Robot path planning method and system in complex environment and medium

Abstract

The invention discloses a robot path planning method and system in a complex environment and a medium, and belongs to the technical field of path planning. According to the invention, based on the point cloud classification model, the classification and identification capability of complex point cloud data in an orchard environment is significantly improved, and fruit trees, static obstacles and dynamic obstacles can be accurately identified; on the basis, global and local path planning strategies are combined: an improved sparrow search algorithm is adopted to carry out global path planning, and an optimal navigation path is rapidly generated; in a high-density fruit tree area and a static obstacle, a local path is optimized by improving an artificial potential field method, and the obstacle avoidance effect is ensured; as for dynamic obstacles, real-time obstacle avoidance is achieved through an improved dynamic window method, and the motion trail of the robot is dynamically adjusted; the path planning efficiency and the environment adaptability of the robot in the complex orchard environment are effectively improved, it is ensured that automatic operation is efficiently and safely completed, and the robot is suitable for the field of modern orchard management and intelligent agriculture.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular, to a method, system and medium for robot path planning in a complex environment. Background Art

[0002] With the rapid development of deep learning technology, especially the breakthroughs in point cloud data processing and computer vision fields, deep learning has shown great potential in path planning in complex environments. The original deep learning model PointNet can automatically extract environmental features from point cloud data through a neural network and perform object classification, providing more accurate environmental perception. However, the original PointNet model mainly relies on the global features of the point cloud and fails to fully consider the local geometric relationships between points. In an orchard environment, the density of fruit trees varies greatly, and traditional models often have difficulty accurately identifying in high-density fruit tree areas and low-density areas, resulting in sub-optimal path planning results. In addition, in the face of dynamic obstacles, the real-time response ability of traditional models is also relatively limited, and they cannot efficiently handle the constantly changing environment, affecting the real-time performance and adaptability of path planning.

[0003] In terms of path planning, existing path planning algorithms can provide good results in static environments, but often perform poorly in dynamic environments. Traditional path planning methods have many limitations in complex environments such as orchards. Traditional algorithms often cannot effectively avoid collisions, resulting in uneven paths or low navigation efficiency. In addition, traditional methods lack the ability to adapt to real-time changes in the environment and are difficult to perform efficient dynamic obstacle avoidance and path adjustment in complex and constantly changing orchard environments. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method, system and medium for robot path planning in a complex environment, which can achieve more efficient and safe navigation in an orchard environment and improve the adaptability and reaction speed of the robot to dynamic environments.

[0005] To achieve the above purpose, the present invention is implemented by the following technical solutions:

[0006] In the first aspect, the present invention provides a method for robot path planning in a complex environment, including:

[0007] Obtaining point cloud data of the orchard environment around the robot and the starting point and ending point of the robot path planning;

[0008] Inputting the point cloud data into a pre-constructed point cloud classification model for target classification and recognition to obtain target categories; the target categories include fruit trees, static obstacles and dynamic obstacles;

[0009] Divide fruit trees into high-density fruit trees and low-density fruit trees according to the point cloud density of the fruit trees;

[0010] Perform path planning for the robot according to the identified high-density fruit trees, low-density fruit trees, static obstacles or dynamic obstacles, so that the robot moves from the starting point of the path planning to the ending point of the path planning;

[0011] The path planning includes:

[0012] For low-density fruit trees, use the improved sparrow search algorithm to calculate the global optimal path from the starting point to the ending point of the robot path planning, and the robot moves according to the global optimal path; in the sparrow search algorithm, dynamically adjust the proportion of explorers using the number of iterations, and optimize the fitness value of sparrow individuals according to the path length, smoothness and obstacle avoidance performance to obtain the improved sparrow search algorithm;

[0013] For high-density fruit trees or static obstacles, use the improved artificial potential field method to calculate the local optimal path of the robot, and the robot moves according to the local optimal path; in the artificial potential field method, optimize the synthetic force using the point cloud density of the fruit trees for high-density fruit trees and the preset density of the static obstacle for static obstacles to obtain the improved artificial potential field method;

[0014] For dynamic obstacles, use the improved dynamic window method to dynamically adjust the motion trajectory of the robot; in the dynamic window method, optimize the range of the dynamic window using the kinematic constraints of the robot to obtain the improved dynamic window method.

[0015] Optionally, the construction of the point cloud classification model includes:

[0016] Obtain the original point cloud deep learning model;

[0017] Add a geometric feature attention module between the input layer and the feature extraction layer of the original point cloud deep learning model to obtain the point cloud classification model.

[0018] Optionally, the processing steps of the point cloud classification model include:

[0019] Calculate the local geometric features of each point in the point cloud data, and generate the attention weights corresponding to the local geometric features of each point;

[0020] Weight the attention weights corresponding to the local geometric features of each point to the local geometric features of each point to obtain the weighted geometric features of each point;

[0021] Perform global geometric feature transformation on the weighted geometric features of each point to obtain the transformed geometric features;

[0022] Classify or segment the transformed geometric features to obtain the target category.

[0023] Optionally, input the point cloud data into a pre-constructed point cloud classification model for target classification and recognition to obtain the target category, including:

[0024] ;

[0025] ;

[0026] ;

[0027] ;

[0028] Perform a global geometric feature transformation on the weighted geometric feature of each point to obtain the transformed geometric feature ;

[0029] ;

[0030] wherein, represents the set of local neighborhood points of each point in the point cloud data; represents the i-th point in the point cloud data; represents the j-th local neighborhood point of each point in the point cloud data; represents from point to its neighborhood point the maximum distance; represents the local geometric feature of the i-th point in the point cloud data; represents the attention weight corresponding to the local geometric feature of the i-th point in the point cloud data; represents the MLP layer; represents the probability of the target category; represents the softmax function; represents the weighted geometric feature represents the transformed geometric feature; represents the norm of the vector.

[0031] Optionally, in the sparrow search algorithm, dynamically adjust the proportion of explorers using the number of iterations, and optimize the fitness value of sparrow individuals according to the path length, smoothness, and obstacle avoidance performance to obtain an improved sparrow search algorithm, including:

[0032] ;

[0033] ;

[0034] wherein, represents the proportion of explorers at the t-th iteration; , respectively represent the initial minimum proportion of explorers and the maximum proportion of explorers; represents the maximum number of iterations; represents the explorer ratio adjustment factor; represents the fitness value function of the sparrow individual; , , represents the weight coefficient; , , respectively represent the normalization functions of path length, smoothness, and obstacle performance.

[0035] Optionally, in the artificial potential field method, for high-density fruit trees, the point cloud density of the fruit trees is used to optimize the synthetic force, and for static obstacles, the preset density of the static obstacles is used to optimize the synthetic force, resulting in an improved artificial potential field method, including:

[0036] ;

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] Among them, represents the target attraction; represents the target attraction constant; represents the target position; represents the current position of the robot; represents the repulsive force of the high-density fruit tree; represents the repulsive force constant; represents a constant; represents the distance between the high-density fruit tree and the robot; represents the point cloud density of the high-density fruit tree; represents the synthetic force of the high-density fruit tree; represents the repulsive force of the nth static obstacle; represents the distance between the nth static obstacle and the robot; represents the preset density of the nth static obstacle; represents the synthetic force of the static obstacle.

[0042] Optionally, in the dynamic window method, the kinematic constraints of the robot are used to optimize the range of the dynamic window, resulting in an improved dynamic window method, including:

[0043] ;

[0044] Among them, , respectively represent the minimum and maximum values of the x-axis speed of the robot; , respectively represent the minimum and maximum values of the y-axis speed of the robot; , respectively represent the minimum and maximum values of the angular velocity of the robot; , respectively represent the x-axis and y-axis speed components of the robot; , respectively represent the maximum accelerations of the robot along the x-axis and y-axis; represents the time step; , respectively represent the preset minimum speed values on the x and y axes; , respectively represent the preset maximum speed values on the x and y axes; represents the angular velocity of the robot; represents the maximum angular acceleration of the robot; , respectively represent taking the maximum or minimum value within the parentheses.

[0045] Optionally, for dynamic obstacles, the improved dynamic window method is adopted to dynamically adjust the motion trajectory of the robot, including:

[0046] Based on each group of speeds within the dynamic window and the robot motion model, predict the motion trajectory of the robot, expressed as:

[0047] ;

[0048] Use the scoring function to score the motion trajectory of the robot, and the scoring function is expressed as:

[0049] ;

[0050] Take the speed and angular velocity corresponding to the motion trajectory with the highest score of the robot as the motion speed of the robot;

[0051] Among them, , respectively represent the new x and y coordinates of the robot after the time step ; represents the new orientation angle of the robot after the time step ; , respectively represent the x and y coordinates of the robot at the current moment; represents the orientation angle of the robot at the current moment; , respectively represent the x-axis and y-axis velocity components of the robot; represents the angular velocity of the robot; represents the time step; , , , respectively represent the weight coefficients corresponding to the target distance score, dynamic obstacle avoidance score, speed smoothness score, and safety boundary score; , , , respectively represent the target distance score, dynamic obstacle avoidance score, speed smoothness score, and safety boundary score.

[0052] In a second aspect, the present invention provides a computer system, including a processor and a storage medium;

[0053] The storage medium is used to store instructions;

[0054] The processor is used to operate according to the instructions to execute the method described in the first aspect.

[0055] In a third aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and when the computer instructions are executed by a processor, the steps of the method described in the first aspect are implemented.

[0056] Compared with the prior art, the beneficial effects achieved by the present invention:

[0057] Based on the point cloud classification model, the present invention significantly improves the classification and recognition ability of complex point cloud data in the orchard environment, and can accurately identify fruit trees, static obstacles, and dynamic obstacles; on this basis, combining global and local path planning strategies: using an improved sparrow search algorithm for global path planning to quickly generate the optimal navigation path; in the high-density fruit tree area and static obstacles, optimizing the local path through an improved artificial potential field method to ensure the obstacle avoidance effect; for dynamic obstacles, using an improved dynamic window method to achieve real-time obstacle avoidance and dynamically adjust the robot's motion trajectory; the present invention effectively improves the path planning efficiency and environmental adaptability of the robot in the complex orchard environment, ensures efficient and safe completion of automated operations, and is applicable to the fields of modern orchard management and intelligent agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 shows a schematic overview of the process of the robot path planning method in a complex environment according to the present invention in an embodiment;

[0059] Figure 2 shows a schematic diagram of the process details of the robot path planning method in a complex environment according to the present invention in an embodiment. Detailed Implementation Manner

[0060] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0061] The term "and / or" is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " generally represents an "or" relationship between the associated objects before and after.

[0062] Embodiment 1

[0063] As Figure 1 shown, this embodiment introduces a robot path planning method in a complex environment, which specifically includes the following steps:

[0064] Step 1: Input the starting point, and perform global path planning through ISSA. Specifically:

[0065] Obtain the current position of the robot and the position of the target point through GPS, determine the starting point and the ending point of the robot path planning; adopt the improved sparrow search algorithm to calculate the global optimal path from the starting point to the ending point of the robot path planning, and the robot moves along the global optimal path to generate an initial navigation path, providing a global navigation reference for the robot. In this embodiment, the improved sparrow search algorithm (Improving Sparrow Search Algorithm, ISSA) is used for global path planning.

[0066] In the sparrow search algorithm, the proportion of explorers is dynamically adjusted using the number of iterations, and the fitness value of sparrow individuals is optimized according to the path length, smoothness, and obstacle avoidance performance, resulting in the improved sparrow search algorithm, that is:

[0067] Initialize parameters and population: mainly including and respectively represent the initial maximum explorer proportion and the minimum explorer proportion, the initial iteration t = 0, represents the maximum number of iterations.

[0068] Update the positions of explorers and followers: The formula for updating the position of explorers is expressed as:

[0069] ;

[0070] Among them, is a dynamic variable in the solution space, representing the specific coordinate of the o-th sparrow in the c-th dimensional space at the (t + 1)-th iteration. represents the specific coordinate of the o-th sparrow in the c-th dimensional space at the t-th iteration, α represents a random number whose value range is between [0, 1], Q is a random number and follows a normal distribution, the randomness enhances the global exploration ability, L is a randomly generated vector used to simulate random perturbations, all elements of which are 1, W represents the warning value whose value range is [0, 1], and SV is used to represent the safety value whose value range is [0.5, 1]; represents the exponential function.

[0071] The formula for updating the position of followers is expressed as:

[0072] ;

[0073] where, represents the position of the individual with the worst fitness in the current population, that is, the individual with the lowest fitness value. represents the best position where the explorer is located at the (t + 1)-th iteration; represents a binary random number with values of 0 or 1, used to control whether the follower performs position update, and m represents the initial number of sparrows.

[0074] Adjust the explorer ratio adaptively according to the number of iterations: In order to balance the global search and local search capabilities in different stages, namely the early stage and the late stage, the explorer ratio can be dynamically adjusted according to the number of iterations.

[0075] Explorer ratio decreases non-linearly with the number of iterations t, and is expressed as:

[0076] ;

[0077] where, and represent the initial minimum explorer ratio and the maximum explorer ratio respectively; represents the maximum number of iterations; represents the explorer ratio adjustment factor, which controls the decay rate of the discoverer ratio.

[0078] This improvement is to prevent the algorithm from falling into the local optimum prematurely. Therefore, when the number of iterations is low, increase the explorer ratio to allow more individuals to perform global search; when the algorithm is approaching convergence, focus on local fine search. At this time, increase the follower ratio to perform local optimization on the current optimal solution.

[0079] After each iteration, calculate the fitness values of all sparrow individuals in the population, where the sparrow individual fitness value function is expressed as:

[0080] ;

[0081] Among them, , , represent weight coefficients; , , respectively represent the normalization functions of path length, smoothness, and obstacle performance.

[0082] Output the optimized smooth path as the global optimal path planning scheme for robot navigation.

[0083] Step 2: LiDAR generates point cloud data: Start the LiDAR to scan the orchard environment around the robot in real time, generate the point cloud data around the robot, and obtain the spatial distribution information of fruit trees, static obstacles, and dynamic obstacles through the point cloud data as the input environment data for path planning. At the same time, preprocess the point cloud data, that is:

[0084] Point cloud data denoising: Use the radius filtering method to remove the isolated points and noise points in the point cloud to ensure the quality of the point cloud data.

[0085] Point cloud coordinate normalization: Perform coordinate normalization processing on the collected point cloud data, map the spatial coordinates of the point cloud to a unified coordinate range, and facilitate feature extraction by the subsequent deep learning model.

[0086] Ground point segmentation: Separate the ground points from the non-ground points, remove the irrelevant ground points, and improve the effectiveness of the point cloud data.

[0087] Downsampling processing: Use voxel downsampling to reduce the amount of point cloud data, improve the model calculation efficiency while maintaining data representativeness.

[0088] Step 3: Point cloud category discrimination by Pointnet model: Input the point cloud data into a pre-constructed point cloud classification (Pointnet) model for target classification and recognition to obtain the target category, specifically:

[0089] The construction of the point cloud classification model includes:

[0090] Obtain the original point cloud deep learning model;

[0091] Add a geometric feature attention module between the input layer and the feature extraction layer of the original point cloud deep learning model to obtain the point cloud classification model.

[0092] As Figure 2 shown, the target category includes fruit trees, static obstacles, and dynamic obstacles;

[0093] The processing steps of the point cloud classification model include:

[0094] The model describes the geometric shape of the point cloud by calculating the local geometric features of each point. For the i-th point in the point cloud data , its local neighborhood point set is expressed as:

[0095] ;

[0096] Among them, represents the j-th local neighborhood point of each point in the point cloud data; represents the maximum distance from point to its neighborhood point ;

[0097] The Euclidean distance between points is used to represent the local geometric features. The local geometric feature can be obtained by calculating the sum of the squares of the distances between points:

[0098] ;

[0099] Among them, is the square of the Euclidean distance between point and its neighborhood point , reflecting the complexity of the local geometric structure, represents the norm of the vector, which is the two-norm in this embodiment, that is, the Euclidean length or modulus length.

[0100] Generate attention weights related to the geometric features of each point through a multi-layer perceptron (MLP). The attention weight corresponding to the local geometric feature of the i-th point in the point cloud data

[0101] ;

[0102] The attention weight is a scalar, representing the importance of point in the current context. A larger indicates that this point is more critical for the model's decision. The MLP consists of multiple fully connected layers, and each layer performs a linear transformation and then a non-linear transformation through an activation function. The input feature vector , after the calculation of the k-th layer, the output can be expressed as:

[0103] ;

[0104] Among them, is the weight matrix of the k-th layer, is the bias term, is the activation function, is the input feature.

[0105] The attention weights corresponding to the local geometric features of each point are weighted to the local geometric features of each point to obtain the weighted geometric feature of each point , that is ;

[0106] The weighted geometric features of each point are subjected to global geometric feature transformation to obtain the transformed geometric feature ;

[0107] The transformed geometric feature after feature transformation is fed into the MLP layer for classification or segmentation tasks. For the classification task, the model outputs the target probability that the point cloud belongs to a certain class , expressed as:

[0108] ;

[0109] Among them, represents the softmax function.

[0110] This improvement enables the model to better adapt to the point cloud data in complex environments such as orchards. It can not only identify different types of obstacles but also provide more accurate and efficient environmental perception and path planning capabilities. At the same time, by introducing the geometric feature attention module, the model can pay more attention to the regions with higher local geometric complexity when processing point cloud data, thereby enhancing the understanding and adaptation capabilities of the environment.

[0111] Discrimination of fruit tree density. Further analyze the fruit tree area in the classification result output by the point cloud classification model, calculate the fruit tree point cloud density, and based on the characteristics of the point cloud density, discriminate the fruit tree area as a low-density area or a high-density area to provide a basis for subsequent path planning.

[0112] Step 4: Differential path planning: The robot moves according to the path planning, moves from the starting point of the path planning to the ending point of the path planning, and stops moving. Specifically:

[0113] For low-density fruit trees, the robot continues to move according to the global optimal path planning scheme.

[0114] For high-density fruit trees or static obstacles, the improved artificial potential field method is adopted to calculate the local optimal path of the robot. The robot moves along the local optimal path to ensure the obstacle avoidance effect. In the traditional artificial potential field method, the repulsive force of static obstacles is usually only related to the distance between the static obstacle and the robot. In this embodiment, the repulsive force is adjusted by the point cloud density of the fruit tree or the preset density of the obstacle. That is, in the artificial potential field method, for high-density fruit trees, the point cloud density of the fruit tree is used to optimize the synthetic force, and for static obstacles, the preset density of the static obstacle is used to optimize the synthetic force, thus obtaining the improved artificial potential field method.

[0115] First, the target attraction force Remains unchanged and is expressed as:

[0116] ;

[0117] Where Represents the target attraction force constant; Represents the target position; Represents the current position of the robot;

[0118] For high-density fruit trees, the point cloud density of the fruit tree is used to optimize the repulsive force of the high-density fruit tree :

[0119] ;

[0120] Where Represents the repulsive force constant; Represents a constant; Represents the distance between the high-density fruit tree and the robot; Represents the point cloud density of the high-density fruit tree;

[0121] The final synthetic force of the high-density fruit tree Is expressed as:

[0122] ;

[0123] For static obstacles, the preset density of the static obstacle is used to optimize the repulsive force of the static obstacle When implementing obstacle avoidance, a fixed density value can be used. The repulsive force of the nth static obstacle Is expressed as:

[0124] ;

[0125] The final synthetic force of the static obstacle Is expressed as:

[0126] ;

[0127] In this improved artificial potential field method, the point cloud density of fruit trees or the preset density of obstacles is introduced as a factor for adjusting the repulsive force on the basis of the original method, so that the repulsive force of the robot increases in the case of high-density fruit trees or static obstacles, avoiding collisions with static obstacles.

[0128] For dynamic obstacles, the improved dynamic window method is adopted to dynamically adjust the motion trajectory of the robot; in the dynamic window method, the range of the dynamic window is optimized by using the kinematic constraints of the robot to obtain the improved dynamic window method.

[0129] The dynamic window method combines the kinematic constraints of the robot and environmental information, and ensures the safety of obstacle avoidance and the effectiveness of path planning by real-time calculating the optimal motion strategy of the robot in the velocity space.

[0130] First, based on the current speed, maximum acceleration and time step of the robot, calculate the dynamic window range:

[0131] ;

[0132] Among them, 、 respectively represent the minimum and maximum values of the x-axis speed of the robot; 、 respectively represent the minimum and maximum values of the y-axis speed of the robot; 、 respectively represent the minimum and maximum values of the angular velocity of the robot; 、 respectively represent the x-axis and y-axis speed components of the robot; 、 respectively represent the maximum accelerations along the x-axis and y-axis of the robot; represents the time step; 、 respectively represent the preset minimum speed values on the x and y axes; 、 respectively represent the preset maximum speed values on the x and y axes; represents the angular velocity of the robot; represents the maximum angular acceleration of the robot; 、 respectively represent taking the maximum or minimum value within the brackets.

[0133] Next, according to each group of speeds within the dynamic window range, based on the robot motion model, predict the motion trajectory of the robot, expressed as:

[0134] ;

[0135] Among them, 、 respectively represent the new x and y coordinates of the robot after a time step; represents the new orientation angle of the robot after a time step; and respectively represent the x and y coordinates of the robot at the current moment; represents the orientation angle of the robot at the current moment; and respectively represent the x-axis and y-axis velocity components of the robot; represents the angular velocity of the robot; represents the time step;

[0136] Then, use the scoring function to score the motion trajectory of the robot, and the scoring function is expressed as:

[0137] ;

[0138] where and and and respectively represent the weight coefficients corresponding to the target distance score, dynamic obstacle avoidance score, speed smoothness score, and safety boundary score; and and and respectively represent the target distance score, dynamic obstacle avoidance score, speed smoothness score, and safety boundary score.

[0139] Take the speed and angular velocity corresponding to the motion trajectory with the highest score of the robot as the motion speed of the robot;

[0140] where and represent the horizontal speed and vertical speed corresponding to the motion trajectory with the highest score of the robot; represents the angular velocity corresponding to the motion trajectory with the highest score of the robot.

[0141] The robot moves from the starting point along the planned path, avoiding obstacles in real time, and finally reaches the end point safely, completing the path navigation task.

[0142] This embodiment improves the path planning ability of the robot in a complex orchard environment through intelligent discrimination of point cloud data and differential path planning strategies, and is applicable to modern orchard management and automated operations.

[0143] Embodiment 2

[0144] This embodiment introduces a computer system, including a processor and a storage medium;

[0145] The storage medium is used to store instructions;

[0146] The processor is used to operate according to the instructions to execute the method described in Embodiment 1.

[0147] Embodiment 3

[0148] This embodiment introduces a computer-readable storage medium, on which computer instructions are stored. When the computer instructions are executed by a processor, the steps of the method described in Embodiment 1 are implemented.

[0149] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0150] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0151] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0152] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide for implementing in the processFigure 1 one process or multiple processes and / or blocks Figure 1 steps of functions specified in one block or multiple blocks.

[0153] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative rather than restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit of the present invention and the scope protected by the claims. All of these fall within the protection scope of the present invention.

Claims

1. A robot path planning method in a complex environment, characterized in that: include: Obtain point cloud data of the orchard environment around the robot and the starting and ending points of the robot's path planning; Inputting the point cloud data into a pre-built point cloud classification model to perform target classification and recognition to obtain target categories; the target categories include fruit trees, static obstacles, and dynamic obstacles; According to the point cloud density of fruit trees, the fruit trees are divided into high-density fruit trees and low-density fruit trees; The robot is path-planned according to the identified high-density fruit trees, low-density fruit trees, static obstacles or dynamic obstacles, so that the robot moves from the starting point of the path planning to the end point of the path planning; The path planning comprises: For low-density fruit trees, an improved sparrow search algorithm is used to calculate the global optimal path from the starting point to the end point of the robot's path planning, and the robot moves according to the global optimal path; in the sparrow search algorithm, the proportion of explorers is dynamically adjusted using the number of iterations, and the fitness value of the sparrow individual is optimized according to the path length, smoothness and obstacle avoidance performance, thus obtaining an improved sparrow search algorithm; For high-density fruit trees or static obstacles, the improved artificial potential field method is used to calculate the local optimal path of the robot, and the robot moves according to the local optimal path; in the artificial potential field method, the point cloud density of the fruit trees is used to optimize the synthetic force for high-density fruit trees, and the preset density of the static obstacles is used to optimize the synthetic force for static obstacles, thus obtaining the improved artificial potential field method; For dynamic obstacles, the improved dynamic window method is used to dynamically adjust the robot's motion trajectory; in the dynamic window method, the kinematic constraints of the robot are used to optimize the range of the dynamic window to obtain the improved dynamic window method.

2. The robot path planning method in a complex environment according to claim 1, characterized in that: The construction of the point cloud classification model includes: Get the original point cloud deep learning model; A geometric feature attention module is added between the input layer and the feature extraction layer of the original point cloud deep learning model to obtain a point cloud classification model.

3. The robot path planning method in a complex environment according to claim 1, characterized in that: The processing steps of the point cloud classification model include: Calculate the local geometric features of each point in the point cloud data and generate attention weights corresponding to the local geometric features of each point; The attention weight corresponding to the local geometric feature of each point is weighted to the local geometric feature of each point to obtain the weighted geometric feature of each point; Performing a global geometric feature transformation on the weighted geometric feature of each point to obtain a transformed geometric feature; The transformed geometric features are classified or segmented to obtain target categories.

4. The robot path planning method in a complex environment according to claim 1 or 3, characterized in that: The point cloud data is input into a pre-built point cloud classification model to perform target classification and recognition to obtain target categories, including: ； ； ； ； The weighted geometric features of each point Perform global geometric feature transformation to obtain transformed geometric features ; ； in, Represents the local neighborhood point set of each point in the point cloud data; Represents the i-th point in the point cloud data; Represents the jth local neighborhood point of each point in the point cloud data; Indicates from point To its area point The maximum distance; Represents the local geometric features of the i-th point in the point cloud data; Represents the attention weight corresponding to the local geometric features of the i-th point in the point cloud data; represents the MLP layer; represents the probability of the target category; represents the softmax function; Representing weighted geometric features Represents transformation geometric features; Represents the norm of a vector.

5. The robot path planning method in a complex environment according to claim 1, characterized in that: In the sparrow search algorithm, the proportion of explorers is dynamically adjusted by the number of iterations, and the fitness value of the sparrow individuals is optimized according to the path length, smoothness and obstacle avoidance performance. The improved sparrow search algorithm is obtained, including: ； ； in, represents the explorer ratio at the tth iteration; , They represent the initial minimum explorer ratio and the maximum explorer ratio respectively; Indicates the maximum number of iterations; represents the explorer ratio adjustment factor; Represents the fitness value function of individual sparrows; , , represents the weight coefficient; , , They represent the normalized functions of path length, smoothness, and barrier performance, respectively.

6. The robot path planning method in a complex environment according to claim 1, characterized in that: In the artificial potential field method, for high-density fruit trees, the point cloud density of the fruit trees is used to optimize the synthetic force, and for static obstacles, the preset density of the static obstacles is used to optimize the synthetic force, and an improved artificial potential field method is obtained, including: ； ； ； ； ； in, Indicates target attractiveness; represents the target attraction constant; Indicates the target location; Indicates the current position of the robot; It indicates the repulsive force of high density fruit trees; represents the repulsive force constant; represents a constant; represents the distance between high-density fruit trees and the robot; Point cloud density representing high-density fruit trees; It indicates the synthetic force of high density fruit trees; represents the repulsive force of the nth static obstacle; Indicates the distance between the nth static obstacle and the robot; Indicates the preset density of the nth static obstacle; Represents the resultant force of a static obstacle.

7. The robot path planning method in a complex environment according to claim 1, characterized in that: In the dynamic window method, the kinematic constraints of the robot are used to optimize the range of the dynamic window, and an improved dynamic window method is obtained, including: ； in, , Respectively represent the minimum and maximum values ​​of the robot's x-axis speed; , Respectively represent the minimum and maximum values ​​of the robot's y-axis speed; , Respectively represent the minimum and maximum values ​​of the robot’s angular velocity; , Represent the x-axis and y-axis velocity components of the robot respectively; , Represent the maximum acceleration of the robot along the x-axis and y-axis respectively; represents the time step; , Respectively represent the preset minimum speed values ​​on the x and y axes; , Respectively represent the preset maximum speed values ​​on the x and y axes; represents the angular velocity of the robot; represents the maximum angular acceleration of the robot; , It means taking the maximum or minimum value in the brackets respectively.

8. The robot path planning method in a complex environment according to claim 1 or 7, characterized in that: For dynamic obstacles, the improved dynamic window method is used to dynamically adjust the robot's motion trajectory, including: According to each set of velocities within the dynamic window, based on the robot motion model, the robot's motion trajectory is predicted, which is expressed as: ； The robot's motion trajectory is scored using a scoring function. It is expressed as: ； The speed and angular velocity corresponding to the robot's highest-scoring motion trajectory are taken as the robot's motion speed; in, , Represents the time step Then the robot's new x and y coordinates; Indicates the time step The robot's new orientation angle afterwards; , Respectively represent the x and y coordinates of the robot at the current moment; Indicates the robot's orientation angle at the current moment; , Represent the x-axis and y-axis velocity components of the robot respectively; represents the angular velocity of the robot; represents the time step; , , , Respectively represent the weight coefficients corresponding to the target distance score, dynamic obstacle avoidance score, speed smoothness score, and safety margin score; , , , They represent target distance score, dynamic obstacle avoidance score, speed smoothness score, and safety margin score respectively.

9. A computer system, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the steps of the method described in any one of claims 1 to 8 are implemented.

Images