Multi-robot collaborative harvesting method and system for woody oil based on group intelligence

Through technical means such as distributed perception network and three-layer decision-making architecture, the reliability and efficiency of the woody oil multi-robot collaborative operating system is solved, efficient and intelligent harvesting is achieved, identification accuracy and energy utilization are improved, and damage rate and costs are reduced.

CN120266681AActive Publication Date: 2025-07-08HARBIN FORESTRY MASCH RES INST STATE FORESTRY & GRASSLAND ADMINISTRATION

Patent Information

Application Number
CN202510405722.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-04-02
Publication Date
2025-07-08
Estimated Expiration
2045-04-02

AI Technical Summary

Technical Problem

The existing multi-robot collaborative operating system has problems such as low reliability of centralized control architecture, high communication dependence, single perception ability, insufficient algorithm adaptability, simple coordination mechanism, no time dimension in planning and no crop characteristics in woody oil harvest, resulting in inefficiency of the system in complex environments.

Method used

We adopt distributed perception network, semantic augmented digital twin model, three-layer decision-making architecture, adaptive task allocation and execution, federated learning architecture and dynamic balance optimization, and obtain data through multi-modal sensors, establish fruit attribute models, implement group collaborative learning and dynamic balance optimization to achieve efficient and intelligent harvesting.

Benefits of technology

It improves system reliability and identification accuracy, improves harvesting efficiency and energy utilization, reduces fruit damage rate and production costs, and significantly improves scalability and adaptability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120266681A_ABST
    Figure CN120266681A_ABST
Patent Text Reader

Abstract

The invention relates to the field of intelligent agricultural robots, in particular to a multi-robot collaborative harvesting method and system for woody oil plants based on swarm intelligence, and provides efficient harvesting of the woody oil plants by constructing a distributed sensing network and a semantic enhanced digital twinborn model, and the system adopts a three-layer decision architecture to formulate a harvesting strategy. The working efficiency is improved through self-adaptive task allocation and collaborative harvesting, multiple robots share experience through federated learning, group collaborative learning is achieved, harvesting performance is optimized, the reliability of the system is improved by 85%, single-point fault risks do not exist, the recognition accuracy is improved by 50% through a multi-modal sensing system, and the system is suitable for popularization and application. And an innovative harvesting solution is provided for the field of intelligent agricultural robots.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent agricultural robots, and particularly to a multi-robot collaborative harvesting method and system for woody oil crops based on swarm intelligence, which is applicable to the intelligent and efficient harvesting operations of woody oil crops such as camellia seeds, olive fruits, and oil-tea camellia seeds. Background Art

[0002] With the improvement of the level of agricultural modernization and the increase in rural labor costs in China, the technology of intelligent agricultural robots has gradually been applied and developed. Woody oil crops are characterized by wide distribution, high value, and high manual harvesting costs, and their intelligent harvesting has become an important research direction in modern agriculture.

[0003] Currently, there have been relevant studies on multi-robot collaborative operation systems. For example, Chinese Patent CN 111240319 B discloses an outdoor multi-robot collaborative operation system and its method. The system adopts a master-slave structure, consisting of a master robot and several slave robots. Each robot is equipped with navigation and positioning, motion control, sensors, communication, and main control modules. The master robot of the system constructs an environmental map through a high-precision lidar, the slave robots receive the map information, and the total control background is responsible for task allocation and path planning.

[0004] However, the above-mentioned existing technologies have the following deficiencies:

[0005] 1. The centralized control architecture makes the system less reliable, and the failure of the master robot or the total control background will cause the entire system to collapse;

[0006] 2. The communication dependence is too high, and communication bottlenecks are likely to occur in complex orchard environments;

[0007] 3. The perception ability is single, and it is difficult to effectively identify the characteristics and maturity of oil crops only relying on lidar;

[0008] 4. The algorithm adaptability is insufficient, and the fixed algorithm combination is difficult to adapt to the complex and changeable environment and task requirements of the orchard;

[0009] 5. The collaborative mechanism is simple, mainly limited to task allocation and obstacle avoidance, lacking in-depth collaboration for harvesting operations;

[0010] 6. The planning has no time dimension, only considering spatial path planning and not considering the optimization of harvesting timing;

[0011] 7. There is no consideration of crop characteristics, and no optimization is carried out for the special growth characteristics of woody oil crops. Summary of the Invention

[0012] The object of the present invention is to provide a multi-robot collaborative harvesting method and system for woody oilseeds based on swarm intelligence, which realizes the efficient and intelligent harvesting of woody oilseed crops through a distributed sensing network, a three-layer decision-making architecture, adaptive task allocation and execution, swarm collaborative learning, and dynamic balance optimization, and overcomes the above-mentioned defects existing in the prior art.

[0013] The present invention proposes a multi-robot collaborative harvesting method for woody oilseeds based on swarm intelligence, including:

[0014] Constructing a distributed sensing network for the oilseed orchard to obtain fruit distribution data, maturity data, and tree structure data; the distributed sensing network consists of multi-modal sensors equipped on multiple robots, and each robot collects local sensing data and shares it;

[0015] Based on the data obtained from the distributed sensing network, establishing a semantic-enhanced digital twin model including fruit attributes;

[0016] According to the digital twin model, generating a harvesting decision through a three-layer decision-making architecture, which includes a global strategy layer, a regional tactics layer, and a local operation layer; among them, the global strategy layer formulates an overall harvesting strategy, the regional tactics layer generates a regional harvesting plan, and the local operation layer determines specific harvesting actions;

[0017] Based on the harvesting decision, performing adaptive task allocation and collaborative harvesting, including: evaluating task value and robot capabilities using a virtual economy model, dynamically allocating harvesting tasks; selecting corresponding harvesting strategies according to fruit characteristics; avoiding conflicts through multi-robot collaborative trajectory planning;

[0018] Collecting the task execution results, and realizing swarm collaborative learning and experience sharing through a federated learning architecture;

[0019] Based on the learning results, performing dynamic balance optimization to seek the global optimum among harvesting quantity, quality, energy efficiency, and uniformity.

[0020] Preferably, the construction of the distributed sensing network for the oilseed orchard includes:

[0021] Configuring the multi-modal sensor combination for each robot, and the multi-modal sensor combination includes an RGB camera, a multi-spectral camera, a lidar, a force and tactile sensor, and an environmental sensor;

[0022] Using the RGB camera and the multi-spectral camera to identify fruit characteristics, and dividing the fruit maturity into 5 grades;

[0023] Using the lidar to generate point cloud data and identify the tree structure;

[0024] Based on the multi-modal sensor data, a fruit distribution density heat map and a harvest difficulty assessment are generated, and the harvest difficulty is quantified on a scale of 1-10 points.

[0025] Preferably, the establishment of the semantic enhanced digital twin model includes:

[0026] Using the distributed SLAM technology to construct a local environment map, and each robot maintains a local map within a range of 50m×50m;

[0027] Map stitching is achieved through ORB feature point matching, and the map overlap is required to be greater than 30%;

[0028] Upgrading the traditional SLAM to semantic SLAM, and introducing semantic labels such as tree bodies, fruits, obstacles, etc.;

[0029] Constructing a complete digital twin model including a tree body geometric model, a fruit attribute database, a terrain obstacle representation, and historical harvest data.

[0030] Preferably, the generation of the harvest decision through the three-layer decision-making architecture further includes:

[0031] Implementing a distributed consensus protocol based on the blockchain concept, adopting a proposal-validation-voting mode, and executing after more than 67% of the robots agree to the plan;

[0032] Introducing a time decay factor λ = e -αt , making the weight of recent data higher;

[0033] Establishing a multi-objective optimization function F(x) = w1·yield(x) + w2·quality(x) - w3·energy consumption(x) - w4·time(x), where w1, w2, w3, and w4 are dynamically adjusted weight parameters;

[0034] Implementing a multi-agent collaborative reasoning system to support distributed information aggregation and joint reasoning.

[0035] Preferably, the execution of adaptive task allocation and collaborative harvesting includes:

[0036] Hierarchically decomposing the harvest tasks into tasks at the park level (L1), regional level (L2), tree body level (L3), and fruit level (L4);

[0037] Optimizing task allocation through a virtual economy model, calculating the task value V(task) = the number of fruits × maturity × market value, and the robot ability score C(robot) = historical efficiency × battery power × actuator accuracy;

[0038] Identifying 5 typical fruit states (easy-to-fall type, firm type, cluster type, blocked type, high-position type), and selecting a dedicated harvest action sequence for each type;

[0039] Implement multi-robot collaborative trajectory planning, predict potential path intersections within 10 seconds, and execute a priority-based avoidance strategy.

[0040] Preferably, the realization of group collaborative learning and experience sharing through the federated learning architecture includes:

[0041] Train the model locally on each robot, aggregating only model parameters rather than raw data;

[0042] Apply differential privacy protection mechanism to add appropriate noise to protect sensitive information;

[0043] Construct a collective experience library that preferentially stores high-value experiences, where the experience representation includes scene features, execution actions, result evaluation, and context information;

[0044] Implement online reinforcement learning, where the state space S includes the robot's position, power, load, and the distribution of surrounding fruits, the action space A includes the moving direction, harvesting actions, and cooperation requests, and the reward function R = the amount of harvested fruits × quality coefficient - energy consumption × energy consumption coefficient - time × time coefficient.

[0045] Preferably, the execution of dynamic balance optimization includes:

[0046] Establish a multi-objective optimization framework to simultaneously optimize the amount of harvested fruits (the number of fruits harvested per hour), quality assurance (the fruit damage rate < 5%), energy efficiency (the number of fruits harvested per unit of power), and harvesting uniformity (the variance of the completion degree of harvesting in different regions);

[0047] Implement adaptive resource allocation, allocate short-distance tasks to robots with low power, and allocate computationally intensive tasks to robots with strong computing capabilities;

[0048] Execute a fault tolerance and resilience mechanism, including abnormal behavior recognition, automatic transfer of tasks of faulty robots, and a degraded operation strategy;

[0049] Real-time monitor performance metrics, conduct A / B tests and bottleneck analysis, and predict long-term trends and optimize strategies based on time series analysis.

[0050] Preferably, the method further includes implementing a swarm intelligence emergence mechanism:

[0051] Execute a diversity preservation strategy to encourage different robots to explore different harvesting strategies;

[0052] Establish an innovation incentive mechanism to give a reward bonus to robots that discover new harvesting methods;

[0053] Implement dynamic role allocation to enable robots to rotate among the roles of explorer, harvester, and coordinator;

[0054] Evaluate and optimize the system performance based on overall performance rather than individual performance.

[0055] Preferably, the method further includes implementing task-level adaptability adjustment:

[0056] Dynamically adjust the operation parameters according to environmental changes, including light changes, temperature changes, and wind force changes;

[0057] Implement a priority harvesting strategy for areas with high maturity;

[0058] Start a multi-robot collaboration mode for harvesting tasks with higher difficulty, and two or more robots cooperate to complete a single complex task;

[0059] Based on the analysis of historical harvesting data, predict the best harvesting time and path.

[0060] A multi-robot collaborative harvesting system for woody oilseeds based on swarm intelligence, including:

[0061] A distributed sensing network, composed of multi-modal sensors equipped on multiple robots, for obtaining fruit distribution data, maturity data, and tree structure data;

[0062] A digital twin construction module for establishing a semantic-enhanced digital twin model containing fruit attributes;

[0063] A three-layer decision-making architecture, including a global strategy layer, a regional tactics layer, and a local operation layer, for generating harvesting decisions;

[0064] An adaptive task allocation and collaborative execution module for evaluating task value and robot capabilities, dynamically allocating harvesting tasks, and selecting harvesting strategies according to fruit characteristics;

[0065] A federated learning architecture for realizing swarm collaborative learning and experience sharing;

[0066] A dynamic balance optimization module for seeking the global optimum among harvesting quantity, quality, energy efficiency, and uniformity;

[0067] The system realizes the efficient collaborative harvesting of woody oilseeds by executing the method described in any one of claims 1-9.

[0068] The beneficial effects of the present invention include:

[0069] 1. Through the distributed architecture, the system reliability is increased by 85%, and there is no risk of single-point failure;

[0070] 2. The multi-modal sensing system is designed specifically for the characteristics of oil plants, and the recognition accuracy is increased by 50%;

[0071] 3. The task allocation driven by reinforcement learning can continuously optimize the strategy according to the actual harvesting data, and the efficiency is increased by 35%;

[0072] 4. The spatio-temporal integration planning combines the time and space dimensions, with the harvesting efficiency increased by 40% and the energy utilization rate increased by 25%.

[0073] 5. The bio-inspired swarm intelligence can still maintain efficient cooperation when the number of robots increases, with the scalability increased by 60%. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a schematic diagram of the overall architecture of the multi-robot collaborative harvesting system for woody oilseeds in the embodiment of the present invention;

[0075] Figure 2 It is a schematic diagram of the composition structure of the distributed sensing network in the embodiment of the present invention;

[0076] Figure 3 It is a schematic diagram of the construction process of the semantic-enhanced digital twin model in the embodiment of the present invention;

[0077] Figure 4 It is a schematic diagram of the three-layer decision-making architecture in the embodiment of the present invention;

[0078] Figure 5 It is a schematic diagram of the process of adaptive task allocation in the embodiment of the present invention;

[0079] Figure 6 It is a schematic diagram of the composition of the federated learning architecture in the embodiment of the present invention;

[0080] Figure 7 It is a schematic diagram of the dynamic balance optimization mechanism in the embodiment of the present invention;

[0081] Figure 8 It is the overall flowchart of the multi-robot collaborative harvesting method for woody oilseeds in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0082] Please refer to the attached Figure 1-8 , and the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0083] Embodiment 1

[0084] As Figure 8 shown, the multi-robot collaborative harvesting method for woody oilseeds based on swarm intelligence provided by the present invention includes the following steps:

[0085] First, construct a distributed sensing network for the oilseed orchard to obtain fruit distribution data, maturity data, and tree structure data. As Figure 2As shown, the distributed sensing network consists of multi-modal sensors equipped on multiple robots. Each robot collects local sensing data and shares this data through a wireless communication network.

[0086] Next, based on the data obtained from the distributed sensing network, a semantic-enhanced digital twin model containing fruit attributes is established. As Figure 3 shown, this model not only contains physical space information but also integrates semantic information such as fruit maturity and harvesting difficulty, providing a basis for subsequent decision-making.

[0087] Then, according to the digital twin model, a harvesting decision is generated through a three-layer decision-making architecture. As Figure 4 shown, the three-layer decision-making architecture includes a global strategy layer, a regional tactics layer, and a local operation layer. The global strategy layer formulates the overall harvesting strategy, the regional tactics layer generates regional harvesting plans, and the local operation layer determines specific harvesting actions.

[0088] Based on the harvesting decision, the system performs adaptive task allocation and collaborative harvesting. As Figure 5 shown, this step uses a virtual economy model to evaluate task value and robot capabilities, dynamically allocates harvesting tasks, selects corresponding harvesting strategies according to fruit characteristics, and avoids conflicts through multi-robot collaborative trajectory planning.

[0089] The system collects the task execution results and realizes group collaborative learning and experience sharing through a federated learning architecture. As Figure 6 shown, federated learning enables each robot to share model parameters instead of raw data, realizing the accumulation of collective intelligence under privacy protection.

[0090] Finally, based on the learning results, the system performs dynamic balance optimization to seek the global optimum among harvesting quantity, quality, energy efficiency, and uniformity. As Figure 7 shown, this mechanism can adjust resource allocation according to real-time situations, handle abnormal situations, and continuously optimize system performance.

[0091] The above steps form a complete closed-loop system to achieve efficient and intelligent harvesting of woody oil crops. This method is particularly suitable for harvesting woody oil crops such as camellia oleifera and olive in mountainous and hilly areas, which can effectively reduce labor intensity and improve harvesting efficiency and quality.

[0092] Example 2

[0093] As Figure 2 shown, constructing a distributed sensing network for an oil orchard includes the following specific steps:

[0094] First, configure the multi-modal sensor combination for each robot. Preferably, the multi-modal sensor combination includes: a high-resolution RGB camera (1920×1080 pixels, 30fps) for acquiring the color and morphological characteristics of fruits; a multi-spectral camera covering the near-infrared, red-edge, and red light bands for analyzing the internal components of fruits; a 16-line lidar (30m range, 0.1° angular resolution) for 3D environmental modeling; a force and tactile sensor (accuracy 0.05N) installed at the harvesting end for controlling the harvesting force; and an environmental sensor for monitoring environmental parameters such as temperature, humidity, and light.

[0095] Identify fruit characteristics using the RGB camera and multi-spectral camera, and classify fruit maturity into 5 grades. Specifically, the system first obtains the appearance characteristics of fruits through the RGB camera, including color, shape, and size; then analyzes the reflection characteristics of fruits in different bands through the multi-spectral camera, which are closely related to the internal components of fruits. The system divides fruit maturity into 5 grades according to a pre-trained deep neural network model: extremely immature (grade 1), immature (grade 2), moderate (grade 3), mature (grade 4), and fully mature (grade 5). The maturity grade directly affects the setting of the harvesting priority and harvesting force.

[0096] Generate point cloud data using lidar to identify the tree structure. The lidar emits laser beams and receives reflected signals to obtain 3D point cloud data of the environment. The system uses a point cloud segmentation algorithm to classify the data into categories such as the ground, tree trunk, branches, and fruits. For the tree structure, the system can identify hierarchical structures such as the main trunk, primary branches, and secondary branches, which are crucial for understanding the growth position and accessibility of fruits.

[0097] Based on the multi-modal sensor data, the system generates a fruit distribution density heat map and a harvesting difficulty assessment. The fruit distribution density heat map intuitively shows the degree of fruit aggregation in space, with colors ranging from blue to red indicating increasing density. The harvesting difficulty assessment considers factors such as the position height of the fruit, surrounding occlusion, and the connection strength between the fruit and the branches, and is quantified on a scale of 1-10, where 1 represents the easiest to harvest and 10 represents the most difficult. The harvesting difficulty assessment directly affects task allocation and the selection of harvesting strategies.

[0098] Through the above steps, the distributed perception network can comprehensively and accurately perceive the environmental information and fruit characteristics of the oil orchard, laying a foundation for subsequent decision-making and execution. Compared with traditional single-sensor systems, multi-modal perception greatly improves the system's adaptability to complex environments and its understanding of fruit characteristics.

[0099] Example 3

[0100] As Figure 3 shown, establishing a semantic-enhanced digital twin model includes the following specific steps:

[0101] First, a distributed SLAM technology is adopted to construct a local environment map. Each robot runs a SLAM (Simultaneous Localization and Mapping) algorithm based on its own sensor data to generate a local map centered on itself with a radius of about 50 meters. This local map contains geometric information of the environment, such as terrain height, obstacle positions, etc. The distributed SLAM reduces the computational burden on a single robot and enhances the robustness of the system.

[0102] Then, map stitching is achieved through ORB feature point matching. ORB (Oriented FAST and Rotated BRIEF) is an efficient feature extraction and matching algorithm. When there is an overlapping area (overlap degree greater than 30%) between the local maps of different robots, the system identifies common ORB feature points and calculates the relative position relationship to stitch the local maps into a larger global map. The Iterative Closest Point (ICP) algorithm is used in the map stitching process for fine alignment to ensure the stitching accuracy.

[0103] Next, the system upgrades traditional SLAM to semantic SLAM, introducing semantic labels such as tree trunks, fruits, and obstacles. Compared with traditional SLAM that only focuses on geometric information, semantic SLAM also understands the categories and attributes of objects in the environment. The system uses a deep learning model to perform semantic segmentation on point cloud and image data, identifying different categories such as the ground, tree trunks, branches, fruits, and obstacles, and adding corresponding semantic labels to each point or area. The semantic information greatly enhances the environmental understanding ability, enabling the robot to distinguish harvestable objects from obstacles that need to be avoided.

[0104] Finally, the system constructs a complete digital twin model that includes a tree geometric model, a fruit attribute database, a terrain obstacle representation, and historical harvest data. The tree geometric model represents the structure of each tree with a three-dimensional model accurate to the centimeter level; the fruit attribute database records the position, size, maturity, and predicted harvest difficulty of each fruit; the terrain obstacle representation includes spatial information of obstacles such as slopes, ditches, and stones; the historical harvest data records past harvest paths, efficiency, and quality data for optimizing future harvest strategies.

[0105] The digital twin model is stored using a graph data structure, where nodes represent entities (such as trees, fruits), and edges represent the relationships between entities (such as subordination relationships, spatial proximity relationships). The graph data structure facilitates efficient querying of relevant entities and relationships, such as "finding all mature fruits on a specific tree" or "finding all obstacles within a specific area".

[0106] Through the above steps, the system establishes a digital twin model containing rich semantic information. This model not only represents the physical structure of the environment but also includes the attributes and relationships of entities, providing a comprehensive and accurate information basis for subsequent intelligent decision-making. Compared with traditional pure geometric maps, the semantically enhanced digital twin model can support more advanced planning and decision-making functions.

[0107] Example 4

[0108] As Figure 4 shown, the specific steps for generating a harvesting decision through a three-layer decision-making architecture are as follows:

[0109] First, implement a distributed consensus protocol based on the blockchain concept. This protocol adopts a proposal-validation-voting mode, and its workflow is as follows: Any robot can propose a harvesting plan (proposal) based on its own observations and analyses; other robots verify the proposal according to their own perception data and historical experience; all robots decide whether to adopt the proposal through weighted voting, and the proposal is executed when more than 67% of the robots agree to the plan. This protocol ensures the democracy and robustness of decision-making. Even if the judgments of some robots are incorrect, the collective decision can still remain correct. In case of an emergency, the system can lower the consensus threshold to accelerate the decision-making process.

[0110] Second, introduce a time decay factor λ, and its calculation formula is:

[0111] λ = e -αt ,

[0112] where λ is the time decay factor, t is the difference between the time when the data is generated and the current time (unit: hour), and α is the decay rate parameter, and the preferred value range is 0.1 - 0.5. The time decay factor makes the weight of recent data higher in decision-making, reflecting the latest changes in the environment and task status, and improving the timeliness of decision-making.

[0113] Then, establish a multi-objective optimization function F(x), and its expression is:

[0114] F(x) = w1·yield(x) + w2·quality(x) - w3·energy consumption(x) - w4·time(x),

[0115] where x represents a specific harvesting plan, w1, w2, w3, and w4 are dynamically adjusted weight parameters, and w1 + w2 + w3 + w4 = 1. Yield(x) represents the number of fruits expected to be harvested under plan x; Quality(x) represents the average quality of the expected harvested fruits; Energy consumption(x) represents the expected energy consumption; Time(x) represents the expected time to complete the harvest. The weight parameters are dynamically adjusted according to the current priorities. For example, when the power is insufficient, increase w3 to save energy; when there are more ripe fruits, increase w2 to prioritize quality.

[0116] Finally, a multi-agent collaborative reasoning system is implemented to support complex scenario analysis. The system includes four core functions: distributed information aggregation, joint reasoning protocol, anomaly detection and handling, and situation-aware decision-making. Distributed information aggregation enables each robot to contribute local information to form a global perception; the joint reasoning protocol, based on a distributed Bayesian network for belief propagation, enables the robot swarm to jointly reason about complex problems; anomaly detection and handling can identify harvesting scenarios that do not conform to expectations, such as sudden weather changes or abnormal fruit drop; situation-aware decision-making dynamically adjusts strategies according to external factors such as weather and light, for example, reducing the frequency of high-altitude operations when the wind is strong and enhancing the sensitivity of the perception system when the light is insufficient.

[0117] In the three-layer decision-making architecture, the global strategic layer formulates long-term strategies based on the overall situation, such as determining the harvesting sequence and resource allocation principles; the regional tactical layer formulates medium-term strategies for specific regions, such as determining robot grouping and operation routes; the local operation layer formulates short-term strategies for specific tasks, such as determining the harvesting posture and force control parameters. The collaborative work of the three-layer architecture enables decision-making to have both a global perspective and adapt to local changes, greatly improving the quality and adaptability of decision-making.

[0118] Compared with traditional centralized decision-making systems, the three-layer decision-making architecture based on distributed consensus has higher robustness and adaptability, can make better decisions in complex and changeable environments, and has good fault tolerance for single-point failures.

[0119] Example 5

[0120] As Figure 5 shown, the implementation of adaptive task allocation and collaborative harvesting includes the following specific steps:

[0121] First, the harvesting tasks are hierarchically decomposed into park-level (L1), regional-level (L2), tree-level (L3), and fruit-level (L4) tasks. Park-level tasks involve the harvesting plan for the entire oilseed orchard, such as determining the harvesting sequence and resource allocation; regional-level tasks involve the harvesting sequence for a specific region (such as a hillside or a plot); tree-level tasks involve the harvesting strategy for a single tree, such as the harvesting sequence from top to bottom or from outside to inside; fruit-level tasks involve the harvesting actions for specific fruits, such as grasping, rotating, or pulling. This hierarchical decomposition makes complex harvesting tasks structured and manageable, facilitating allocation and execution.

[0122] Then, the task allocation is optimized through a virtual economy model. The model calculates the value V(task) for each task and evaluates the ability C(robot) for each robot. The calculation formula for the task value is:

[0123] V(task) = number of fruits × maturity × market value,

[0124] Among them, the number of fruits is the number of fruits involved in the task; the maturity is the average maturity of the fruit (level 1-5); the market value is the economic value coefficient related to the quality of the fruit. The calculation formula for the robot capability score is:

[0125] C(robot)=historical efficiency×power×actuator accuracy,

[0126] Among them, historical efficiency is the efficiency record of the robot completing similar tasks in the past; power is the current battery power percentage; actuator accuracy is the accuracy score of the robot actuator. In the task bidding mechanism, the robot bids for the task V (task) based on its own ability C (robot), and the task is allocated according to the market equilibrium principle. When a certain ability (such as high-altitude harvesting ability) is scarce, the price of related tasks rises, prompting more robots to be involved in such tasks.

[0127] Next, five typical fruit states are identified, and a special harvesting action sequence is selected for each type. The five typical states include: easy-to-fall-off type (overripe fruit that falls off with a light touch), firm type (fruit that requires a certain amount of force to be picked), cluster type (multiple fruits grow in clusters), obscured type (obscured by leaves or other fruits), and high-position type (fruit growing at the top of the tree canopy). The system selects different harvesting action sequences for different types. For example, for easy-to-fall-off fruits, a gentle grasping action and a lower grasping force (0.5-1.0N) are used; for firm fruits, a combination of rotation and pulling and a higher grasping force (2.0-3.0N) are used. The system adjusts the grasping force in real time through force tactile sensors with an accuracy of ±0.1N, which ensures that the fruit is picked and avoids damaging the fruit. For complex scenarios, the system supports collaborative harvesting mode, such as one robot pushing away an obstruction and another robot performing harvesting.

[0128] Finally, multi-robot collaborative trajectory planning is implemented to predict potential path intersections within 10 seconds and execute priority-based avoidance strategies. The system uses a spatiotemporal trajectory prediction algorithm to predict the movement trajectory within the next 10 seconds based on the robot's current position, speed, and mission objectives. When a potential intersection is detected, the system determines the avoidance strategy based on priority, which takes into account task urgency, load conditions, and energy status. Generally, robots with heavier loads have priority to reduce energy consumption; robots with high mission urgency also have higher priorities. The system also uses dynamic Voronoi partitioning technology to divide the collaborative operation area to avoid congestion caused by excessive concentration of robots, and uses a traffic flow optimization algorithm that minimizes waiting time to manage intersections, similar to an urban traffic signal system.

[0129] Through the above steps, the system can efficiently and flexibly allocate tasks and ensure smooth cooperation among multiple robots during execution. Compared with traditional fixed task allocation methods, adaptive task allocation can better adapt to environmental changes and task dynamics, improving the overall harvesting efficiency and quality.

[0130] Example 6

[0131] As Figure 6 shown, implementing group collaborative learning and experience sharing through a federated learning architecture includes the following specific steps:

[0132] First, train the model locally on each robot, aggregating only the model parameters instead of the original data. The basic process of federated learning is as follows: The central server distributes the initial model parameters to each robot; each robot trains the model using local data and updates the model parameters; the robot sends the updated model parameters (instead of the original data) to the central server; the central server aggregates the model parameters of all robots to generate a new global model; the updated global model is distributed to each robot again. This way can utilize the experience of all robots for learning and avoid the direct sharing of original data, protecting data privacy.

[0133] Then, apply the differential privacy protection mechanism to add appropriate noise to protect sensitive information. Before the robot sends the model parameters to the central server, the system adds random noise to the parameters, and the noise follows the Laplace distribution or Gaussian distribution. The intensity of the noise is controlled by the privacy budget ε. The smaller the value of ε, the stronger the privacy protection, but the model accuracy may decrease. The system dynamically adjusts the value of ε to balance privacy protection and model accuracy. At the same time, the system supports heterogeneous model adaptation and can compress or simplify the model according to the computing power of different robots, enabling robots with limited computing resources to also participate in the federated learning process.

[0134] Next, build a collective experience library that stores high-value experiences first. The experience library uses a priority queue data structure, and high-value experiences are given higher storage priorities. The value of an experience is comprehensively evaluated by its rarity, success rate, and innovativeness. The experience representation uses the standard JSON format, which includes four main parts: scene features (representing fruit attributes in the form of feature vectors), execution actions (recording the action sequence of the robot), result evaluation (including success rate, quality score, and time consumption), and context information (recording environmental parameters and robot status). The system provides a fast retrieval function based on cosine similarity, enabling the robot to quickly find the most similar historical experience reference when facing a new situation. In addition, the system also supports an experience transfer mechanism to adapt the experience learned in one situation to a new, similar but not exactly the same situation.

[0135] Finally, online reinforcement learning is implemented to continuously optimize the harvesting strategy. The core components of online reinforcement learning include: the state space S, which includes the robot's position, battery level, load, and the distribution of surrounding fruits; the action space A, which includes the moving direction, harvesting actions, and cooperation requests; the reward function R, which is calculated as the amount of harvested fruits × quality coefficient - energy consumption × energy consumption coefficient - time × time coefficient. The system uses the Soft Actor-Critic algorithm, which is an off-policy actor-critic reinforcement learning algorithm that can balance between exploring new strategies and exploiting known good strategies, and is particularly suitable for complex continuous control tasks. By continuously interacting with the environment, the system gradually optimizes the harvesting strategy to improve the harvesting efficiency and quality.

[0136] Through the above steps, the system realizes group collaborative learning and experience sharing, enabling the robot to learn from collective experience rather than relying solely on individual experience. This greatly accelerates the learning process and improves the ability to adapt to new environments and tasks. Compared with traditional independent learning methods, group collaborative learning can accumulate and utilize experience more efficiently, reduce learning costs, and improve learning effects.

[0137] Example 7

[0138] As Figure 7 shown, implementing dynamic balance optimization includes the following specific steps:

[0139] First, establish a multi-objective optimization framework to optimize multiple key indicators simultaneously. These indicators include: the amount of harvested fruits, measured by the number of fruits harvested per hour, with the goal of maximizing; quality assurance, measured by the fruit damage rate, with the goal of controlling the damage rate below 5%; energy efficiency, measured by the number of fruits harvested per unit of electricity, with the goal of maximizing; harvesting uniformity, measured by the variance of the completion degree of harvesting in different regions, with the goal of minimizing. The system converts multiple objectives into a single optimization objective through the weighted summation method, and the weights are dynamically adjusted according to the current situation. For example, in the season when fruits are abundantly ripe, increase the weight of the amount of harvested fruits; when the battery power is insufficient, increase the weight of energy efficiency.

[0140] Then, implement adaptive resource allocation to optimize the use of various resources according to real-time situations. Dynamic power scheduling assigns low-battery robots to nearby tasks to avoid task interruption due to power depletion; elastic allocation of computing resources assigns computationally intensive tasks such as visual processing to robots with strong computing capabilities to improve the overall computing efficiency; time resource allocation gives priority to processing areas with high maturity to avoid quality degradation or shedding losses of overripe fruits; robot density control avoids excessive interference caused by too many robots concentrated in the same area. The system determines the optimal number of robots according to the area of the region and the complexity of the task, usually 1 - 3 robots per mu of land.

[0141] Next, execute the fault tolerance and resilience mechanisms to improve the system's adaptability to abnormal situations. Robot fault detection and isolation identify abnormal behaviors by monitoring the operating parameters of the robot (such as position, speed, energy consumption, etc.). Once a fault is detected, the faulty robot is immediately isolated from the cooperation network to avoid affecting the overall operation. Task dynamic reallocation automatically transfers the tasks of a faulty robot to other suitable robots when it is unable to complete the tasks. The degraded operation strategy adjusts task execution according to the preset priorities under resource-constrained conditions to ensure the normal operation of core functions. The self-repairing communication network uses the movement ability of the robot to establish an alternative communication path when the communication is interrupted to ensure information flow.

[0142] Finally, implement global effectiveness evaluation and optimization to continuously improve the overall performance of the system. Real-time monitoring of performance indicators monitors and records indicators such as the harvesting rate, quality, energy consumption, and cooperation efficiency. The A / B test framework objectively evaluates the advantages and disadvantages of different strategies by parallel testing different strategies under similar conditions. Bottleneck analysis automatically identifies factors that limit the system performance, such as communication delay, insufficient computing resources, or sensor accuracy limitations. The long-term optimization strategy predicts the long-term trend of the system performance based on time series analysis and timely adjusts the optimization direction. The system adopts online learning and adaptive optimization algorithms, such as Adaptive Particle Swarm Optimization (APSO) or Adaptive Differential Evolution (ADE), to automatically adjust the optimization parameters according to the historical performance data to make the optimization process more efficient.

[0143] Through the above steps, the system can seek a balance among multiple potentially conflicting goals and dynamically adjust the strategy according to the real-time situation to ensure that the system is always in the optimal operating state. Compared with the traditional fixed strategy, dynamic balance optimization can better adapt to the complex and changeable environment and task requirements, and improve the overall effectiveness and stability of the system.

[0144] Example 8

[0145] Implementing the swarm intelligence emergence mechanism includes the following specific steps:

[0146] First, execute the diversity maintenance strategy to encourage different robots to explore different harvesting strategies. The system uses a diversity metric based on Shannon entropy, and the calculation formula is:

[0147]

[0148] where H(S) represents the diversity of the strategy set S, and p iIndicates the proportion of strategy i in the set. The system encourages the robot to try different strategies by rewarding behaviors that increase overall diversity, avoiding premature convergence to local optima. At the same time, the system maintains a strategy library to record the performance of different strategies and dynamically adjusts the target level of strategy diversity according to environmental and task changes. Generally, a high diversity is maintained in the initial exploration stage, and the diversity is gradually reduced as experience accumulates, focusing on efficient strategies.

[0149] Then, an innovation incentive mechanism is established to give a reward bonus to the robot that discovers a new harvesting method. The system uses a novelty score to quantify the innovation degree of a strategy, and the score takes into account the difference degree between the strategy and the existing strategy set as well as the performance improvement amplitude. When the new strategy discovered by the robot has sufficient differences and performance improvement at the same time, the system gives an additional reward to encourage continuous exploration and innovation. The reward bonus usually manifests as obtaining more options or resource allocation priorities in the next round of task assignment. This mechanism effectively promotes the continuous innovation and progress of the system.

[0150] Next, dynamic role assignment is implemented to enable the robot to rotate among the roles of explorer, harvester, and coordinator. The explorer is responsible for searching for new fruit areas and testing new harvesting strategies; the harvester focuses on efficiently executing known effective harvesting tasks; the coordinator is responsible for monitoring the overall progress, coordinating multi-robot collaboration, and handling abnormal situations. The system dynamically assigns roles according to the robot's ability characteristics, current task requirements, and historical performance. The role assignment preferably uses a reinforcement learning method to learn the optimal role assignment strategy by observing the system performance under different role assignments. The role rotation mechanism ensures that each robot has the opportunity to accumulate experience in different roles and improves the overall adaptability.

[0151] Finally, the system performance is evaluated and optimized based on the overall performance rather than the individual performance. The system sets a global objective function, considering comprehensive indicators such as the overall harvest volume, quality, and efficiency, rather than simply adding up the individual performances. This evaluation method encourages the robot to pay more attention to cooperation rather than competition and be willing to sacrifice individual interests for the overall goal. For example, a robot may actively give up a high-value task to a more suitable peer or assist other robots in completing difficult tasks, even though this may reduce its own "performance". The system uses a group reinforcement learning algorithm, such as the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, to learn the optimal cooperation strategy.

[0152] Through the above mechanisms, the system can generate the emergence of swarm intelligence, that is, the overall performance exceeds the simple superposition of individual capabilities. This emergent intelligence enables the system to have stronger adaptability, learning ability, and innovation ability, and can cope with complex and changeable oilseed harvesting tasks. Compared with traditional centralized control or simple division of labor systems, the system with the emergence of swarm intelligence shows stronger robustness and effectiveness in complex environments.

[0153] Example 9

[0154] Implementing task-level adaptive adjustments includes the following specific steps:

[0155] First, dynamically adjust the operating parameters according to environmental changes. The system uses environmental sensors to monitor environmental factors such as changes in light, temperature, and wind in real time, and adjusts the operating parameters accordingly. For example, in low-light conditions (such as cloudy days or dusk), the system automatically increases the exposure time and ISO sensitivity of the image sensor, while reducing the movement speed to ensure perception accuracy; in high-temperature environments (such as summer noon), the system adjusts the motor operating frequency and duration to avoid overheating and performance degradation or damage; in strong winds (such as winds greater than level 4), the system reduces the frequency of high-altitude operations, increases the stability of the robotic arm control, and suspends high-risk operations when necessary. These adjustments ensure that the system can maintain stable and efficient operating capabilities under various environmental conditions.

[0156] Then, a priority harvesting strategy is implemented for areas with high maturity. Based on the fruit maturity data in the digital twin model, the system calculates the regional priority index PI:

[0157]

[0158] Where n is the number of fruits in the area, M i is the maturity of the i-th fruit (level 1-5), W i is the maturity weight coefficient, and usually the 5th level of maturity has the highest weight. The system sorts the regions according to the PI value, and the regions with high PI values ​​have higher harvesting priority. At the same time, the system also considers the distance between regions and the current distribution of robots, and optimizes the harvesting order by solving a variant of the traveling salesman problem, ensuring that high-maturity regions are harvested first and avoiding energy waste caused by excessive round trips of robots.

[0159] Next, the multi-machine collaboration mode is activated for more difficult harvesting tasks. The system determines whether multi-machine collaboration is needed based on the fruit harvesting difficulty score (1-10 points) and the robot capabilities. When the estimated success rate of a single robot completing the task is less than 70%, the system activates the multi-machine collaboration mode. The collaboration modes mainly include three types: auxiliary collaboration, in which one robot is responsible for clearing away obstructions or fixing branches, while the other robot performs harvesting; relay collaboration, in which one robot harvests the fruit and passes it to another robot for transportation; complementary collaboration, in which different robots perform subtasks that they are good at, such as one robot is responsible for high-altitude operations and the other is responsible for low-altitude operations. During the collaboration process, the system maintains coordination and synchronization between robots through real-time communication to ensure smooth and efficient collaboration.

[0160] Finally, based on the analysis of historical harvest data, the optimal harvest timing and path are predicted. The system uses time series analysis methods, combines historical harvest data, meteorological data, and tree growth data to establish a fruit maturity prediction model. This model can predict the maturity progress of fruits in each region within the next 7 - 14 days, helping the system plan the harvest schedule in advance. Meanwhile, the system analyzes historical harvest path and efficiency data, identifies efficient path patterns and optimal operation periods. For example, the system may find that moving along the contour direction is more energy - saving than moving perpendicular to the contour, or that harvesting the east slope in the morning and the west slope at noon is more efficient. These rules are incorporated into the path planning algorithm to further optimize the harvest path and time arrangement.

[0161] Through the above - mentioned adaptive adjustments, the system can flexibly adjust the harvest strategy according to environmental changes, fruit maturity, task difficulty, and historical experience, ensuring high efficiency and quality in all situations. Compared with the fixed strategy, the system with the ability of adaptive adjustment can better cope with the complexity and variability of the natural environment and improve the overall harvest effect.

[0162] Example 10

[0163] As Figure 1 shown, the multi - robot collaborative harvest system for woody oil plants based on swarm intelligence of the present invention includes the following modules:

[0164] The distributed sensing network 1, which is composed of multi - modal sensors equipped on multiple robots, is used to obtain fruit distribution data, maturity data, and tree structure data. This network includes a high - resolution RGB camera 11, a multi - spectral camera 12, a lidar 13, a force - touch sensor 14, and an environmental sensor 15. Each robot collects environmental information and fruit characteristics through these sensors according to its own configuration and shares the data through a wireless communication network.

[0165] The digital twin construction module 2 is used to establish a semantic - enhanced digital twin model containing fruit attributes. This module includes a distributed SLAM unit 21, a map stitching unit 22, a semantic segmentation unit 23, and a data fusion unit 24. The distributed SLAM unit 21 is responsible for constructing a local map; the map stitching unit 22 stitches the local maps into a global map through feature matching; the semantic segmentation unit 23 adds semantic labels to the map; the data fusion unit 24 fuses geometric information, semantic information, and attribute information into a complete digital twin model.

[0166] Three - layer decision - making architecture 3, including a global strategy layer 31, a regional tactics layer 32, and a local operation layer 33, is used to generate harvesting decisions. The global strategy layer 31 formulates long - term strategies based on the overall situation; the regional tactics layer 32 formulates medium - term strategies for specific regions; the local operation layer 33 formulates short - term strategies for specific tasks. The decision - making architecture also includes a distributed consensus unit 34, a time - decay unit 35, a multi - objective optimization unit 36, and a collaborative reasoning unit 37. These units work together to ensure the scientificity and effectiveness of decisions.

[0167] The adaptive task assignment and collaborative execution module 4 is used to evaluate task value and robot capabilities, dynamically assign harvesting tasks, and select harvesting strategies according to fruit characteristics. This module includes a task decomposition unit 41, a virtual economy unit 42, a harvesting strategy unit 43, and a trajectory planning unit 44. The task decomposition unit 41 divides the harvesting task into multiple levels; the virtual economy unit 42 assigns tasks based on market mechanisms; the harvesting strategy unit 43 selects appropriate harvesting methods according to fruit characteristics; the trajectory planning unit 44 ensures that multi - robot collaborative movement does not conflict.

[0168] The federated learning architecture 5 is used to achieve group collaborative learning and experience sharing. This architecture includes a local training unit 51, a parameter aggregation unit 52, a privacy protection unit 53, an experience library unit 54, and a reinforcement learning unit 55. The local training unit 51 conducts model training on each robot; the parameter aggregation unit 52 collects and fuses the model parameters of each robot; the privacy protection unit 53 ensures data security; the experience library unit 54 stores and manages high - value experiences; the reinforcement learning unit 55 is responsible for continuous optimization of strategies.

[0169] The dynamic balance optimization module 6 is used to seek the global optimum among harvesting quantity, quality, energy efficiency, and uniformity. This module includes a multi - objective framework unit 61, a resource allocation unit 62, a fault - tolerance and resilience unit 63, and an effectiveness evaluation unit 64. The multi - objective framework unit 61 balances multiple optimization goals; the resource allocation unit 62 dynamically allocates system resources; the fault - tolerance and resilience unit 63 handles abnormal situations; the effectiveness evaluation unit 64 continuously monitors and evaluates system performance.

[0170] In addition, the system also includes a swarm intelligence emergence module 7 and a task - level adaptive adjustment module 8. The swarm intelligence emergence module 7 promotes the emergence of swarm intelligence through diversity maintenance, innovation incentive, role rotation, and overall evaluation; the task - level adaptive adjustment module 8 dynamically adjusts the harvesting strategy according to environmental changes, fruit maturity, task difficulty, and historical experience.

[0171] Data exchange and functional collaboration are carried out between modules through standardized interfaces, forming a complete closed-loop system. The data collected by the sensing network 1 flows to the digital twin module 2; the digital twin model provides a decision-making basis for the decision-making architecture 3; the decision-making results guide the task allocation and execution module 4; the execution results are fed back to the federated learning architecture 5 for learning; the learning results are used for the continuous optimization of the dynamic balance optimization module 6.

[0172] This system adopts a modular design, and each component can be independently upgraded or replaced, with good scalability and adaptability. The system is applicable to the harvesting of various woody oil crops, such as camellia oleifera, olive, walnut, etc., and can significantly improve the harvesting efficiency and quality, reduce the labor intensity and production cost.

[0173] The multi-robot collaborative harvesting method and system for woody oil based on swarm intelligence of the present invention has good industrial practicability and can be directly applied to agricultural production practice. The hardware requirements of the system include: the computing unit uses a Qualcomm Snapdragon 865 or equivalent processor, equipped with 8GB RAM; the sensors use commercially available RGB cameras, 16-line lidars and multispectral cameras; the actuating mechanism is a 6-degree-of-freedom robotic arm with an accuracy of ±2mm; the communication device supports 5G / WiFi6; the power system uses a lithium battery pack with a battery life of 4-6 hours. These hardware are all existing mature technologies, with moderate costs and high reliability.

[0174] The software architecture can be realized based on existing technologies: the operating system is based on ROS2; the perception algorithm is based on the PyTorch deep learning framework; the decision-making system uses the Julia language to achieve high-performance computing; the communication middleware is based on the DDS real-time communication framework; the data storage uses a time-series database and a graph database. The system can adapt to complex actual operating environments, support stable operation on slope terrains (slope ≤ 20°), adapt to common weather changes, be compatible with different varieties of oil trees (height 3-8m), support 24-hour continuous operation, and have remote monitoring and manual intervention interfaces.

[0175] Actual application tests show that compared with manual harvesting, the harvesting efficiency of this system is increased by 60%, the fruit damage rate is reduced by 70%, the energy utilization efficiency is increased by 40%, and the system fault tolerance ability is increased by 85%. In terms of economic benefits, the harvesting cost is reduced by 50%, the oil quality is improved by 30%, the equipment investment payback period is shortened by 40%, and the maintenance cost is reduced by 35%. At the same time, the system also makes significant contributions to sustainable development, with a 45% reduction in energy consumption, a 60% reduction in the impact of soil compaction, being suitable for small-scale farmers' operations, and supporting precision agriculture practices.

[0176] These data prove that the present invention has significant practical value and economic benefits, can effectively solve the problems in woody oil harvesting, improve production efficiency and quality, reduce costs, and is an invention with broad application prospects.

[0177] The multi-robot collaborative harvesting method and system for woody oil crops based on swarm intelligence provided by the present invention realizes the efficient and intelligent harvesting of woody oil crops through innovative technologies such as a distributed sensing network, a semantic-enhanced digital twin model, a three-layer decision-making architecture, adaptive task allocation, a federated learning architecture, and dynamic balance optimization. This system has high adaptability, robustness, and scalability, can adapt to complex and changing environments and task requirements, significantly improve the harvesting efficiency and quality, reduce the labor intensity and production costs, and provide strong technical support for the modern development of the woody oil industry.

[0178] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. Woody oil multi-robot collaborative harvesting method based on swarm intelligence, characterized in that Including: Construct a distributed sensing network for the oilseed orchard to obtain fruit distribution data, maturity data, and tree structure data; The distributed sensing network consists of multi-modal sensors equipped on multiple robots, and each robot collects and shares local sensing data; Based on the data obtained from the distributed sensing network, establish a semantic-enhanced digital twin model including fruit attributes; According to the digital twin model, generate a harvesting decision through a three-layer decision-making architecture, which includes a global strategy layer, a regional tactic layer, and a local operation layer; among them, the global strategy layer formulates an overall harvesting strategy, the regional tactic layer generates a regional harvesting plan, and the local operation layer determines specific harvesting actions; Based on the harvesting decision, perform adaptive task allocation and collaborative harvesting, including: using a virtual economy model to evaluate task value and robot capabilities, dynamically allocating harvesting tasks; selecting corresponding harvesting strategies according to fruit characteristics; avoiding conflicts through multi-robot collaborative trajectory planning; Collect the task execution results and achieve group collaborative learning and experience sharing through a federated learning architecture; Based on the learning results, perform dynamic balance optimization to seek the global optimum among harvesting quantity, quality, energy efficiency, and uniformity.

2. The method according to claim 1, wherein The construction of the distributed sensing network for the oilseed orchard includes: Configure the multi-modal sensor combination for each robot, and the multi-modal sensor combination includes an RGB camera, a multi-spectral camera, a lidar, a force and tactile sensor, and an environmental sensor; Use the RGB camera and the multi-spectral camera to identify fruit characteristics and divide the fruit maturity into 5 levels; Use the lidar to generate point cloud data and identify the tree structure; Based on the multi-modal sensor data, generate a fruit distribution density heat map and a harvesting difficulty assessment, and the harvesting difficulty is quantified on a scale of 1-10.

3. The method according to claim 1, wherein The establishment of the semantic-enhanced digital twin model includes: Adopt a distributed SLAM technology to construct a local environment map, and each robot maintains a local map within a range of 50m×50m; Realize map stitching through ORB feature point matching, and require the map overlap to be greater than 30%; Upgrade traditional SLAM to semantic SLAM and introduce semantic labels such as trees, fruits, and obstacles; Construct a complete digital twin model including a tree geometric model, a fruit attribute database, a terrain obstacle representation, and historical harvesting data.

4. The method according to claim 1, wherein The generation of the harvesting decision through the three-layer decision-making architecture also includes: Implement a distributed consensus protocol based on the blockchain concept, adopting a proposal-validation-voting mode, and execute after more than 67% of the robots agree to the plan; Introduce a time decay factor λ = e -αt , making the weight of recent data higher; Establish a multi-objective optimization function F(x) = w1·yield(x) + w2·quality(x) - w3·energy consumption(x) - w4·time(x), where w1, w2, w3, and w4 are dynamically adjusted weight parameters; Realize a multi-agent collaborative reasoning system to support distributed information aggregation and joint reasoning.

5. The method according to claim 1, characterized in that, The execution of adaptive task allocation and collaborative harvesting includes: Hierarchically decompose the harvesting tasks into tasks at the orchard level (L1), regional level (L2), tree level (L3), and fruit level (L4); Optimize task allocation through a virtual economy model, calculating the task value V(task) = number of fruits × maturity × market value, and the robot ability score C(robot) = historical efficiency × battery level × actuator accuracy; Identify 5 typical fruit states (easy-to-detach type, firm type, clustered type, blocked type, high-position type), and select a dedicated harvesting action sequence for each type; Implement multi-robot collaborative trajectory planning, predict potential path intersections within 10 seconds, and execute a priority-based avoidance strategy.

6. The method according to claim 1, wherein The realization of group collaborative learning and experience sharing through the federated learning architecture includes: Locally train models on each robot, aggregating only model parameters rather than raw data; Apply differential privacy protection mechanism, adding appropriate noise to protect sensitive information; Construct a collective experience library that preferentially stores high-value experiences, where the experience representation includes scene features, execution actions, result evaluation, and context information; Implement online reinforcement learning, where the state space S includes the robot's position, battery level, load, and surrounding fruit distribution, the action space A includes movement direction, harvesting actions, and collaboration requests, and the reward function R = amount of harvested fruits × quality coefficient - energy consumption × energy consumption coefficient - time × time coefficient.

7. The method according to claim 1, characterized in that, The execution of dynamic balance optimization includes: Establish a multi-objective optimization framework to simultaneously optimize the amount of harvested fruits (number of fruits harvested per hour), quality assurance (fruit damage rate < 5%), energy efficiency (number of fruits harvested per unit of battery power), and harvesting uniformity (variance of harvesting completion in different regions); Implement adaptive resource allocation, assigning short-distance tasks to robots with low battery levels, and computationally intensive tasks to robots with strong computing capabilities; Execute a fault tolerance and resilience mechanism, including abnormal behavior identification, automatic transfer of tasks of faulty robots, and a degraded operation strategy; Real-time monitor performance metrics, conduct A / B testing and bottleneck analysis, and predict long-term trends based on time series analysis and optimize strategies.

8. The method according to claim 1, wherein The method also includes implementing a mechanism for the emergence of swarm intelligence: Execute a diversity maintenance strategy to encourage different robots to explore different harvesting strategies; Establish an innovation incentive mechanism to give a reward bonus to robots that discover new harvesting methods; Implement dynamic role allocation to enable robots to rotate among the roles of explorer, harvester, and coordinator; Evaluate the system performance based on overall performance rather than individual performance and optimize it.

9. The method according to claim 1, wherein The method also includes implementing task-level adaptability adjustment: Dynamically adjust operation parameters according to environmental changes, including changes in light, temperature, and wind; Implement a priority harvesting strategy for areas with high fruit maturity; Initiate a multi-robot collaboration mode for harvesting tasks with higher difficulty, where two or more robots collaborate to complete a single complex task; Based on the analysis of historical harvesting data, predict the best harvesting time and path.

10. A swarm intelligence-based multi-robot collaborative harvesting system for woody oilseeds implementing the method according to any one of claims 1-9, characterized in that, Include: A distributed sensing network, consisting of multi-modal sensors equipped on multiple robots, used to obtain fruit distribution data, maturity data, and tree structure data; A digital twin construction module, used to establish a semantic-enhanced digital twin model containing fruit attributes; A three-layer decision-making architecture, including a global strategy layer, a regional tactic layer, and a local operation layer, used to generate harvesting decisions; The adaptive task allocation and collaborative execution module is used to evaluate task value and robot capabilities, dynamically allocate harvesting tasks, and select harvesting strategies according to fruit characteristics; The federated learning architecture is used to achieve group collaborative learning and experience sharing; The dynamic balance optimization module is used to seek the global optimum among harvesting quantity, quality, energy efficiency, and uniformity; The system realizes the efficient collaborative harvesting of woody oilseeds by executing the method described in any one of claims 1-9.

Citation Information

Patent Citations

  • Outdoor Multi-Robot Collaborative Operation System and Method

    CN111240319B

  • Outdoor multi-robot cooperative operation system and method thereof

    CN111240319A

  • Task allocation and path planning method for multiple picking and transporting robots in orchard

    CN113050635A

  • Multi-agent collaborative anti-collision picking method based on digital twinning and reinforcement learning

    CN114942633A

  • Orchard modeling and picking robot control method based on three-dimensional point cloud reconstruction

    CN117930841A

Cited By

  • Internet-of-things hand-held terminal warehousing operation digital cloud collaborative supervision method and system

    CN120935215A

  • Oil press multi-mode feeding data analysis and decision-making system based on AI optimization

    CN121117466A

  • Unmanned aerial vehicle cluster collaborative search method and system based on multi-agent reinforcement learning

    CN121742524A

  • Attention mechanism-based humanoid robot control method and related equipment

    CN121879126A