersed fruit clusters (such as sparsely arranged apple clusters), the weighting coefficient can be reduced to the range of 0.2 to 0.3.

[0144] For example, when picking strawberries, the risk of position displacement is extremely high because strawberries are low in hardness, have long and soft stems, and have high cluster density. Therefore, the weighting coefficient for the position displacement risk index should be set to 0.6. When picking apples, the hardness is high, the stems are moderately rigid, the stems are relatively short, and the clusters are relatively dispersed. The weighting coefficient for the position displacement risk index can be set to 0.25.

[0145] The position offset risk index weight is updated based on the control parameter correction amount. Specifically, when the changes in the clamping force correction amount, approach speed correction amount, or clamping time correction amount in the control parameter correction amount exceed the preset threshold, it indicates that the current position offset risk is high or low. The position offset risk index weight should be dynamically adjusted in real time to reflect the changes in the fruit position offset risk during the actual harvesting process and optimize the robot path planning.

[0146] The update method maps the clamping force correction, approach speed correction, and clamping time correction to the position displacement risk index weights using proportional coefficients. For example, for every 0.1 Newton increase in the clamping force correction, the position displacement risk index weight coefficient increases by 0.05; for every 5 mm / s decrease in the approach speed correction, the weight coefficient increases by 0.05; and for every 0.1 second increase in the clamping time correction, the weight coefficient increases by 0.05. Taking the apple (number 3) example, an increase of 0.2 Newtons in clamping force (weight increase of 0.1), a decrease of 10 mm / s in approach speed (weight increase of 0.1), and an increase of 0.3 seconds in clamping time (weight increase of 0.15) result in a total weight coefficient increase of 0.35. Since the original position displacement risk index weight coefficient was 0.25, the updated position displacement risk index weight coefficient is 0.6, reflecting the increased risk of fruit position displacement.

[0147] The updated weight coefficients are recorded in real time in the multi-objective optimization function parameter data table. The data table includes the weight coefficients of the position offset risk index, the weight coefficients of the robotic arm movement distance, the weight coefficients of the robotic arm posture change, the update timestamp, and the fruit number, ensuring that the updated multi-objective optimization function is used for real-time dynamic replanning and execution of the robot's picking path.

[0148] The updated multi-objective optimization function is used to dynamically replan the corrected robot harvesting path to obtain the adjusted robot harvesting path.

[0149] Dynamic replanning first extracts information about unexecuted picking path nodes from the revised robot picking path, including the node's spatial coordinates, posture angle data, and corresponding fruit numbers. The path optimization algorithm is then re-executed to reorder and optimize the path nodes. The execution process of the path optimization algorithm is as follows: for each unpicked path node, a comprehensive evaluation value of a multi-objective optimization function is calculated. Using updated weight coefficients for position offset risk, robotic arm movement distance, and posture change, the index value for each node is calculated separately and combined with the path planning results of adjacent nodes to determine the overall optimality of the node sequence. For example, the dynamic replanning process is illustrated using the path nodes for unpicked fruits numbered 4, 5, and 6 in a cluster of apples. In the corrected path node sequence, the spatial coordinates of fruit number 4 are (400.0, 250.0, 100.0) mm, fruit number 5 is (420.0, 240.0, 110.0) mm, and fruit number 6 is (430.0, 255.0, 105.0) mm. The initial corrected picking order is fruit number 4-fruit number 5-fruit number 6. After updating, the weight coefficient of the position offset risk index increased from 0.25 to 0.6, while the weight coefficients of the robotic arm movement distance and attitude change were each 0.2. The path optimization algorithm recalculates the index values ​​for each path. For example, starting from the current robot end effector position (380.0, 240.0, 120.0) mm, it calculates the robotic arm movement distance, attitude change, and position offset risk index for fruit numbers 4, 5, and 6 respectively, and calculates the comprehensive evaluation value for each fruit node based on the updated weights. Assuming the calculation results are: the sum of the comprehensive evaluation values ​​corresponding to the initial sequence numbered 4-5-6 is 80, and the sum of the comprehensive evaluation values ​​for the sequence numbered 5-6-4 is 60, which is smaller, the new node picking sequence is determined to be numbered 5-6-4. This method achieves dynamic replanning, and the adjusted robot picking path has the minimum positional offset risk, the most reasonable robotic arm movement distance, and the optimal robotic arm posture change characteristics.

[0150] The replanned path results are recorded in the updated picking path data table. The data table includes the path node number, the optimized picking order, the target space coordinates corresponding to each node, the posture data of the robotic arm end effector, and the comprehensive evaluation value of the path optimization, so that the robot control system can call and execute it in real time.

[0151] Based on the control parameter correction and the adjusted robot picking path, the control parameters and path planning for the robot's next picking action are adjusted in real time.

[0152] Dynamic replanning yields an adjusted robot harvesting path and control parameter corrections, which are used to adjust the robot's execution method for the next harvesting action in real time. The control parameter corrections include clamping force correction, approach speed correction, and clamping time correction. The adjusted harvesting path then determines the spatial coordinates and orientation angle of the target fruit for the next harvesting action.

[0153] Real-time adjustment methods include real-time adjustment of control parameters and real-time fine-tuning of path planning. Real-time adjustment of control parameters means the robot control system updates the clamping force, approach speed, and clamping time based on the data from the control parameter correction. Real-time adjustment of the clamping force is accomplished by the force feedback control unit inside the robot's clamping mechanism. This unit measures the force on the fruit handle during clamping in real time and dynamically adjusts the actual clamping force output by the actuator to ensure a precise match between the actual clamping force and the corrected target value. For example, taking apple number 5 as an example, the initial clamping force is 2 Newtons, and the clamping force correction is an increase of 0.2 Newtons. During real-time adjustment, the clamping actuator continuously adjusts the actual output torque based on the real-time measurement data from the force feedback sensor, maintaining the actual clamping force within the set range of approximately 2.2 Newtons.

[0154] The approach speed is adjusted in real time by dynamically regulating the motor drive current or control voltage through the robotic arm speed controller, ensuring that the actual movement speed of the robotic arm's end effector when approaching the fruit matches the corrected speed setting value. The gripping time is adjusted in real time by setting the holding time of the gripping action in real time through the timing unit within the robot control system, precisely controlling the duration of the gripping action.

[0155] Real-time path planning adjustment refers to the robot's harvesting control system performing path tracking and fine-tuning control based on the adjusted harvesting path. The real-time adjustment method includes the path tracking controller performing real-time error compensation based on the actual execution position measured in real-time by the position feedback sensor and the adjusted planned path position. It also fine-tunes the position and speed of each joint of the robotic arm to ensure that the actual motion path precisely matches the planned path, eliminating path execution deviations. For example, when actually harvesting apple number 5, the real-time feedback of the actual position of the robotic arm's end effector is (419.0, 240.5, 109.5) mm, which slightly deviates from the planned position of (420.0, 240.0, 110.0) mm. After calculating the position deviation, the real-time path tracking controller sends fine-tuning commands to each joint motor for minor adjustments, correcting the end effector position to the planned position in real-time, thereby achieving precise path execution.

[0156] The above real-time adjustment data is recorded in the real-time control log data table in the robot control system. The data table includes the real-time adjusted clamping force, approach speed, clamping time, and the real-time execution position coordinates and attitude angles of the robot's picking path, ensuring that the robot's picking action is stable, efficient, accurate and reliable.

[0157] To more clearly and completely illustrate the data correlation and logical coherence between steps involved in the control method of the harvesting robot of the present invention, the following data table ( Figure 2 The "Robot Fruit Picking Action Control Parameters and Path Planning Data Table" discloses all key data involved in the robot's fruit picking action. The meaning of each column in the table is as follows:

[0158] Picking action number: indicates the sequence of consecutive picking actions performed by the robot;

[0159] Target Fruit Number: Represents the number of the target fruit harvested each time;

[0160] Initial position coordinates and attitude angles: These represent the target grasping position and attitude calculated by the robot during initial path planning;

[0161] Corrected position coordinates and attitude angles: These represent the target grasping position and attitude dynamically corrected based on the fruit stalk stress transfer path data.

[0162] Clamping force, approach speed, and clamping time: Real-time corrected control parameters calculated based on abnormal change characteristic data.

[0163] 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 and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0164] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0165] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in 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. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0166] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0167] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0168] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0169] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0170] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0171] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations 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. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0172] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A control method of a picking robot, characterized by, Comprise the following steps: S1: Real-time acquisition of three-dimensional point cloud data of all fruits in the fruit cluster area, and analyze the spatial position relationship and fruit stalk connection structure of the fruits, output the spatial connection structure data of all fruit stalks in the fruit cluster; S2: Based on the spatial connection structure data of all fruit stalks in the fruit cluster, analyze the stress transmission path after the picking action acts on the fruit stalk of the target fruit, predict the position deviation of the adjacent un-picked fruit, and output the fruit stalk stress transmission path data; S3: Based on the spatial connection structure data of all fruit stalks in the fruit cluster, plan the initial path sequence of the picking robot for continuous picking of fruits, and output the initial picking path of the robot; S4: Based on the fruit stalk stress transmission path data and the initial picking path of the robot, dynamically correct the spatial position of the next fruit picking action, and output the corrected picking path of the robot; S5: Based on the actual position deviation data after each picking action of the robot is executed, identify the abnormal change characteristics of the fruit position and attitude, output the abnormal change characteristics data of the fruit position and attitude, specifically: Obtain the actual grasping pose data after the current target fruit is picked and the target grasping pose data recorded in the corrected picking path of the robot for the current target fruit; Calculate the three-dimensional position deviation vector and attitude angle deviation vector between the actual grasping pose data and the target grasping pose data to form the current position deviation data; Differential operation is performed on the current position deviation data and the spatial predicted deviation in the fruit stalk stress transmission path data for the current target fruit, and the fruit position error residual vector is obtained; Threshold discrimination and cluster analysis are performed on the fruit position error residual vector to extract the abnormal change characteristics of the fruit position and attitude; Combine the abnormal change characteristics and the current position deviation data to generate the abnormal change characteristics data of the fruit position and attitude; S6: According to the abnormal change characteristic data and the corrected picking path of the robot, real-time adjust the control parameters and path planning of the next picking action of the robot.

2. The control method of a picking robot according to claim 1, characterized in that, S1, specifically: Real-time acquisition of three-dimensional point cloud data of all fruits in the fruit cluster area, and pre-processing to obtain pre-processed three-dimensional point cloud data; Based on the pre-processed three-dimensional point cloud data, identify the spatial position of the fruits in the fruit cluster area and the geometric shape of the fruits, and analyze the spatial position relationship between the fruits in the fruit cluster area and the fruit stalk connection structure characteristics; Based on the spatial position relationship between the fruits in the fruit cluster area and the fruit stalk connection structure characteristics, establish the spatial connection topology between all fruit stalks in the fruit cluster area, and output the spatial connection structure data of all fruit stalks in the fruit cluster area.

3. The control method of a picking robot according to claim 2, characterized in that, S2, specifically: Based on the spatial connection structure data of all fruit stalks in the fruit cluster area, establish a finite element mechanics analysis model of the fruit stalk connection structure in the fruit cluster area; Based on the finite element mechanics analysis model, analyze the torque and stress transmission path generated when the picking robot clamps the fruit stalk of the target fruit during picking; Based on the torque and stress transmission path, predict the position deviation of the adjacent un-picked fruit due to the stress transmission effect of the fruit stalk connection structure; According to the position deviation of the adjacent un-picked fruit, output the fruit stalk stress transmission path data.

4. The control method of a picking robot according to claim 3, characterized in that, S3, specifically: Based on the spatial connection structure data of all fruit stalks in the fruit cluster region, a multi-objective optimization function including mechanical arm movement distance, mechanical arm attitude change amount, and position offset risk index is constructed; Under the constraints of mechanical arm joint movement range, fruit spatial collision avoidance, and continuous picking sequence, a multi-objective path optimization algorithm is used to generate a plurality of candidate continuous picking path sequences; The comprehensive evaluation value of the multi-objective optimization function is calculated for each candidate continuous picking path sequence; The candidate continuous picking path sequence with the optimal comprehensive evaluation value is determined as the initial picking path of the robot.

5. The control method of a picking robot according to claim 4, characterized in that, S4, specifically: For the next target picking action in the initial picking path of the robot, the spatial prediction offset of the corresponding target fruit is extracted from the fruit stalk stress transfer path data; The target grasping pose of the next target fruit at the picking time is calculated based on the spatial prediction offset, and a path correction amount is generated; The initial picking path of the robot is incrementally updated using the path correction amount to form an updated path node sequence; The updated path node sequence is checked for continuity and reachability to generate a corrected robot picking path that satisfies the kinematic constraints of the mechanical arm.

6. The control method of a picking robot according to claim 5, wherein S6, specifically: Based on the abnormal change feature data, the control parameter correction amount of the next target picking action is calculated; Based on the control parameter correction amount, the position offset risk index weight in the multi-objective optimization function is updated; The updated multi-objective optimization function is used to dynamically re-plan the corrected robot picking path to obtain an adjusted robot picking path; Based on the control parameter correction amount and the adjusted robot picking path, the control parameters and path planning of the next picking action of the robot are adjusted in real time.

Images