CN118092290B

CN118092290B - Multi-plant vehicle cooperative control system based on path planning

Info

Publication number: CN118092290B
Application number: CN202410481975.4A
Authority: CN
Inventors: 张昊博; 王俊杰; 黄岩军; 邵昱龙; 庄仲元
Original assignee: Tongji University
Current assignee: Tongji University
Priority date: 2024-04-22
Filing date: 2024-04-22
Publication date: 2024-08-02
Anticipated expiration: 2044-04-22
Also published as: CN118092290A

Abstract

The present invention provides a multi-planting vehicle cooperative control system based on path planning, including a positioning subsystem for capturing and filtering each reflective mark point on multiple capture targets in the intended agricultural planting area, and obtaining the three-dimensional spatial coordinates of each reflective mark point in different time measurement units; a region division subsystem for determining the operating range according to the three-dimensional spatial coordinates of the preset boundary points in the three-dimensional spatial coordinates of each reflective mark point, and evenly arranging multiple agricultural robots in multiple convex polygonal planting areas divided by landscape plant fields or special terrain farmlands within the operating range; a decision subsystem for generating an overall decision based on a cultivation recommendation model, a crop growth status monitoring model, a local parameter detection model and a global model; and a path planning subsystem for performing path planning based on multiple convex polygonal planting areas to obtain a fully covered operating path. The present invention solves the problem of insufficient accuracy of empirical models.

Description

Multi-plant vehicle cooperative control system based on path planning

Technical Field

The embodiment of the invention relates to the technical field of agricultural planting, in particular to a cooperative control system of a plurality of planting vehicles based on path planning.

Background

In agricultural planting, an empirical model-based agricultural decision is generally used, which refers to a decision process for guiding agricultural planting management by establishing a mathematical model or rule using past planting experience and data. This decision method relies on past observations and practices to extract rules and patterns by analyzing and summarizing historical data, which are then converted into decision models for guiding future planting activities.

Typically, empirical model-based agronomic decisions take into account a number of factors, such as crop growth cycle, soil type, climatic conditions, pest occurrence, etc., as well as the experience and knowledge of the grower. After analysis and comprehensive consideration of the factors, specific suggestions of planting management, such as optimal sowing time, proper fertilization scheme, proper irrigation water quantity and the like, can be obtained. The agricultural decision based on the experience model can generally help farmers and growers to make more scientific and reasonable planting management decisions, improve the yield and quality of crops, reduce the production cost and further promote sustainable development of agriculture.

However, empirical models in agronomic decisions based on empirical models are typically built up based on personal experiences and observations of farmers or growers and may be limited by factors such as region, climate, soil and crop variety. Such models may only be applicable to specific regions or conditions, and predictions and decisions for different regions or conditions may be inaccurate. And experience models are often influenced by subjective consciousness and experience of individuals, decisions depend on judgment and memory of the individuals, and subjective deviation exists. Different farmers or growers may have different experience and preferences, resulting in impaired consistency and accuracy of decisions. On the other hand, the experience model can only cover the own experience range of farmers or growers, and lacks global data and information. Such locality may lead to decisions ignoring a wider range of factors and trends, failing to make a comprehensive decision.

Thus, although the empirical model has a certain reference value in agricultural planting, there is still a problem of insufficient accuracy.

Disclosure of Invention

In view of the above, an embodiment of the present invention provides a cooperative control system for a plurality of vehicles based on path planning to at least partially solve the above-mentioned problems.

The embodiment of the invention provides a path planning-based cooperative control system for a plurality of planting vehicles, which comprises a positioning subsystem, a positioning subsystem and a positioning subsystem, wherein the positioning subsystem is used for capturing and filtering each reflective marker point on a plurality of capturing targets in an agricultural planting area to obtain three-dimensional space coordinates of each reflective marker point on different time measurement units; the regional division subsystem is used for determining an operation range according to the three-dimensional space coordinates of the preset boundary points in the three-dimensional space coordinates of each reflective mark point, and uniformly arranging a plurality of agricultural robots in a plurality of convex polygon planting regions divided by a landscape plant field or a special terrain field in the operation range; the decision subsystem is used for generating an overall decision according to the cultivation recommendation model, the crop growth condition monitoring model, the local parameter detection model and the global model, and the overall decision is used for coordinating and distributing the operation tasks of the plurality of agricultural robots; and the path planning subsystem is used for planning paths based on the plurality of convex polygon planting areas to obtain a full-coverage operation path so that the plurality of agricultural robots can complete the operation task.

In one implementation, the positioning subsystem includes: the capturing module is used for covering the agricultural planting-planned area and capturing three-dimensional space positions of reflecting mark points on a plurality of capturing targets in the agricultural planting-planned area, and comprises a plurality of motion capturing sensors which are respectively arranged on a centralized server in an external working space and a plurality of agricultural robots on the ground; and the filtering module is used for filtering the captured reflective marker points through a Kalman filtering algorithm to obtain three-dimensional space coordinates of the reflective marker points on different time measurement units.

In another implementation, the uniformly disposing the plurality of agricultural robots in the plurality of convex polygon planting areas divided by the landscape plant field or the special terrain field within the working range includes: and uniformly arranging the plurality of agricultural robots in the plurality of convex polygon planting areas according to the sizes and the shapes of the plurality of convex polygon planting areas and the respective operation performances of the plurality of agricultural robots.

In another implementation, the decision subsystem includes: the first decision module is arranged on the plurality of agricultural robots and is used for generating a plurality of preliminary decisions according to the cultivation recommendation model, the crop growth condition monitoring model and the local parameter detection model carried by the plurality of agricultural robots; the second decision module is arranged on the centralized server and is used for generating an overall decision according to the plurality of preliminary decisions and the global model of the centralized server.

In another implementation, the generating a plurality of preliminary decisions according to the cultivation recommendation model, the crop growth condition monitoring model, and the local parameter detection model onboard the plurality of agricultural robots includes: constructing a cultivation recommendation model based on vision and decision tree fusion, wherein the cultivation recommendation model is used for outputting a recommended crop list according to soil condition information, surrounding environment information and climate season information of a current area where the plurality of agricultural robots are located; constructing a crop growth condition monitoring model based on a neural network, wherein the crop growth condition monitoring model is used for outputting crop growth condition indexes according to crop images and crop growth data, and the crop growth condition indexes are used for evaluating and/or predicting the health state and growth stage of crops; constructing a local parameter detection model, wherein the local parameter detection model is used for outputting model parameters of a tillable area, a recommended crop and agricultural product characteristic data according to the recommended planting list, the crop growth condition index and the agricultural product characteristic data, and the soil condition information, the surrounding environment information, the climate season information, the crop image, the crop growth data and the agricultural product characteristic data of the current area are all acquired by the plurality of agricultural robots; generating the plurality of preliminary decisions based on model parameters of the tillable area, recommended crop and agricultural product feature data.

In another implementation, the generating the overall decision from the plurality of preliminary decisions and the global model of the centralized server includes: inputting the plurality of preliminary decisions and model parameters of the characteristic data of the cultivated area, the recommended crop and the agricultural product into a global model of the centralized server for updating, and obtaining updated model parameters of the characteristic data of the cultivated area, the recommended crop and the agricultural product; and inputting the model parameters of the updated tillable area, recommended crop and agricultural product characteristic data into the local parameter detection model for iterative training, and generating the overall decision.

In another implementation, the system further includes: and the multiple planting vehicle cooperative deep learning subsystem is used for training the multiple agricultural robots through a deep Q network algorithm according to the current working states, the interactive communication states and the real-time position states of the multiple agricultural robots.

In another implementation, the system further includes: and the multi-plant vehicle cooperative training subsystem is used for training the plurality of agricultural robots through a classification algorithm.

In another implementation, the path planning subsystem includes: the terrain data acquisition module is used for acquiring three-dimensional terrain data of the sowing operation pavement in the plurality of convex polygon planting areas and generating a simulated three-dimensional terrain map according to the three-dimensional terrain data and the rand function; the control optimization module is used for planning the motion trail of the plurality of agricultural robots according to the three-dimensional terrain data and a preset optimization function to obtain a planned motion trail; the simulation module is used for simulating the planned motion trail in the simulated three-dimensional topographic map to obtain optimal control parameters; and the execution mechanism module is used for driving the plurality of agricultural robots to move according to the planned motion trail and collecting real-time position data according to the optimal control parameters so as to generate a full-coverage operation path.

In another implementation, the system further includes: and the communication subsystem is used for communicating with each subsystem through the SX1262 wireless communication module, and the SX1262 wireless communication module is mounted on the plurality of agricultural robots.

In summary, the scheme of the embodiment of the invention simplifies the path planning in each sub-area by dividing a plurality of convex polygon planting areas, and effectively improves the coverage efficiency of the whole agricultural robot cluster based on the path planning; generating an overall decision by a decision subsystem according to a cultivation recommendation model, a crop growth condition monitoring model, a local parameter detection model and a global model, and solving the problem of insufficient accuracy of agricultural decisions based on an empirical model; on the other hand, the intelligent agricultural system is based on a group intelligent technology, combines the actual requirements of the agricultural field, utilizes collaborative learning to enable subsystems of different types to mutually cooperate and share information, seeks the optimal solution of various agricultural problems, realizes intelligent agriculture integrating intelligent agricultural machinery, agricultural production management and agricultural big data analysis, and is beneficial to improving the agricultural production efficiency and the agricultural product quality and reducing the agricultural production cost.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the embodiments of the present invention, and other drawings may be obtained according to these drawings for a person having ordinary skill in the art.

Fig. 1 is a schematic diagram of a system structure according to the present invention.

FIG. 2 is a flow chart of an architecture of the present invention.

Fig. 3 is a flow chart of the deep Q network algorithm of the present invention.

FIG. 4 is a schematic view of a two-dimensional implantable region of the present invention.

Fig. 5 is a schematic diagram of the conventional algorithm and the novel algorithm optimization path planning according to the present invention.

Fig. 6 is a schematic diagram of a conventional algorithm and a novel algorithm for optimizing a path planning convergence curve.

Fig. 7 is a three-dimensional path diagram of the invention generated based on the rand function.

Fig. 8 is a graph of an optimal two-dimensional path generated based on the rand function according to the present invention.

Fig. 9 is a schematic diagram of MATLAB simulation iteration effect of the present invention.

FIG. 10 is a pseudo code diagram of the path planning of the present invention.

Detailed Description

For a clearer understanding of technical features, objects, and effects of embodiments of the present application, a specific implementation of the embodiments of the present application will be described with reference to the accompanying drawings.

In this document, "exemplary" means "serving as an example, instance, or illustration," and any illustrations, embodiments, described herein as "exemplary," should not be construed as a more preferred or advantageous solution.

For simplicity of the drawing, the figures show only the parts relevant to the application by way of example, and they do not represent the actual structure of the product. In addition, for simplicity and ease of understanding, components having the same structure or function in some of the figures are shown by way of example only as one or more of them, or only as one or more of them are labeled.

In order to better understand the technical solutions in the embodiments of the present invention, the following description will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which are derived by a person skilled in the art based on the embodiments of the present invention, shall fall within the scope of protection of the embodiments of the present invention.

The implementation of the embodiments of the present invention will be further described below with reference to the accompanying drawings.

Referring to fig. 1 and 2, a path planning-based cooperative control system 100 for a plurality of vehicles according to the present invention includes:

The positioning subsystem 110 is used for capturing and filtering each reflective marker point on a plurality of capturing targets in the agricultural planting planned area to obtain three-dimensional space coordinates of each reflective marker point on different time measurement units;

The regional division subsystem 120 is used for determining an operation range according to the three-dimensional space coordinates of the preset boundary points in the three-dimensional space coordinates of each reflective mark point, and uniformly arranging a plurality of agricultural robots in a plurality of convex polygon planting regions divided by a landscape plant field or a special terrain field in the operation range;

a decision subsystem 130 for generating an overall decision for coordinating and distributing the work tasks of the plurality of agricultural robots based on the cultivation recommendation model, the crop growth status monitoring model, the local parameter detection model, and the global model;

And the path planning subsystem 140 is used for planning paths based on the convex polygon planting areas to obtain a full-coverage operation path so that the plurality of agricultural robots can complete the operation task.

It should be understood that the plurality of motion capture sensors are cameras connected to the cloud monitoring platform and disposed in an external working space, i.e., a planting space, and cameras disposed on a ground agricultural robot, i.e., a planting vehicle.

It should also be appreciated that the capture objects may be diverse, such as obstacles in space, boundary points of a preset area, i.e., an agricultural intended planting area, etc. In an agricultural environment, an obstacle may include a tree, a building, a device, or other immovable object. By capturing the position and shape of these obstacles, the capture module may help the agricultural robot or the autonomous vehicle avoid collisions, enabling safe navigation.

Illustratively, there are two types of capture objects: a dynamic capture object and a static capture object, the static capture object comprising:

boundary marking points: for defining boundaries of a farm field, a work area, or other particular area.

An obstacle: such as trees, buildings or other fixed installations, or static objects that need to be collision-protected.

The dynamic capture targets include:

Agricultural machinery equipment: such as agricultural intelligent trolleys, agricultural robots and the like.

The number of capture objects depends on the specific application scenario and requirements, among other things. For example, in a small-scale farm, it may be necessary to monitor only a small number of glistenings. In more complex environments, such as large-scale farms or precision agricultural applications, it may be desirable to track and record the location of multiple reflector spots and monitor a large number of entities.

For capturing a dynamic capture target, a kalman filtering algorithm is often used for smoothing and predicting a motion track of the dynamic capture target, and can predict a future position of the dynamic capture target according to a current state and historical motion data of the dynamic capture target and correct newly obtained observation data of the dynamic capture target, so that a more stable and continuous three-dimensional coordinate sequence of the dynamic capture target is obtained.

The positioning subsystem can also be used for tracking the positions and motion states of a plurality of agricultural robots and carrying out identification and interconnection among the plurality of agricultural robots.

It will be appreciated that the use of geometric algorithms to automatically break down an overall region into a plurality of convex polygons aims to ensure that each sub-region is convex, thereby simplifying path planning within each sub-region, deploying one or more agricultural robots within each convex polygon planting region, and ensuring that the agricultural robots are able to collect data for the respective region, such as soil moisture, crop growth conditions, etc., may avoid situations where some regions are overworked and others are ignored. Meanwhile, the allocation scheme is adjusted according to the operation performance of the agricultural robot, so that the operation efficiency can be further improved.

In area division, the method focuses on the spatial decomposition and geometric characteristics of the area, divides the area into a plurality of convex polygons with limited quantity, and plans planting paths in the plurality of convex polygons. The regional division is an important step of coverage path planning, and can divide complex regions so as to obtain the matching of the complex regions with preset rule subregions, thereby realizing the coordination among agricultural robots. Meanwhile, when the subareas are concave, the coverage path planning is not easy to directly carry out, and the proper area division can simplify the coverage path planning problem and reduce the energy consumption and the working time of planting.

It should be understood that the agricultural product characteristic data includes crop morphological characteristic data, crop fruit characteristic data, crop seed characteristic data, environmental data about the cultivated land, and the like.

Optionally, inputting the updated model parameters of the tillable area, the recommended crop and the agricultural product feature data into the local parameter detection model for iterative training, and generating the overall decision specifically includes: inputting the updated model parameters of the tillable area, the recommended crop and the agricultural product characteristic data into the local parameter detection model for iterative training, and outputting a model prediction result; and generating an overall decision based on a group intelligent algorithm according to the perception data and model prediction results acquired by a plurality of agricultural robots, and performing task allocation according to the overall decision so as to ensure that the agricultural robots adapt to actual conditions in the agricultural fields and ensure the high efficiency and coverage rate of the operation.

It should be appreciated that the first decision module is disposed on a plurality of agricultural robots to generate a plurality of preliminary decisions, which may be understood as the agricultural robots making the preliminary decisions, and the second decision module is disposed on a centralized server to generate the overall decisions, which may be understood as the centralized server generating the overall decisions.

It should also be understood that the perceived data is data obtained from the network by the agricultural robot, including satellite positioning data, data of the terrain of the planting area, the plantable area, soil moisture content such as water and fertility, etc., meteorological data, etc.

It should also be understood that the overall decision refers to the final decision that is obtained by comprehensively analyzing and optimizing the data and preliminary decisions of all agricultural robots through a centralized server. The overall decision is used to coordinate and distribute the job tasks of multiple agricultural robots, aiming to maximize overall efficiency while taking into account coordination and cooperation between individual agricultural robots.

For example, if the agricultural robot a-vehicle makes a preliminary decision that the agricultural robot a-vehicle should perform seeding work, after the a-vehicle returns its preliminary decision and model parameters to the centralized server, the centralized server may perform overall decision, coordinate and distribute a plurality of operation tasks of the agricultural robot, for example, determine whether to call the B-vehicle or the C-vehicle nearby to assist in seeding or to perform other work according to the magnitude of the seeding workload of the a-vehicle at this time.

In another implementation, the system further includes: the multiple planting vehicle cooperative deep learning subsystem 150 is configured to train the multiple agricultural robots through a deep Q network algorithm according to the current working states, the interactive communication states and the real-time position states of the multiple agricultural robots.

Illustratively, referring to FIG. 3, the multi-plant cooperative deep learning subsystem employs a deep Q network algorithm (DQN) that approximates a state-action value function using a neural network (referred to as a deep Q network). The neural network receives the current working states, the interactive communication states and the real-time position states of a plurality of agricultural robots as inputs and outputs the Q value of the actions of the agricultural robots. The Q value refers to an indicator for a given state and combination of actions that measures the quality of the action in that state.

Specifically, the Q value represents the expected future jackpot after taking some action in the current state. In the context of an agricultural robot, each action may correspond to an operation or behavior, such as adjusting a travel speed, steering angle, amount of fertilizer applied, etc. The Q-value agricultural robot selects which action can maximize future benefits or benefits for a given state (e.g., current operating state, interactive communication state, and real-time location state, etc.).

In the training process of the agricultural robot, the DQN gradually approaches the optimal Q value function through the interaction of the agricultural robot and the environment and the continuous updating of the weight of the neural network. Thus, in practical application, the agricultural robot can select the action with the highest Q value according to the current state (the current working state, the interactive communication state, the real-time position state and the like) and the trained deep Q network, so that decision operation such as preliminary decision is realized.

The training process of DQN includes two phases: an offline phase and an online phase. In the offline phase, the agricultural robot randomly samples a batch of experiences (i.e., a series of collected data of state data, motion data, etc., originally measured by the agricultural robot) from the experience playback buffer and uses the experiences to update the depth Q network. In the online phase, the agricultural robot uses the current state data (current working state, interactive communication state and real-time position state) and the depth Q network to select and perform the best action, and stores the new experience in the experience playback buffer. Therefore, training is performed on a plurality of agricultural robots through the deep Q network algorithm, so that the decision capability of the agricultural robots is improved, and the decision accuracy of the decision subsystem is further improved.

In another implementation, the system further includes: the multiple plant-on-vehicle co-training subsystem 160 is used to train the multiple agricultural robots through a classification algorithm.

Illustratively, the multi-plant vehicle cooperative training subsystem performs calculation based on a two-classification algorithm and adopts a knowledge and data dual-drive mode. The principle is as follows:

In the tag data total training library M, data is extracted as behavioural data a and non-behavioural data B, and two classifiers X1 and X2 are trained on (a and B are two views of the data). Randomly selecting z samples from N of the unlabeled data total training library, putting the samples into another library L, and circulating for K times: training classifier X1 with part A of M; training classifier X2 with part B of M; marking all elements in L by X1, selecting p positive marks and q negative marks with highest confidence, marking all elements in L by X2, and selecting p positive marks and q negative marks with highest confidence; adding the 2p+2q marker samples selected above to M; and randomly selecting unlabeled data from N, and repeating the cycle until the cycle of N being an empty set is terminated. The estimation of the classification confidence coefficient is different due to different learning algorithms, and the posterior probability can be converted into the classification confidence coefficient when a Bayes classifier is selected; when using a support vector machine as a classifier, then the granularity may be translated into classification confidence, etc.

It should be appreciated that the overall training library M of tag data: the data in this library has been manually marked or annotated, meaning that each data sample corresponds to a known label or result, so that the actual measured local parameter detection model can be learned and adjusted by comparing its predicted and actual labels.

Unlabeled data total training library N: in contrast, the data in this library has not been marked or annotated. These data are typically used in semi-supervised or unsupervised learning scenarios to help the actual measured local parametric test model discover more potential information and features, thereby improving the generalization ability of the local parametric test model.

In practice, the marking data include data such as images of crop types that have been classified, soil samples of known types, boundaries, paths of agricultural robot movements, and decisions made. Unlabeled data may be a large amount of raw farmland image data, soil data, climate data, and unlabeled decisions.

It should also be appreciated that the two views (i.e., viewing the data in two angles A, B) are:

Behavior data a: this portion of data generally refers to data related to a particular behavior or operation of the agricultural robot. For example, this may include a movement trajectory of the agricultural robot, a type of task performed, a record of success or failure of the operation, and the like. The behavior data reflects the actual behavior of the agricultural robot in the agricultural field and the executed tasks, and is an important basis for evaluating the performance and the working efficiency of the agricultural robot.

Non-behavioural data B: this portion of the data may then include other information not directly related to the agricultural robot behavior, such as environmental parameters (soil humidity, temperature, light, etc.), agricultural robot status, or other data related to the farm environment (e.g., soil type, crop growth, etc.). The non-behavior data provides background information of the agricultural robot operating environment, which helps to more fully understand the influencing factors of the agricultural robot behavior.

The training of the plurality of agricultural robots through the two-classification algorithm is mainly to improve the recognition capability of the agricultural robots to different data through training the classifier. Although this process also involves training, it differs from the training purposes and methods of DQN. The training of the classification algorithm is mainly used for improving the accuracy of the classifier, so that the classifier can better distinguish the behavior data from the non-behavior data, the classifier is marked by manual classification and then stored in the local parameter model of the agricultural robot as a reference, and the accuracy of the agricultural robot on the identification data can be improved by training the classifier.

In summary, an agricultural robot trained by the DQN algorithm will be able to autonomously select an optimal action according to the current state. Meanwhile, through training of a classification algorithm, the agricultural robot can also recognize and process different types of data more accurately. The training processes jointly promote the intelligent level of the agricultural robot, so that the agricultural robot can autonomously make decisions in a complex environment.

For ease of understanding, prior art path planning algorithms are illustrated prior to describing the path planning of the present invention in detail.

In the prior art, a single intelligent vehicle, i.e. an agricultural robot, is exemplified. Firstly, initializing the initial position of the intelligent trolley, and analyzing constraint conditions and performance indexes of tasks according to model prediction results, such as obtaining planting difficulty of various plants or a plantable area of terrains, and evaluating the fitness function of each intelligent trolley. In order to better simulate the field complex planting environment condition, referring to fig. 4, a 50×50 plantable area diagram is created by adopting a rand random function, wherein a white part in the diagram represents an area where the intelligent trolley can freely move, and a black part represents a large obstacle which cannot be surmounted, the invention uses a novel optimization algorithm to plan a travelling path of the intelligent trolley when encountering the obstacle, and compares the result with the existing traditional optimization algorithm to obtain the difference of the two in path planning, and simultaneously draws a path planning convergence curve of the two algorithms, particularly referring to fig. 5 and 6. Finally, the shortest path obtained by optimizing the traditional algorithm is as follows: 89.028; the shortest path obtained by optimizing the novel algorithm is as follows: 86.1476.

The working principles of the conventional algorithm optimizing and the novel algorithm optimizing in the two-dimensional space are described below:

Step one: initializing parameters and variables, and setting the number of search agents, namely agricultural robots, the maximum iteration number, the lower and upper bounds of a search space, the dimensionality of a problem and an objective function. Initializing the positions and scores of the optimal, suboptimal and third optima, and setting a convergence curve.

Step two: initializing the position of a search agent, traversing each column according to the grid map of the planting land to find a free grid, wherein the free grid is a plantable grid. For each search agent, a free grid is randomly selected as the initial location.

Step three: iteratively searching for an optimal solution, for each iteration, performing the following operations: all search agents are traversed, whether the positions of the search agents exceed the boundary of the search space is checked, and if so, the search agents are pulled back into the boundary. The fitness, i.e. the objective function value, of the current location of each search agent is calculated. And updating the positions and the scores of the optimal, suboptimal and third optimal according to the fitness.

Step four: calculating a Levy random vector, updating the position of the search agent, calculating a cosine value according to the current iteration number, and generating the Levy random vector. The position of the search agent is adjusted using cosine values and Levy random vectors to explore different regions of the search space.

Step five: and recording a convergence curve, and recording the current optimal score into the convergence curve after each iteration is finished so as to facilitate the convergence condition of a subsequent analysis algorithm.

Step six: and outputting a result, namely outputting an optimal position, an optimal score and a convergence curve when the maximum iteration number is reached.

However, in the actual seeding production process, the road surface tends to be uneven, which affects the running stability and seeding accuracy of the agricultural robot. And considering the actual size of the agricultural robot and the change of the gravity center of the road surface running on different heights, the two-dimensional space-based path planning operation is low in efficiency and is not suitable for the agricultural production nowadays, so that the invention introduces a path planning subsystem which can adapt to the three-dimensional terrain change and comprises:

The terrain data acquisition module is used for acquiring three-dimensional terrain data of the sowing operation pavement in the plurality of convex polygon planting areas and generating a simulated three-dimensional terrain map according to the three-dimensional terrain data and the rand function;

The control optimization module is used for planning the motion trail of the plurality of agricultural robots according to the three-dimensional terrain data and a preset optimization function to obtain a planned motion trail;

The simulation module is used for simulating the planned motion track in the simulated three-dimensional topographic map through MATLAB to obtain optimal control parameters;

And the execution mechanism module is used for driving the plurality of agricultural robots to move according to the planned motion trail and collecting real-time position data according to the optimal control parameters so as to generate a full-coverage operation path.

And obtaining a three-dimensional topographic map according to the rand function, and further adopting an optimization function to perform MATLAB simulation to obtain a result which is shown in fig. 7 and is an optimal three-dimensional path map generated based on the rand function. The best two-dimensional path graph generated based on the rand function is shown in fig. 8. A schematic diagram of MATLAB simulation iteration effect is shown in fig. 9.

The motion trail planning of a plurality of agricultural robots adopts an optimization and improvement algorithm, and the working principle of the improved optimization algorithm of the path planning subsystem which is introduced by the invention and can adapt to three-dimensional terrain change is described below:

In practical application, three-dimensional topographic data of the seeding operation pavement is firstly obtained through a topographic data obtaining module. The algorithm is initialized according to the input parameters (such as population size, maximum iteration number, search space upper and lower bounds, dimension). A population X is initialized, where each individual represents a candidate solution in the search space. And calculating the fitness value of each individual in the initial population. And initializing a global optimal solution and global optimal fitness.

Wherein each individual refers to the position of one agricultural robot at each moment. The individual represents the position X (i) where the individual is located at that time, and the planned path of the agricultural robot is performed by calculating the position X (i+1) where the individual should be located at the next time.

In the improved optimization algorithm, the location of each agricultural robot is considered an "individual" and the algorithm plans the path of the agricultural robot by constantly optimizing the locations of those individuals. For each agricultural robot, the algorithm decides whether to update its path by comparing the fitness of its new location with the fitness of the original location. And if the adaptability of the new position is higher, namely the path is better, the new position is accepted as the position of the agricultural robot at the next moment.

Comprehensively considering the influence degree of various factors on the algorithm effect, the invention respectively quantifies the influence degree into the following constant values: first, the path length factor takes up an important role, with an impact weight of 0.4; secondly, the size of the agricultural robot and the size of the obstacle have obvious influence on the algorithm effect, and the influence weight is 0.3; finally, various environmental information factors are also a non-negligible part, which also have an impact weight of 0.3. These weight values will be used in the algorithm design and optimization process to ensure that various influencing factors are adequately considered and balanced.

Some symbols and variables are defined below:

X (i): representing the position vector of the ith individual in the population.

Popsize: population size.

GbSol: a position vector of the globally optimal solution.

Fitness (i): fitness value of the i-th individual.

U1 and U2: a randomly generated vector or matrix for location update.

Randn: a standard normal distribution of random numbers.

Rand: random numbers uniformly distributed in the [0,1] interval.

Randi (Popsize): random integers uniformly distributed in the [1, pop ] interval are used to randomly select an individual from a population.

Α, r2, tf, t, tmax, dim: alpha represents the convergence rate and is used for controlling the algorithm convergence speed; r2 represents random numbers uniformly distributed in the range of [0, 1), and is used for adjusting the step length and the direction of solution updating, so that the randomness and the diversity of the searching process are increased; tf represents a parameter between 0 and 1 for making a trade-off between two different mechanisms of the algorithm; t represents the current iteration number; tmax represents the maximum number of iterations, i.e., the maximum number of attempts by the algorithm to find the optimal solution; dim is the dimension of the problem, i.e. the number of variables that need to be optimized.

St, yt, ft, vt: ft represents an adjustment factor based on the fitness of the current individual and a random number, which is calculated using the fitness value of the current individual, a randomly selected location, and a random number. Ft is used to further adjust the search step size and search direction to enhance the exploratory capabilities of the algorithm. St denotes a weighting factor based on the fitness of the current individual, which is calculated using the fitness value of the current individual, fitness (i), and the sum of all individual fitness, st is used to adjust the search step size and direction, helping the algorithm to adapt better to the characteristics of the problem. Yt represents a dynamically adjusted parameter for controlling the balance between the exploration and development of the algorithm, which is calculated based on the current number of iterations t and the maximum number of iterations Tmax. As the iteration proceeds, the value of Yt will gradually decrease, helping the algorithm to pay more attention to development rather than exploration in the later stages of the iteration. Vt represents the location of the current individual.

The improved optimization algorithm enters an iterative process and loops until the maximum number of iterations is reached. In each iteration, the following is performed for each individual in the population:

The random number rand is generated for deciding to enter the a or B stage.

If stage A is entered, the first mechanism or the second mechanism is selected based on the random number to update the location of the individual.

A first mechanism:

An intermediate point is calculated that is the average of the current individual and another randomly selected individual in the population.

Updating the position of the current individual, and weighting the difference value between the random number generated by normal distribution and the global optimal solution and the intermediate point.

Where y represents the average of the current individual and another randomly selected individual in the population, i.e., the average of the locations, and X (i+1) represents the location of the individual at the next time.

A second mechanism:

Intermediate points are also calculated.

Updating the position of the current individual, and weighting the difference values of the current position, the middle point and the other two random individuals in the population by using the random number U1.

If B stage is entered, a third or fourth mechanism is selected to update the location of the individual based on the random number preset condition.

A third mechanism:

A weight is calculated based on the fitness of the individual.

A step size is calculated and multiplied by Yt and St.

Updating the current individual position, weighting the current position, the positions of random individuals in the population and the difference thereof by using the random number U1, and subtracting the step size.

A fourth mechanism:

St is calculated from the fitness of the individual.

Ft is calculated based on the global optimal solution and the locations of the random individuals in the population.

The step size S is updated and multiplied by Yt and Ft.

Updating the position of the current individual, weighting by using the difference between the global optimal solution and the current position, and subtracting the step S.

Boundary processing: the improved optimization algorithm checks whether each dimension of each individual exceeds the upper and lower bounds ub and lb of the search space. If a dimension exceeds a boundary, the algorithm resets the value of that dimension to a random value within the boundary.

Fitness evaluation and optimal solution update: for each newly generated individual, the improved optimization algorithm calculates its fitness value.

The improved optimization algorithm compares the fitness value of the new individual with the fitness value before it and the fitness value of the globally optimal solution to decide whether to update the individual optimal solution Xp or the globally optimal solutions GbSol and GbFit.

Population size adjustment: after each iteration is completed, the improved optimization algorithm adjusts the population size Popsize according to the current iteration number t and some preset parameters. This way of dynamically sizing the population is to balance the ability to explore and utilize during the search.

Outputting information of the optimal solution: the improved optimization algorithm calls the objective function to obtain the best solution. And outputting corresponding information according to the feasibility of the optimal solution. At the end of each iteration, the current global optimum fitness is added to the convergence curve.

Illustratively, referring to FIG. 10, a pseudo-code diagram of path planning according to the present invention describes the specific steps of an improved optimization algorithm:

First, initial parameters including a population size (N), a maximum number of iterations (t_max), a final time (t_f), and a minimum population size (n_mn) are set. Then, the position of the solution is randomly initialized and is marked as (X [ i ]), wherein (i) represents the number of the individuals in the population, and the value range is 1 to (N). Meanwhile, the current iteration number (t) is set to 0. Then, the main cycle is entered. As long as the current iteration number (T) is smaller than the maximum iteration number (t_ { max }), the iterative optimization is continued. In each iteration, the fitness of each solution in the current population is first evaluated. And determining the optimal solution in the current group according to the fitness value, and recording the optimal solution as (x_ { best }). The parameter (y_t) is then updated according to a specific formula taking into account the relation between the current iteration number and the maximum iteration number. Meanwhile, the size (N) of the group is adjusted according to the requirement of an algorithm. Next, an update operation is performed for each individual in the population. First, the relevant intermediate variables are calculated. Then, two random numbers (r_1) and (r_2) are generated.

If (r_1) is less than (r_2), the first phase is entered. In the first stage, two random numbers (r_3) and (r_4) are regenerated. If (r_3) is less than (r_4), then updating the solution's location using a first mechanism; otherwise, the second mechanism is adopted for updating.

If (r_1) is not less than (r_2), the second stage is entered. In the second phase, two new random numbers (r_5) and (r_6) are generated. If (r_5) is less than (r_6), then updating the solution's location using a third mechanism; otherwise, a fourth mechanism is used for updating. After each update of the position of the solution, the fitness of the new solution is evaluated. And if the adaptability of the new solution is better than that of the original solution, replacing the original solution with the new solution. After all individual updates are completed, the number of iterations (T) is increased by 1 and the next iteration is continued until the maximum number of iterations (t_ { max }) is reached. Finally, the found optimal solution (x_ { best }) is returned, and the operation of the algorithm is stopped. By the method, the algorithm can find a globally optimal solution or an approximately optimal solution in a given iteration number.

Specifically, the agricultural integrated intelligent planting trolley is taken as a main body to describe, and the trolley can continuously update the position and the speed of the trolley according to the current position and the speed of the trolley and simultaneously referring to the optimal position information in the whole intelligent trolley group. This process is essentially a dynamic movement process where the cart finds the optimal solution in a wide search space. When the algorithm operation reaches the preset maximum iteration number or the task execution cost reaches the set threshold, the improved optimization algorithm stops operation and outputs the optimal solution obtained through optimization calculation. Then, the agricultural integrated intelligent planting trolley accurately and efficiently performs agricultural planting operation according to the task allocation instruction of the optimal solution.

In the task execution process, the coverage efficiency and the decomposition of the planting areas are comprehensively considered, and the invention ensures that each sub-planting area is of a convex polygon, and based on the convex polygon, the coverage efficiency of the whole robot cluster is effectively improved. In addition, the width supporting line direction of the sub-planting area is approximately taken as the main axis direction, and meanwhile, the minimum turning radius of the agricultural integrated intelligent planting trolley is fully considered, so that fine coverage path planning is performed. The planning mode not only can reduce the total running length of the trolley, but also can effectively shorten the maneuvering length during turning, thereby further improving the working efficiency.

The regional division method provided by the invention has high calculation speed and can quickly obtain a better feasible solution. In practical application, after the method is adopted for regional division, the implementation difficulty of coverage path planning can be obviously reduced, so that the agricultural integrated intelligent planting trolley or the agricultural robot can more intelligently and efficiently complete planting tasks.

In another implementation, the system further includes: the communication subsystem 170 is configured to communicate with each subsystem through an SX1262 wireless communication module, where the SX1262 wireless communication module is mounted on the multiple agricultural robots.

The communication subsystem 170 is exemplified to make the subsystems connected, the wireless network power management system based on Zigbee technology transmits signals to the air through the SX1262 wireless communication module carried by the agricultural robot, the transmitted signals are received by the wireless communication modules carried by other agricultural robots and the server control platform, and the subsystems output and receive through continuous reciprocating signals, so as to realize wireless communication.

The present invention may be a system, method, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for causing a processor to implement aspects of the present invention.

The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, punch cards or in-groove structures such as punch cards or grooves having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media, as used herein, are not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses through fiber optic cables), or electrical signals transmitted through wires.

The computer readable program instructions described herein may be downloaded from a computer readable storage medium to a respective computing/processing device or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. The network interface card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in the respective computing/processing device.

The computer program instructions for carrying out operations of the present invention may be assembly instructions, instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as SMALLTALK, C ++, python, etc., and conventional procedural programming languages, such as the "C" language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, aspects of the present invention are implemented by personalizing electronic circuitry, such as programmable logic circuitry, field Programmable Gate Arrays (FPGAs), or Programmable Logic Arrays (PLAs), with state information for computer readable program instructions, which can execute the computer readable program instructions.

Various aspects of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable medium having the instructions stored therein includes an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

The foregoing description of embodiments of the invention has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various embodiments described. The terminology used herein was chosen in order to best explain the principles of the embodiments, the practical application, or the technical improvements in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.

Claims

1. A multi-vehicle cooperative control system based on path planning, characterized by comprising:

A positioning subsystem is used to capture and filter each reflective mark point on a plurality of capture targets in the planned agricultural planting area to obtain the three-dimensional spatial coordinates of each reflective mark point in different time measurement units;

A region division subsystem is used to determine the operation range according to the three-dimensional space coordinates of the preset boundary points in the three-dimensional space coordinates of the reflective marking points, and evenly arrange multiple agricultural robots in multiple convex polygonal planting areas divided by landscape plant fields or special terrain farmlands within the operation range;

A decision-making subsystem, used to generate an overall decision based on the farming recommendation model, the crop growth status monitoring model, the local parameter detection model and the global model, wherein the overall decision is used to coordinate and allocate the operation tasks of the multiple agricultural robots;

A path planning subsystem, used for performing path planning based on the multiple convex polygonal planting areas to obtain a full coverage operation path so that the multiple agricultural robots can complete the operation tasks;

The decision subsystem includes: a first decision module, which is arranged on the multiple agricultural robots and is used to generate multiple preliminary decisions according to the farming recommendation model, the crop growth status monitoring model and the local parameter detection model carried by the multiple agricultural robots; a second decision module, which is arranged on a centralized server and is used to generate an overall decision according to the multiple preliminary decisions and the global model of the centralized server;

The generating of multiple preliminary decisions according to the farming recommendation model, the crop growth status monitoring model and the local parameter detection model carried by the multiple agricultural robots includes: constructing the farming recommendation model based on vision and decision tree fusion, the farming recommendation model is used to output a recommended crop list according to the soil condition information, surrounding environment information and climate season information of the current area where the multiple agricultural robots are located; constructing the crop growth status monitoring model based on a neural network, the crop growth status monitoring model is used to output crop growth status indicators according to crop images and crop growth data, and the crop growth status indicators are used to evaluate and/or predict the health status and growth stage of crops; constructing a local parameter detection model, the local parameter detection model is used to output model parameters of arable areas, recommended crops and agricultural product characteristic data according to the recommended crop list, the crop growth status indicators and agricultural product characteristic data, the soil condition information, the surrounding environment information, the climate season information, the crop images, the crop growth data and the agricultural product characteristic data of the current area are all collected by the multiple agricultural robots; generating the multiple preliminary decisions according to the model parameters of the arable areas, recommended crops and agricultural product characteristic data;

The method of generating an overall decision based on the multiple preliminary decisions and the global model of the centralized server includes: inputting the multiple preliminary decisions and the model parameters of the arable area, recommended crops and agricultural product characteristic data into the global model of the centralized server for updating, thereby obtaining updated model parameters of the arable area, recommended crops and agricultural product characteristic data; and inputting the updated model parameters of the arable area, recommended crops and agricultural product characteristic data into the local parameter detection model for iterative training, thereby generating the overall decision.

2. The system according to claim 1, characterized in that the positioning subsystem comprises:

A capture module is used to cover the planned agricultural planting area and capture the three-dimensional spatial position of each reflective mark point on a plurality of capture targets in the planned agricultural planting area, wherein the capture module includes a plurality of motion capture sensors, and the plurality of motion capture sensors are respectively arranged on a centralized server in an external working space and on a plurality of agricultural robots on the ground;

The filtering module is used to filter the captured reflective marking points through the Kalman filtering algorithm to obtain the three-dimensional spatial coordinates of each reflective marking point in different time measurement units.

3. The system according to claim 1, characterized in that the multiple agricultural robots are evenly arranged in multiple convex polygonal planting areas divided by landscape plant fields or special terrain farmlands within the operating range, comprising:

According to the sizes and shapes of the multiple convex polygonal planting areas and the respective operating performances of the multiple agricultural robots, the multiple agricultural robots are evenly arranged in the multiple convex polygonal planting areas.

4. The system according to claim 1, characterized in that the system further comprises:

The multiple planting-vehicle collaborative deep learning subsystem is used to train the multiple agricultural robots through a deep Q network algorithm according to the current working status, interactive communication status and real-time position status of the multiple agricultural robots.

5. The system according to claim 1, characterized in that the system further comprises:

The multi-planting vehicle collaborative training subsystem is used to train the multiple agricultural robots through a binary classification algorithm.

6. The system according to claim 1, wherein the path planning subsystem comprises:

A terrain data acquisition module, used to acquire three-dimensional terrain data of the sowing operation road surface in the plurality of convex polygonal planting areas, and generate a simulated three-dimensional terrain map according to the three-dimensional terrain data and a rand function;

A control optimization module, used for planning the motion trajectories of the plurality of agricultural robots according to the three-dimensional terrain data and a preset optimization function to obtain the planned motion trajectories;

A simulation module, used for simulating the planned motion trajectory in the simulated three-dimensional terrain map to obtain optimal control parameters;

The actuator module is used to drive the multiple agricultural robots to move according to the planned motion trajectory and collect real-time position data to generate a full coverage operation path based on the optimal control parameters.

7. The system according to any one of claims 1 to 6, characterized in that the system further comprises:

The communication subsystem is used to communicate with each subsystem through the SX1262 wireless communication module, and the SX1262 wireless communication module is mounted on the multiple agricultural robots.