Multi-robot collaborative scheduling method based on end-side collaboration

Abstract

The invention discloses a multi-robot collaborative scheduling method based on end-side collaboration, and the method remarkably improves the collaborative efficiency and robustness of a system when the system processes complex tasks and deals with dynamic environment changes on the basis of a conventional industrial park multi-robot system through introducing an end-side cloud collaborative swarm intelligence algorithm. According to the invention, the bottleneck in a traditional centralized control mode is broken through, and the problems of reduced cooperation efficiency and out-of-control system caused by system scale enlargement and task complexity are avoided; efficient collaborative operation of the multi-robot system in a large-scale and multi-task scene is realized; besides, according to the method, through real-time task monitoring, dynamic scheduling optimization and global map updating, the flexibility and adaptability of the system in a complex dynamic environment are ensured, so that each robot can make a decision autonomously and quickly respond to task requirements, and finally optimal configuration of resources and accurate execution of tasks are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of multi-robot collaborative scheduling, and specifically, relates to a multi-robot collaborative scheduling method based on end-edge collaboration. Background Art

[0002] In the application scenarios of modern industrial parks, the development of automation technology is advancing at an unprecedented speed, driving the intelligent and efficient production process. With the continuous improvement of automation levels, enterprises have begun to introduce large-scale mobile robot systems to replace traditional manual operations and management. These mobile robots not only undertake a variety of complex tasks such as material handling, production line supply, and warehouse management, but can also achieve autonomous navigation and collaborative operations throughout the park. However, when the number of these robots grows to hundreds or even thousands, the complexity of the system also increases dramatically.

[0003] In such a large system, the traditional simple logical thinking and decision-making mode can no longer meet the increasingly complex task requirements. In the past, a single robot or a small-scale system could be operated and managed through preset rules and centralized control, but this approach is insufficient in large-scale, multi-task scenarios. When performing tasks, each robot must not only consider its own path planning and resource allocation, but also effectively collaborate with other robots to avoid collisions, blockages, and resource competition. However, as the scale of the system expands, this simple logical processing method gradually exposes its shortcomings and is difficult to cope with complex environments that change in real time.

[0004] In this case, the efficiency of group collaboration in the system may drop significantly, and conflicts between robots, task delays, and waste of resources will occur frequently, causing the overall operation of the system to become inefficient or even out of control. As the complexity of tasks increases and the dynamic changes in the environment intensify, the system is prone to bottlenecks and cannot respond and adjust in time, which may eventually lead to the interruption or paralysis of the entire production process.

[0005] In addition, as the scale of the system expands, the disadvantages of the centralized control model become more obvious. The centralized decision-making process often has delays and is difficult to handle a large number of concurrent task requirements, which will further slow down the system response speed and reduce the overall operational efficiency. At the same time, the scalability and flexibility of the system are also limited, making it difficult to quickly adapt to changing needs within the industrial park.

[0006] In view of this, the present invention is proposed. Summary of the invention

[0007] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a multi-robot collaborative scheduling method based on end-edge collaboration, which solves the problems raised in the above-mentioned background technology.

[0008] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0009] A multi-robot collaborative scheduling method based on end-edge collaboration includes: using an end agent, an edge agent, and a cloud agent to interact, and is characterized by comprising the following steps:

[0010] Step (1): The terminal agent receives the task instruction and collects environmental data to preliminarily analyze the matching degree between its own computing power and the task requirements;

[0011] Step (2): The edge agent receives the task request from the end agent, generates a suboptimal scheduling solution based on deep reinforcement learning and convolutional neural network, and sends the solution to the end agent for execution;

[0012] Step (3): The end agent executes the task according to the suboptimal scheduling scheme and evaluates the task execution status in real time through the task cost estimator. If the task progress is abnormal, feedback information is sent to the edge agent.

[0013] Step (4): After receiving the feedback information from the end agent, the edge agent performs secondary optimization on the suboptimal scheduling scheme through the water wave optimization algorithm fused with genetic operators, generates a global optimal scheduling scheme, and sends it to the end agent for execution;

[0014] Step (5): The cloud agent periodically receives the accumulated task execution data from the edge agent and performs non-real-time complex AI model training and global optimization analysis. Finally, the optimized AI model is distributed to the edge agent for inference calculation of subsequent tasks.

[0015] Step (6): The cloud agent integrates the regional maps uploaded by multiple edge agents and the real-time data of the end agent to generate a global map of the industrial park and conducts simulation tests based on the global map to verify the accuracy and availability of the map;

[0016] Step (7): Based on the simulation test results, the cloud agent identifies the unexplored areas in the global map, generates corresponding exploration task instructions, and sends the task instructions to the end agent through the edge agent.

[0017] Optionally, the steps of receiving the task instruction and collecting environmental data to preliminarily analyze the matching degree between its own computing capability and the task requirement are as follows:

[0018] The end agent receives task instructions from the cloud agent through the wireless communication network and collects surrounding environment data in real time through the integrated sensor robot system;

[0019] Evaluate its own computing power based on built-in computing resources to determine currently available computing resources and remaining energy levels;

[0020] Comprehensively analyze environmental data, task requirements and its own computing power to determine whether it has the ability to complete the task independently. If the evaluation result shows that the computing resources of the end agent are insufficient, part of the computing requirements will be transferred to the edge agent.

[0021] Optionally, the steps of receiving a task request from an end agent, generating a suboptimal scheduling solution based on deep reinforcement learning and convolutional neural network, and sending the solution to the end agent for execution are as follows:

[0022] The edge agent receives the task request from the end agent through the wireless communication network;

[0023] The edge agent parses the received task request data, extracts the key feature parameters of the task, and processes these features through a convolutional neural network to map the complex input data into a low-dimensional space;

[0024] The edge agent uses a deep reinforcement learning model to evaluate the state of the task and calculate the reward values ​​of different operations. Then, the reinforcement learning model generates a suboptimal scheduling plan based on the current state and historical data and sends the generated suboptimal scheduling plan back to the end agent through the wireless communication network for execution by the end agent.

[0025] While the end agent is executing the suboptimal scheduling plan, the edge agent continuously monitors the execution of the task. If it detects changes in the environment or anomalies in the task execution, the edge agent will adjust the scheduling plan in real time and send the updated instructions to the end agent.

[0026] Optionally, the end agent executes the task according to the suboptimal scheduling scheme, and evaluates the task execution status in real time through the task cost estimator. If the task progress is abnormal, the steps of sending feedback information to the edge agent are as follows:

[0027] During the task execution process, the terminal agent monitors the progress of the task in real time through the built-in task cost estimator and comprehensively analyzes the execution status of the current task through the task cost estimator to evaluate the matching degree between the actual execution status and the expected suboptimal scheduling plan. If the task execution cost exceeds expectations or the task progress is abnormal, the task cost estimator will identify these abnormalities;

[0028] Once an abnormal situation is detected during task execution, the end agent immediately generates an abnormality report and sends the abnormality report and current task execution data to the edge agent via the wireless communication network;

[0029] After receiving the abnormal information, the edge agent will re-evaluate the task according to the abnormal situation and regenerate the scheduling plan or take other corrective measures;

[0030] While waiting for feedback from the edge agent or a new scheduling plan, the end agent can choose to temporarily stop task execution or continue to execute certain parts of the task without affecting the main task until new instructions are received.

[0031] Optionally, after receiving the feedback information from the end agent, the edge agent performs secondary optimization on the suboptimal scheduling scheme by using the water wave optimization algorithm fused with genetic operators, generates a global optimal scheduling scheme, and sends it to the end agent for execution in the following steps:

[0032] Receive feedback information from the end agent through the wireless communication network and analyze the feedback data sent by the end agent to identify key problems and deviations in task execution;

[0033] Adjust the task model in the suboptimal scheduling plan based on feedback information, and redefine task priority, path planning, and resource allocation elements to adapt to the current task status and environmental changes;

[0034] The water wave optimization algorithm is used to simulate the energy diffusion process during water wave propagation. The diffusion law is used to optimize the path selection and task allocation, and the adjusted task model is preliminarily optimized. On the basis of the water wave optimization algorithm, the edge agent integrates the key operators in the genetic algorithm to help the algorithm jump out of the local optimal solution and find a better global optimal solution.

[0035] By combining the water wave optimization algorithm and genetic operators, the edge agent generates a quadratic optimization scheduling plan. After generating the quadratic optimization scheduling plan, the plan will be verified to ensure its feasibility under the current environment and resource constraints. If necessary, the edge agent will fine-tune the plan.

[0036] The verified global optimal scheduling plan is sent to the end agent through the wireless communication network for execution by the end agent;

[0037] While the end agent is executing the global optimal scheduling solution, the side agent continuously monitors the task execution status. If an anomaly occurs again or the environment changes further, the side agent restarts the optimization process.

[0038] Optionally, the cloud agent periodically receives accumulated task execution data from the edge agent and performs non-real-time complex AI model training and global optimization analysis, and finally distributes the optimized AI model to the edge agent for reasoning calculation of subsequent tasks;

[0039] The cloud agent periodically receives the accumulated task execution data from the edge agent through high-speed wired communication and pre-processes the received task execution data;

[0040] Based on the preprocessed data, the cloud agent extracts key features and labels the data to create training and test sets;

[0041] Perform training of complex AI models based on accumulated data to predict various situations in task execution and optimize future scheduling strategies;

[0042] After the model training is completed, the cloud intelligence will verify the model through cross-validation, simulation testing, etc., and evaluate the performance of the model in actual tasks;

[0043] Based on the trained AI model, the cloud agent performs global optimization analysis, comprehensively considers the coordinated execution of multiple tasks, resource allocation, and environmental changes, and generates a global optimization strategy;

[0044] The cloud agent will send the global optimization strategy generated through optimization analysis to the edge agent to guide the execution and scheduling of subsequent tasks. As the task execution and environmental changes continue, the cloud agent will regularly update the AI ​​model and continuously optimize the global strategy to meet new challenges and task requirements.

[0045] Optionally, the cloud agent integrates the regional maps uploaded by multiple edge agents and the real-time data of the terminal agent to generate a global map of the industrial park; the cloud agent performs simulation tests based on the global map to verify the accuracy and availability of the map, and identifies the steps of unexplored areas:

[0046] The cloud agent receives uploaded regional maps from multiple edge agents through a high-speed wired communication network. These maps are generated by the edge agents in the specific areas they cover.

[0047] The cloud agent calibrates the coordinates based on the overlapped parts of the regional maps, then processes the boundaries of each map to eliminate the errors in the overlapping areas, and further accurately calibrates the relative positions of the regional maps through key landmarks;

[0048] After completing the fusion of the regional maps, the cloud agent combines the real-time data of the end agent with the fused regional map to generate a more detailed global map and processes the integrated data to generate a global map of the industrial park;

[0049] Simulation tests are performed based on the generated global map to verify the accuracy and usability of the global map and to identify potential problems in the map through simulation.

[0050] Optionally, the cloud agent identifies the unexplored area in the global map based on the simulation test results, generates corresponding exploration task instructions, and sends the task instructions to the end agent through the edge agent in the following steps:

[0051] After the simulation test is completed, the cloud agent identifies unexplored areas within the region by analyzing the non-obstacle areas and unlabeled areas at the map boundary;

[0052] After identifying the unexplored area, the cloud agent feeds this information back to the edge agent, and sends new task instructions to the end agent through the edge agent, so that the end agent can continue to explore the unknown area;

[0053] While the end agent is performing the exploration task, the edge agent will continuously monitor the progress of the task and dynamically adjust the task instructions based on the actual execution situation;

[0054] After the end agent completes the task, it feeds back the execution results to the edge agent, which then summarizes the feedback information and uploads it to the cloud agent to update the global map.

[0055] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:

[0056] By introducing a swarm intelligence algorithm that collaborates with the edge and the cloud, the collaborative efficiency and robustness of the system in processing complex tasks and responding to dynamic environmental changes have been significantly improved on the basis of the traditional multi-robot system in the industrial park. The present invention not only breaks through the bottleneck of the traditional centralized control mode, but also avoids the decline in collaborative efficiency and system out-of-control problems caused by the expansion of the system scale and the complexity of tasks. At the same time, through distributed intelligent decision-making and real-time data feedback mechanisms, it realizes the efficient collaborative operation of multi-robot systems in large-scale, multi-task scenarios. In addition, the method of the present application also ensures the flexibility and adaptability of the system in complex dynamic environments through real-time task monitoring, dynamic scheduling optimization and global map updates, so that each robot can make autonomous decisions and respond quickly to task requirements, and ultimately achieve the optimal allocation of resources and precise execution of tasks.

[0057] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0059] In the figure:

[0060] Figure 1 This is a flowchart of the multi-robot collaborative scheduling method;

[0061] Figure 2 The swarm intelligence system architecture that integrates the edge and cloud;

[0062] Figure 3 This is a diagram of the real-time scheduling algorithm for multi-robot collaboration;

[0063] Figure 4 Framework diagram of multi-robot collaborative mapping method.

[0064] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0065] The present invention will now be described in further detail with reference to the accompanying drawings.

[0066] See also Figure 1-4 As shown, in this embodiment, a multi-robot collaborative scheduling method based on end-edge collaboration is provided, which adopts the interaction between end agent, edge agent and cloud agent ( Figure 2 ), including the following steps:

[0067] Step (1): The terminal agent receives the task instruction and collects environmental data to preliminarily analyze the matching degree between its own computing power and the task requirements;

[0068] Step (2): The edge agent receives the task request from the end agent, generates a suboptimal scheduling solution based on deep reinforcement learning and convolutional neural network, and sends the solution to the end agent for execution;

[0069] Step (3): The end agent executes the task according to the suboptimal scheduling scheme and evaluates the task execution status in real time through the task cost estimator. If the task progress is abnormal, feedback information is sent to the edge agent.

[0070] Step (4): After receiving the feedback information from the end agent, the edge agent performs a secondary optimization on the suboptimal scheduling plan through the water wave optimization algorithm fused with genetic operators to generate a global optimal scheduling plan, which is then sent to the end agent for execution. The end agent evaluates the task progress in real time during the task execution process and feeds back to the edge agent for adjustment in abnormal situations. The edge agent can also perform a secondary optimization on the scheduling plan based on the new feedback information to ensure the flexibility and robustness of task execution.

[0071] Step (5): The cloud agent periodically receives the accumulated task execution data from the edge agent and performs non-real-time complex AI model training and global optimization analysis. Finally, the optimized AI model is distributed to the edge agent for inference calculation of subsequent tasks.

[0072] Step (6): The cloud agent integrates the regional maps uploaded by multiple edge agents and the real-time data of the end agent to generate a global map of the industrial park and conducts simulation tests based on the global map to verify the accuracy and availability of the map;

[0073] Step (7): Based on the simulation test results, the cloud agent identifies the unexplored areas in the global map, generates corresponding exploration task instructions, and sends the task instructions to the end agent through the edge agent.

[0074] Through deep learning, reinforcement learning and optimization algorithms, the robot system can intelligently generate suboptimal and global optimal scheduling solutions, reducing the need for human intervention and improving the accuracy of task allocation. At the same time, this intelligent task allocation reduces resource waste and enables various intelligent agents to work together more efficiently.

[0075] In this embodiment, the steps of receiving the task instruction and collecting environmental data to preliminarily analyze the matching degree between its own computing capability and the task requirement are as follows:

[0076] The end agent receives task instructions from the cloud agent through the wireless communication network and collects surrounding environment data in real time through the integrated sensor robot system; through the wireless communication network (such as 5G / WiFi6), the end agent can quickly obtain the instructions issued by the cloud agent to ensure the timeliness and accuracy of the task. At the same time, the integrated sensor robot system collects environmental data in real time, which can provide reliable basic information for subsequent task execution, allowing the end agent to make reasonable decisions based on the current environment. This approach not only improves the efficiency of task execution, but also ensures the adaptability and responsiveness of the end agent in dynamic and complex environments.

[0077] Based on the built-in computing resources, the end agent evaluates its own computing power to determine the currently available computing resources and remaining energy level; by evaluating its own computing power through the built-in computing resources, the end agent can better manage its own resources and avoid execution failures due to insufficient resources during task execution. At the same time, this evaluation process provides a key basis for the end agent's decision-making, enabling it to decide whether it can complete the task independently based on its current capabilities. In this way, the end agent can optimize the task execution process and maximize the use of its own resources through reasonable resource allocation and management, thereby improving the success rate and efficiency of task execution.

[0078] By comprehensively analyzing environmental data, task requirements, and its own computing power, the end agent can determine whether it has the ability to complete the task independently. If the evaluation result shows that the computing resources of the end agent are insufficient, part of the computing requirements will be transferred to the edge agent. By comprehensively analyzing environmental data, task requirements, and its own computing power, the end agent can make a judgment on whether to perform the task independently. This process ensures that the execution of the task will not fail due to insufficient computing resources, and improves the robustness and task success rate of the robot system. In addition, when the end agent determines that its own resources are insufficient, it can actively transfer part of the computing tasks to the edge agent for processing, making full use of the overall resource advantages of the robot system. This dynamic allocation of resources and flexible adjustment of tasks not only improves the efficiency of the robot system, but also ensures the smooth completion of tasks in various environments.

[0079] refer to Figure 3 As shown, in this embodiment, the steps of receiving a task request from an end agent, generating a suboptimal scheduling solution based on deep reinforcement learning and convolutional neural network, and sending the solution to the end agent for execution are as follows:

[0080] The edge agent receives task requests from the end agent through the wireless communication network; through the wireless communication network (such as 5G / WiFi6), the edge agent can receive task requests from the end agent in a timely manner. This efficient communication method ensures that each agent in the robot system can quickly respond to changing task requirements and environmental conditions. The timely response capability of the edge agent is the basis for the efficient operation of the entire multi-agent robot system, and can maintain the continuity and accuracy of tasks in complex and dynamic environments.

[0081] The edge agent parses the received task request data, extracts the key feature parameters of the task, and processes these features through a convolutional neural network, mapping the complex input data into a low-dimensional space; by parsing the task request data, the edge agent can identify the key feature parameters in the task, such as task type, environmental complexity, resource requirements, etc. The extraction of these features is the basis of task scheduling and optimization, and the application of convolutional neural networks (CNNs) further improves the efficiency and accuracy of this process. CNNs can map complex high-dimensional data into low-dimensional space, which not only reduces the amount of data processing computation, but also enables subsequent deep learning models to be trained and reasoned more efficiently.

[0082] The edge agent uses a deep reinforcement learning model to evaluate the state of the task and calculate the reward values ​​of different operations. Then, the reinforcement learning model generates a suboptimal scheduling plan based on the current state and historical data and sends the generated suboptimal scheduling plan back to the end agent through the wireless communication network for the end agent to execute. By applying the deep reinforcement learning model, the edge agent can dynamically evaluate the task state and calculate the potential benefits of different operations. This reward-based evaluation mechanism enables the edge agent to select the optimal operation strategy and generate a suboptimal scheduling plan. Moreover, this process not only utilizes historical data, but also combines the current task state, ensuring that the generated scheduling plan has high practicality and efficiency.

[0083] While the end agent is executing the suboptimal scheduling plan, the side agent continuously monitors the execution of the task. If it detects changes in the environment or anomalies in the task execution, the side agent will adjust the scheduling plan in real time and send the updated instructions to the end agent. During the execution of the task, the environment and conditions may change. The side agent can detect these changes or anomalies in time through continuous monitoring. This monitoring mechanism provides a guarantee for the smooth execution of the task, because once a problem is found, the side agent can immediately adjust the scheduling plan to avoid task failure or inefficiency. In addition, the ability to adjust in real time enables the robot system to flexibly respond to various uncertainties in complex environments, improving the robustness and success rate of task execution.

[0084] refer to Figure 3 As shown, in this embodiment, the end agent performs the task according to the suboptimal scheduling scheme, and evaluates the task execution status in real time through the task cost estimator. If the task progress is abnormal, the steps of sending feedback information to the edge agent are as follows:

[0085] During the task execution process, the end agent monitors the progress of the task in real time through the built-in task cost estimator and comprehensively analyzes the execution status of the current task through the task cost estimator, evaluates the match between the actual execution and the expected suboptimal scheduling plan, and if the task execution cost exceeds expectations or the task progress is abnormal, the task cost estimator will identify these abnormalities; this monitoring and evaluation mechanism ensures that the task can be carried out as expected. If there is a deviation, the robot system can identify it in time. This timeliness can effectively prevent the occurrence of task failure or waste of resources. At the same time, the comprehensive analysis capability of the task cost estimator enables the end agent to evaluate the difference between the actual execution and the plan, so that it can take quick action when the task progress is abnormal. This mechanism improves the robustness and accuracy of task execution.

[0086] Once an abnormality is detected during task execution, the end agent immediately generates an abnormality report and sends the abnormality report and current task execution data to the edge agent through the wireless communication network; by generating an abnormality report, the end agent can not only describe the current abnormal state in detail, but also provide key data in execution for edge agent analysis. This fast feedback mechanism shortens the time to identify and respond to problems, allowing the robot system to make adjustments or corrections in the shortest possible time to avoid greater losses. In addition, through communication with the edge agent, the end agent can use more powerful computing resources and more advanced analysis capabilities to obtain optimized solutions, thereby improving the success rate of the task and the overall stability of the robot system.

[0087] After receiving the abnormal information, the edge agent will re-evaluate the task according to the abnormal situation, and regenerate the scheduling plan or take other corrective measures; the re-evaluation helps the robot system to fully understand the root cause and impact of the abnormality, so as to formulate a more effective response strategy. The edge agent can regenerate the scheduling plan based on the current abnormal situation, optimize the execution path and resource allocation of the task, and ensure that the task can continue smoothly. In addition, the edge agent can also take other corrective measures, such as reallocating tasks, adjusting resources, and modifying execution strategies, to cope with the challenges of abnormal situations.

[0088] While waiting for feedback from the edge agent or a new scheduling plan, the end agent can choose to temporarily stop the task execution or continue to execute certain parts of the task without affecting the main task until new instructions are received. The adaptive and flexible execution capabilities of the end agent in the face of uncertainty. When the end agent recognizes anomalies in task execution and sends feedback to the edge agent, it will not wait passively while waiting for a new scheduling plan, but will take the best action strategy based on the actual situation. The end agent can choose to temporarily stop the task execution to prevent resource waste or erroneous operations from causing bigger problems; on the other hand, if some parts of the task can be executed independently and do not affect the completion of the main task, it can also choose to continue to execute these parts. This flexible execution strategy ensures the efficiency of the robot system. Even when waiting for new instructions, the end agent can maximize the use of time and resources, thereby reducing task interruptions and delays.

[0089] refer to Figure 3 As shown, in this embodiment, after the edge agent receives the feedback information of the end agent, the steps of performing secondary optimization on the suboptimal scheduling scheme by using the water wave optimization algorithm integrating the genetic operator to generate the global optimal scheduling scheme and sending it to the end agent for execution are as follows:

[0090] Receive feedback information from the end agent through the wireless communication network and analyze the feedback data sent by the end agent to identify key problems and deviations in task execution;

[0091] According to the feedback information, the task model in the suboptimal scheduling scheme is adjusted, and the task priority, path planning and resource allocation elements are redefined to adapt to the current task status and environmental changes; the task model is adjusted according to the feedback information, so that the edge agent can dynamically respond to the actual situation in task execution, thereby optimizing resource allocation and path planning. This adjustment mechanism enables the robot system to quickly adjust its strategy in the face of environmental changes and uncertainties in task execution to ensure that the task can be completed smoothly. This flexibility not only improves the robustness of the robot system, but also enables the edge agent to maximize the use of robot system resources by optimizing task priorities and resource allocation, thereby improving the overall task execution efficiency.

[0092] The water wave optimization algorithm is applied to simulate the energy diffusion process during water wave propagation, and the diffusion law is used to optimize the path selection and task allocation. The adjusted task model is initially optimized. On the basis of the water wave optimization algorithm, the edge agent integrates the key operators in the genetic algorithm to help the algorithm jump out of the local optimal solution and find a better global optimal solution. By applying the water wave optimization algorithm, the edge agent can simulate the energy diffusion process and achieve preliminary optimization of path selection and task allocation. More importantly, by integrating the key operators in the genetic algorithm, the edge agent can effectively jump out of the local optimal solution and explore a better global solution. This advanced combination of optimization algorithms ensures the efficient operation of the robot system in a complex and dynamic environment, makes the task scheduling plan more comprehensive and accurate, and improves the overall execution effect and resource utilization of the robot system.

[0093] By combining the water wave optimization algorithm and genetic operators, the edge agent generates a secondary optimized scheduling plan. After generating the secondary optimized scheduling plan, the plan will be verified to ensure its feasibility under the current environment and resource constraints. If necessary, the edge agent will fine-tune the plan. Through secondary optimization, the edge agent generates a finely tuned scheduling plan, which will be strictly verified after generation to ensure its feasibility and applicability. This verification and fine-tuning mechanism enables the edge agent to provide high-quality scheduling plans in complex task environments and ensure the accuracy and reliability of task execution. This cautious attitude minimizes the risks and errors in task execution and ensures the stable operation of the robot system.

[0094] The verified global optimal scheduling scheme is sent to the end agent through the wireless communication network for execution by the end agent; it is quickly sent to the end agent through the wireless communication network to ensure that the end agent can obtain the latest execution instructions in time. This fast task delivery and response mechanism enables the end agent to quickly adjust the task execution strategy to avoid task failure or resource waste due to delays.

[0095] While the end agent is executing the global optimal scheduling solution, the side agent continuously monitors the task execution. If an anomaly occurs again or the environment changes further, the side agent restarts the optimization process. Through continuous monitoring, the side agent can grasp the status of task execution in real time, and once a new anomaly or environmental change is detected, a new optimization process can be started immediately. This real-time monitoring and dynamic adjustment mechanism ensures the flexibility and adaptability of task execution, allowing the robot system to maintain efficient and stable operation in a complex and dynamic environment.

[0096] See also Figure 4 As shown, in this embodiment, the cloud agent periodically receives the accumulated task execution data from the edge agent and performs non-real-time complex AI model training and global optimization analysis, and finally distributes the optimized AI model to the edge agent for reasoning calculation of subsequent tasks.

[0097] The cloud agent periodically receives the accumulated task execution data from the edge agent through high-speed wired communication and pre-processes the received task execution data;

[0098] Based on the preprocessed data, the cloud agent extracts key features and labels the data to create training sets and test sets. By extracting key features, the cloud agent can grasp the most important information in the task execution data, which directly affects the training effect of the AI ​​model. Data labeling is an indispensable part of supervised learning. It classifies the data and assigns corresponding labels to ensure that the model can correctly understand the meaning of the data.

[0099] The training of complex AI models based on accumulated data is used to predict various situations in task execution and optimize future scheduling strategies. Through the accumulated task execution data, cloud agents can train complex AI models that can predict various situations that may occur in task execution, enabling the robot system to foresee potential problems and make corresponding adjustments before the actual task execution. Through this forward-looking scheduling strategy, the robot system can maximize the success rate of task execution, reduce resource waste, and effectively cope with complex task environments.

[0100] After the model training is completed, the cloud agent will verify the model through cross-validation, simulation testing, etc., and evaluate the performance of the model in actual tasks; through cross-validation and simulation testing, the cloud agent can evaluate the performance of the model in different tasks and environments to ensure its reliability in actual applications. This verification process can identify potential weaknesses of the model and provide feedback for further optimization.

[0101] Based on the trained AI model, the cloud agent performs global optimization analysis, comprehensively considers the coordinated execution of multiple tasks, resource allocation, and environmental changes, and generates a global optimization strategy;

[0102] The cloud agent sends the global optimization strategy generated through optimization analysis to the edge agent to guide the execution and scheduling of subsequent tasks. As the task execution and environmental changes continue, the cloud agent will regularly update the AI ​​model and continuously optimize the global strategy to meet new challenges and task requirements. By sending the global optimization strategy to the edge agent, the cloud agent ensures that each part of the robot system can perform tasks according to the optimal solution. This top-down strategy distribution mechanism ensures efficient coordination and optimal resource allocation of the robot system. In addition, as the task is executed and the environment changes, the cloud agent will regularly update the AI ​​model and optimize the global strategy. This continuous optimization mechanism ensures that the robot system can always meet new challenges and task requirements, and maintain the long-term stability and efficiency of the robot system.

[0103] See also Figure 4 As shown, in this embodiment, the cloud agent integrates the regional maps uploaded by multiple edge agents and the real-time data of the terminal agent to generate a global map of the industrial park; the cloud agent performs simulation tests based on the global map to verify the accuracy and availability of the map, and the steps of identifying unexplored areas are:

[0104] The cloud agent receives uploaded regional maps from multiple edge agents through a high-speed wired communication network. These maps are generated by edge agents in the specific areas they cover. The cloud agent calibrates the coordinates based on the overlaps between the regional maps, then processes the boundaries of each map to eliminate errors in the overlapping areas, and further accurately calibrates the relative positions of the regional maps through key landmarks. By receiving regional maps from multiple edge agents, the cloud agent is able to integrate each local information to generate a global map covering the entire industrial park. This process takes advantage of the local advantages of each edge agent to ensure that the map can reflect the unique characteristics of each area in detail. By processing overlaps, calibrating coordinates, and eliminating errors, the cloud agent generates a highly accurate map to ensure seamless connection between regions.

[0105] After completing the fusion of regional maps, the cloud agent combines the real-time data of the end agent with the fused regional map to generate a more detailed global map and processes the integrated data to generate a global map of the industrial park; simulation tests can identify potential problems in the map, such as coordinate deviations, inaccurate boundaries, uncovered areas, etc., so that the cloud agent can further optimize the map before application.

[0106] Simulation tests based on the generated global map verify the accuracy and usability of the global map and identify potential problems in the map through simulation. The identification capability enables the robotic system to foresee possible challenges before the mission begins and solve them in advance, thereby reducing possible failures or delays during mission execution.

[0107] See also Figure 4 As shown, in this embodiment, the cloud agent identifies the unexplored area in the global map based on the simulation test results, generates corresponding exploration task instructions, and sends the task instructions to the end agent through the edge agent in the following steps:

[0108] After the simulation test is completed, the cloud agent identifies unexplored areas within the region by analyzing the non-obstacle areas and unmarked areas on the map boundary; by identifying unexplored areas in advance, the robot system can effectively plan subsequent exploration tasks to ensure that every corner of the industrial park is fully covered and monitored, improving the overall efficiency and coverage of the task.

[0109] After identifying the unexplored area, the cloud agent feeds this information back to the edge agent, and sends new task instructions to the end agent through the edge agent, so that the end agent can continue to explore the unknown area; when the end agent performs the task, it can focus on further exploration of the unexplored area, making the exploration process of the robot system more efficient and targeted. This flexibility and pertinence of task allocation helps improve the adaptability and execution efficiency of the robot system in complex environments.

[0110] While the end agent is performing the exploration task, the edge agent will continuously monitor the progress of the task and dynamically adjust the task instructions according to the actual execution situation; through continuous data feedback and global map updates, the robot system can continuously optimize and improve its own operating status. This process not only improves the real-time and accuracy of the global map, but also provides more reliable basic data for the planning and execution of subsequent tasks.

[0111] After the end agent completes the task, it feeds back the execution result to the edge agent, which then summarizes the feedback information and uploads it to the cloud agent to update the global map. This collaborative working mode ensures the overall efficiency and robustness of the robot system, enabling the multi-robot system to achieve stable and efficient task execution in complex environments.

[0112] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.

Claims

1. A multi-robot collaborative scheduling method based on end-edge collaboration, using end agents, edge agents, and cloud agents to interact, characterized in that: The following steps are involved: Step (1): The terminal agent receives the task instruction and collects environmental data to preliminarily analyze the matching degree between its own computing power and the task requirements; Step (2): The edge agent receives the task request from the end agent, generates a suboptimal scheduling solution based on deep reinforcement learning and convolutional neural network, and sends the solution to the end agent for execution; Step (3): The end agent executes the task according to the suboptimal scheduling scheme and evaluates the task execution status in real time through the task cost estimator. If the task progress is abnormal, feedback information is sent to the edge agent. Step (4): After receiving the feedback information from the end agent, the edge agent performs secondary optimization on the suboptimal scheduling scheme through the water wave optimization algorithm fused with genetic operators, generates a global optimal scheduling scheme, and sends it to the end agent for execution; Step (5): The cloud agent periodically receives the accumulated task execution data from the edge agent and performs non-real-time complex AI model training and global optimization analysis. Finally, the optimized AI model is distributed to the edge agent for inference calculation of subsequent tasks. Step (6): The cloud agent integrates the regional maps uploaded by multiple edge agents and the real-time data of the end agent to generate a global map of the industrial park and conducts simulation tests based on the global map to verify the accuracy and availability of the map; Step (7): Based on the simulation test results, the cloud agent identifies the unexplored areas in the global map, generates corresponding exploration task instructions, and sends the task instructions to the end agent through the edge agent.

2. The multi-robot collaborative scheduling method based on end-edge collaboration according to claim 1 is characterized in that: The steps for the terminal agent to receive the task instruction and collect environmental data to initially analyze the matching degree between its own computing power and the task requirements are as follows: The end agent receives task instructions from the cloud agent through the wireless communication network and collects surrounding environment data in real time through the integrated sensor robot system; Evaluate its own computing power based on built-in computing resources to determine currently available computing resources and remaining energy levels; Comprehensively analyze environmental data, task requirements and its own computing power to determine whether it has the ability to complete the task independently. If the evaluation result shows that the computing resources of the end agent are insufficient, part of the computing requirements will be transferred to the edge agent.

3. The multi-robot collaborative scheduling method based on end-edge collaboration according to claim 1 is characterized in that: The steps for the edge agent to receive the task request from the end agent, generate a suboptimal scheduling plan based on deep reinforcement learning and convolutional neural network, and send the plan to the end agent for execution are as follows: The edge agent receives the task request from the end agent through the wireless communication network; The edge agent parses the received task request data, extracts the key feature parameters of the task, and processes these features through a convolutional neural network to map the complex input data into a low-dimensional space; The edge agent uses a deep reinforcement learning model to evaluate the state of the task and calculate the reward values ​​of different operations. Then, the reinforcement learning model generates a suboptimal scheduling plan based on the current state and historical data and sends the generated suboptimal scheduling plan back to the end agent through the wireless communication network for execution by the end agent. While the end agent is executing the suboptimal scheduling plan, the side agent continuously monitors the execution of the task. If it detects environmental changes or abnormalities in task execution, the side agent will adjust the scheduling plan in real time and send the updated instructions to the end agent.

4. The multi-robot collaborative scheduling method based on end-edge collaboration according to claim 1 is characterized in that: The end agent executes the task according to the suboptimal scheduling scheme and evaluates the task execution status in real time through the task cost estimator. If the task progress is abnormal, the feedback information is sent to the edge agent as follows: During the task execution process, the terminal agent monitors the progress of the task in real time through the built-in task cost estimator and comprehensively analyzes the execution status of the current task through the task cost estimator to evaluate the matching degree between the actual execution status and the expected suboptimal scheduling plan. If the task execution cost exceeds expectations or the task progress is abnormal, the task cost estimator will identify these abnormalities; Once an abnormal situation is detected during task execution, the end agent immediately generates an abnormality report and sends the abnormality report and current task execution data to the edge agent via the wireless communication network; After receiving the abnormal information, the edge agent will re-evaluate the task according to the abnormal situation and regenerate the scheduling plan or take other corrective measures; While waiting for feedback from the edge agent or a new scheduling plan, the end agent can choose to temporarily stop task execution or continue to execute certain parts of the task without affecting the main task until new instructions are received.

5. The multi-robot collaborative scheduling method based on end-edge collaboration according to claim 1 is characterized in that: After receiving the feedback information from the end agent, the edge agent performs secondary optimization on the suboptimal scheduling scheme through the water wave optimization algorithm fused with genetic operators, generates a global optimal scheduling scheme, and sends it to the end agent for execution in the following steps: Receive feedback information from the end agent through the wireless communication network and analyze the feedback data sent by the end agent to identify key problems and deviations in task execution; Adjust the task model in the suboptimal scheduling plan based on feedback information, and redefine task priority, path planning, and resource allocation elements to adapt to the current task status and environmental changes; The water wave optimization algorithm is used to simulate the energy diffusion process during water wave propagation. The diffusion law is used to optimize the path selection and task allocation, and the adjusted task model is preliminarily optimized. On the basis of the water wave optimization algorithm, the edge agent integrates the key operators in the genetic algorithm to help the algorithm jump out of the local optimal solution and find a better global optimal solution. By combining the water wave optimization algorithm and genetic operators, the edge agent generates a quadratic optimization scheduling plan. After generating the quadratic optimization scheduling plan, the plan will be verified to ensure its feasibility under the current environment and resource constraints; The verified global optimal scheduling solution is sent to the end agent through the wireless communication network for execution by the end agent; While the end agent is executing the global optimal scheduling solution, the side agent continuously monitors the task execution status. If an anomaly occurs again or the environment changes further, the side agent restarts the optimization process.

6. The multi-robot collaborative scheduling method based on end-edge collaboration according to claim 1 is characterized in that: The cloud agent periodically receives the accumulated task execution data from the edge agent and performs non-real-time complex AI model training and global optimization analysis, and finally distributes the optimized AI model to the edge agent for reasoning calculation of subsequent tasks. The cloud agent periodically receives the accumulated task execution data from the edge agent through high-speed wired communication and pre-processes the received task execution data; Based on the preprocessed data, the cloud agent extracts key features and labels the data to create training and test sets; Perform training of complex AI models based on accumulated data to predict various situations in task execution and optimize future scheduling strategies; After the model training is completed, the cloud intelligence will verify the model through cross-validation and simulation testing to evaluate the performance of the model in actual tasks; Based on the trained AI model, the cloud agent performs global optimization analysis, comprehensively considers the coordinated execution of multiple tasks, resource allocation, and environmental changes, and generates a global optimization strategy; The cloud agent will send the global optimization strategy generated through optimization analysis to the edge agent to guide the execution and scheduling of subsequent tasks. As the task execution and environmental changes continue, the cloud agent will regularly update the AI ​​model and continuously optimize the global strategy.

7. The multi-robot collaborative scheduling method based on end-edge collaboration according to claim 1 is characterized in that: The cloud agent integrates the regional maps uploaded by multiple edge agents and the real-time data of the end agent to generate a global map of the industrial park. The cloud agent performs simulation tests based on the global map to verify the accuracy and availability of the map and identify the unexplored areas. The steps are as follows: The cloud agent receives uploaded regional maps from multiple edge agents through a high-speed wired communication network. These maps are generated by the edge agents in the specific areas they cover. The cloud agent calibrates the coordinates based on the overlapped parts of the regional maps, then processes the boundaries of each map to eliminate the errors in the overlapping areas, and further accurately calibrates the relative positions of the regional maps through key landmarks; After completing the fusion of the regional maps, the cloud agent combines the real-time data of the end agent with the fused regional map to generate a more detailed global map and processes the integrated data to generate a global map of the industrial park; Simulation tests are performed based on the generated global map to verify the accuracy and usability of the global map and to identify potential problems in the map through simulation.

8. The multi-robot collaborative scheduling method based on end-edge collaboration according to claim 1 is characterized in that: According to the simulation test results, the cloud agent identifies the unexplored areas in the global map, generates corresponding exploration task instructions, and sends the task instructions to the end agent through the edge agent. The steps are as follows: After the simulation test is completed, the cloud agent identifies unexplored areas within the region by analyzing the non-obstacle areas and unlabeled areas at the map boundary; After identifying the unexplored area, the cloud agent feeds this information back to the edge agent, and sends new task instructions to the end agent through the edge agent, so that the end agent can continue to explore the unknown area; While the end agent is performing the exploration task, the edge agent will continuously monitor the progress of the task and dynamically adjust the task instructions based on the actual execution situation; After the end agent completes the task, it feeds back the execution results to the edge agent, which then summarizes the feedback information and uploads it to the cloud agent to update the global map.

Images