A robot intelligent group control method based on machine learning

Abstract

The present invention belongs to the field of multi-robot group control, and specifically is a robot intelligent group control method based on machine learning, the method comprising: synchronization and communication, map drawing, elimination of repeated areas and map merging, closed-loop detection, multi-robot task allocation and performance evaluation and optimization. The present invention adopts SLAM technology and a closed-loop detection algorithm based on cyclic learning, which improves the efficiency of multi-robot exploration of unknown areas, enables multi-robots to perform more accurate regional correction for more complex unknown environments, and improves the adaptability of multi-robots to unknown environments; the present invention adopts an optimized greedy algorithm to allocate tasks to multi-robots, comprehensively considers multiple factors of tasks and robots, searches task allocation space more comprehensively, finds a better allocation scheme, and improves the quality and efficiency of task allocation results. At the same time, the optimized greedy algorithm can allocate tasks to robots in a real-time environment, and improves the real-time performance of multi-robot collaborative control.

Description

Technical Field

[0001] The present invention belongs to the field of multi-robot group control, and specifically is a robot intelligent group control method based on machine learning. Background Art

[0002] Robot swarm control refers to multiple robots working together in a shared environment to complete specific tasks or achieve common goals through mutual cooperation and coordination. Robot swarm control is often used in industrial automation, logistics distribution, and exploration and rescue missions. The challenges faced by robot swarm control include coordination and communication, task allocation and collaborative decision-making, collision avoidance and path planning, etc. In order to achieve effective collaborative control, it is necessary to design appropriate algorithms and control strategies, while taking into account the interaction between robots and the dynamic changes of the environment. The existing robot swarm control methods are inefficient, and when facing unknown environments, they have poor coordination and are difficult to complete collaborative exploration. Summary of the invention

[0003] In view of the problem that the existing robot group control methods have poor coordination when facing unknown environments, the present invention adopts SLAM technology and a closed-loop detection algorithm based on cyclic learning to explore unknown areas and correct the explored areas, thereby reducing the workload of manual intervention and manual repair, improving the efficiency of unknown area exploration, and at the same time increasing the scalability and adaptability of the closed-loop detection algorithm, being able to perform more accurate regional correction for more complex unknown environments, and improving the adaptability of multiple robots to unknown environments; in view of the problem that the existing robot group control methods are inefficient, the present invention adopts an optimized greedy algorithm to assign tasks to multiple robots, which can comprehensively consider multiple factors of tasks and robots, and through continuous iteration and in-depth search, more comprehensively search the task allocation space, find a better allocation plan, and improve the quality and efficiency of task allocation results. At the same time, the optimized greedy algorithm can assign tasks to robots in a real-time environment, thereby improving the real-time performance of multi-robot collaborative control.

[0004] The technical solution adopted by the present invention is as follows: a robot intelligent group control method based on machine learning, specifically comprising the following steps:

[0005] Step S1: Synchronization and communication, initializing the time steps and exploration positions of all robots, using the time synchronization protocol to assimilate the time steps of all robots;

[0006] Step S2: Mapping, each robot uses SLAM technology to generate a dynamic map of the exploration area and its corresponding positions, and each robot explores the exploration area;

[0007] Step S3: Eliminate duplicate areas and merge maps. The exploration areas of all robots that have completed exploration are merged after removing duplicates to obtain the total exploration area.

[0008] Step S4: closed loop detection, using a closed loop detection algorithm based on machine learning to correct the inconsistency of the total exploration area and obtain a final correction map;

[0009] Step S5: multi-robot task allocation, using the greedy algorithm of in-depth optimization to allocate tasks to multiple robots and obtain a task allocation list;

[0010] Step S6: Performance evaluation and optimization, evaluating the performance of the closed-loop detection algorithm and the greedy algorithm for deepening optimization, and optimizing according to actual conditions.

[0011] Furthermore, step S1 specifically includes the following steps:

[0012] Step S11: Determine the communication architecture, and determine the communication architecture and communication method between multiple robots;

[0013] Step S12: Design a communication protocol and define the communication protocol between robots;

[0014] Step S13: establishing a communication channel, establishing a communication channel in the robot system;

[0015] Step S14: Transmitting status and data, the robot transmits its own status information and perception data through the communication channel;

[0016] Step S15: Synchronization mechanism: A synchronization mechanism is introduced to ensure the consistency of time and action between robots.

[0017] Step S16: Data fusion and cooperative perception, using a data fusion algorithm to fuse the perception data of multiple robots to obtain more comprehensive and accurate environmental perception data.

[0018] Further, step S4 specifically includes the following steps:

[0019] Step S41: feature extraction, using edge feature extraction method to extract the features of the total explored area at each moment;

[0020] Step S42: Initialize the machine learning algorithm, select the recurrent neural network as the feature extraction and matching model, and set the loop stop condition;

[0021] Step S43: feature matching, calculating the similarity between the features of the total explored area at the current moment and the features of the total explored area at the previous moment, and obtaining a feature matching result;

[0022] Step S44: loop detection, using the RANSAC method to perform closed-loop detection on the feature matching results to obtain closed-loop detection results;

[0023] Step S45: closed loop correction, when a closed loop exists in the closed loop detection result, the closed loop is corrected using a nonlinear least squares method to obtain a correction map;

[0024] Step S46: repositioning, repositioning the position of each robot according to the correction map;

[0025] Step S47: Machine learning, repeating steps S43 to S46 until the loop stop condition is reached to obtain the final correction map.

[0026] Further, step S5 specifically includes the following steps:

[0027] Step S51: Data collection, collecting data on all tasks, task priorities, robot parameters, robot capabilities and constraints;

[0028] Step S52: Data initialization, setting all tasks and robots to unassigned status, and creating an empty task list;

[0029] Step S53: setting the search depth limit and the maximum number of iterations;

[0030] Step S54: iteratively deepen the optimization greedy algorithm to obtain a task allocation list;

[0031] Further, step S54 specifically includes the following steps:

[0032] Step S541: setting the current search depth to 1 and starting iterative deepening search;

[0033] Step S542: Execute the task selection and robot selection steps of the greedy algorithm of deepening optimization, and select the task and robot according to the current search depth;

[0034] Step S543: assign the selected task to the selected robot, mark the task as assigned, increase the search depth value by 1, and update the task list;

[0035] Step S544: Check whether the current search depth reaches the set search depth limit. If so, jump to step S547; otherwise, continue to step S545;

[0036] Step S545: increase the current search depth, return to step S542, and continue to deepen the search;

[0037] Step S546: If the search reaches the maximum number of iterations or the task cannot be assigned any more, jump to step S547;

[0038] Step S547: Go back to the previous search depth and cancel the previous task allocation;

[0039] Step S548: Check whether there are any unassigned tasks. If all tasks have been assigned, jump to step S549; otherwise, continue to execute step S545;

[0040] Step S549: Output the result, output the final task list, and define it as a task allocation list.

[0041] Further, step S6 specifically includes the following steps:

[0042] Step S61: define performance indicators and clarify the indicators for evaluating performance;

[0043] Step S62: Data set preparation, using robot group control simulation software to simulate the data of unknown areas, all robot tasks, task priorities, robot parameters, robot capabilities and constraints to obtain a simulation data set, and define the simulation data set as a training set;

[0044] Step S63: running an evaluation, using the training set to run the closed-loop detection algorithm and the greedy algorithm for deepening optimization, and evaluating the effects of the closed-loop detection algorithm and the greedy algorithm for deepening optimization;

[0045] Step S64: Analyze the effects of the closed-loop detection algorithm and the greedy algorithm of the deepening optimization to obtain analysis results;

[0046] Step S65: Optimizing algorithms and parameters, optimizing the closed-loop detection algorithm and the greedy algorithm for deepening optimization according to the analysis results;

[0047] Step S66: Optimization verification, verifying the performance of the closed-loop detection algorithm and the greedy algorithm with in-depth optimization, using actual scenarios for verification, and obtaining actual verification results;

[0048] Step S67: loop iteration, continuously evaluating and optimizing the closed-loop detection algorithm and the greedy algorithm for deepening optimization according to the actual verification results.

[0049] The beneficial results achieved by the present invention using the above scheme are as follows:

[0050] (1) In view of the problem that the existing robot group control methods have poor coordination when facing unknown environments, the present invention adopts SLAM technology and a closed-loop detection algorithm based on cyclic learning to explore unknown areas and correct the explored areas, thereby reducing the workload of manual intervention and manual repair, improving the efficiency of unknown area exploration, and increasing the scalability and adaptability of the closed-loop detection algorithm. It can perform more accurate regional correction for more complex unknown environments, thereby improving the adaptability of multiple robots to unknown environments;

[0051] (2) To address the problem of low efficiency of existing robot group control methods, the present invention adopts an optimized greedy algorithm to assign tasks to multiple robots. It can comprehensively consider multiple factors of tasks and robots. Through continuous iteration and in-depth search, it can more comprehensively search the task allocation space and find a better allocation scheme, thereby improving the quality and efficiency of task allocation results. At the same time, the optimized greedy algorithm can assign tasks to robots in a real-time environment, thereby improving the real-time performance of multi-robot collaborative control. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of a robot intelligent group control method based on machine learning provided by the present invention;

[0053] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0055] Example 1: See Figure 1 The present invention provides a robot intelligent group control method based on machine learning, which specifically includes the following steps:

[0056] Step S1: Synchronization and communication, initializing the time steps and exploration positions of all robots, using the time synchronization protocol to assimilate the time steps of all robots;

[0057] Step S2: Mapping, each robot uses SLAM technology to generate a dynamic map of the exploration area and its corresponding positions;

[0058] Step S3: Eliminate duplicate areas and merge maps. The exploration areas of all robots that have completed exploration are merged after removing duplicates to obtain the total exploration area.

[0059] Step S4: closed loop detection, using a closed loop detection algorithm based on machine learning to correct the inconsistency of the total exploration area and obtain a final correction map;

[0060] Step S5: multi-robot task allocation, using the greedy algorithm of in-depth optimization to allocate tasks to multiple robots and obtain a task allocation list;

[0061] Step S6: Performance evaluation and optimization, evaluating the performance of the closed-loop detection algorithm and the greedy algorithm for deepening optimization, and optimizing according to actual conditions.

[0062] Embodiment 2: This embodiment is based on the above embodiment, and step S1 specifically includes the following steps:

[0063] Step S11: Determine the communication architecture, select centralized communication as the communication architecture of the multi-robot system, and determine Bluetooth as the communication protocol and communication method between the multi-robots;

[0064] Step S12: Design a communication protocol, define TCP / IP as the communication protocol between robots, use JSON as the message format, set the communication frequency to 1HZ, and select Bluetooth 1.6 as the communication method to ensure that the protocol can support real-time communication and data transmission, and has reliability and fault tolerance;

[0065] Step S13: establishing a communication channel, using Bluetooth to establish a communication channel in the robot system so that the robots can communicate with each other;

[0066] Step S14: Transmitting status and data. The robot transmits its own status information and perception data through the communication channel, including the position of each robot, the real-time speed of each robot, and the speed sensor data, to ensure the accuracy and real-time nature of the data so that other robots can obtain the latest information.

[0067] Step S15: Synchronization mechanism, introducing a clock synchronization protocol to ensure the consistency of time and action between robots, wherein the Network Time Protocol NTP is selected as the selected protocol;

[0068] Step S16: Data fusion and cooperative perception, using data fusion algorithm to fuse the perception data of multiple robots to obtain more comprehensive and accurate environmental perception data, and using the perception information and task allocation of multiple robots to achieve common understanding of the environment and tasks and collaborative decision-making.

[0069] Embodiment 3: This embodiment is based on the above embodiment, and step S4 specifically includes the following steps:

[0070] Step S41: feature extraction, using edge feature extraction method to extract the features of the total explored area at each moment;

[0071] Step S42: Initialize the machine learning algorithm, select the recurrent neural network as the feature extraction and matching model, set the loop stop condition, set the number of hidden layer neurons to 10, set the learning rate to 0.02%, and set the number of iterations to 10;

[0072] Step S43: feature matching, calculating the similarity between the features of the total explored area at the current moment and the features of the total explored area at the previous moment, and obtaining a feature matching result;

[0073] Step S44: loop detection, using the RANSAC method, setting the number of RANSAC iterations to 15, the number of samples to 10, the accuracy threshold to 99%, performing closed-loop detection on the feature matching results, and obtaining closed-loop detection results;

[0074] Step S45: closed loop correction, when a closed loop exists in the closed loop detection result, the closed loop is corrected using a nonlinear least squares method to obtain a correction map;

[0075] Step S46: repositioning, repositioning the position of each robot according to the correction map;

[0076] Step S47: Machine learning, repeating steps S43 to S46 until the loop stop condition is reached to obtain the final correction map.

[0077] By performing the above operations, in order to solve the problem of poor coordination of existing robot group control methods when facing unknown environments, the present invention adopts SLAM technology and a closed-loop detection algorithm based on cyclic learning to explore unknown areas and correct the explored areas, thereby reducing the workload of manual intervention and manual repair, improving the efficiency of unknown area exploration, and at the same time increasing the scalability and adaptability of the closed-loop detection algorithm, which can perform more accurate area correction for more complex unknown environments, thereby improving the adaptability of multiple robots to unknown environments.

[0078] Embodiment 4: This embodiment is based on the above embodiment, and step S5 specifically includes the following steps:

[0079] Step S51: Data collection, collecting data on all tasks, task priorities, robot parameters, robot capabilities and constraints;

[0080] Step S52: Data initialization, setting all tasks and robots to unassigned status, and creating an empty task list;

[0081] Step S53: setting the search depth limit to 10 and the maximum number of iterations to 20;

[0082] Step S54: Iterate the greedy algorithm of deepening optimization to obtain a task allocation list, which specifically includes the following steps:

[0083] Step S541: setting the current search depth to 1 and starting iterative deepening search;

[0084] Step S542: Execute the task selection and robot selection steps of the greedy algorithm of deepening optimization, and select the task and robot according to the current search depth;

[0085] Step S543: assign the selected task to the selected robot, mark the task as assigned, increase the search depth value by 1, and update the task list;

[0086] Step S544: Check whether the current search depth reaches the set search depth limit. If so, jump to step S547; otherwise, continue to step S545;

[0087] Step S545: increase the current search depth, return to step S542, and continue to deepen the search;

[0088] Step S546: If the search reaches the maximum number of iterations or the task cannot be assigned any more, jump to step S547;

[0089] Step S547: Go back to the previous search depth and cancel the previous task allocation;

[0090] Step S548: Check whether there are any unassigned tasks. If all tasks have been assigned, jump to step S549; otherwise, continue to execute step S545;

[0091] Step S549: Output the result, output the final task list, and define it as a task allocation list.

[0092] By executing the above operations, in order to solve the problem of low efficiency of existing robot group control methods, the present invention adopts an optimized greedy algorithm to assign tasks to multiple robots, which can comprehensively consider multiple factors of tasks and robots, and through continuous iteration and in-depth search, more comprehensive search of task allocation space can be carried out to find a better allocation plan, thereby improving the quality and efficiency of task allocation results. At the same time, the optimized greedy algorithm can assign tasks to robots in a real-time environment, thereby improving the real-time performance of multi-robot collaborative control.

[0093] Embodiment 5: This embodiment is based on the above embodiment, and the performance evaluation and optimization specifically includes the following steps:

[0094] Step S61: define performance indicators and clarify the indicators for evaluating performance. For closed-loop detection, consider the closed-loop detection accuracy, missed detection rate and false detection rate indicators. For the optimized greedy algorithm, consider the task completion time, task balance and robot utilization rate indicators.

[0095] Step S62: Data set preparation, using ROS as the robot group control simulation software, simulating the data of unknown areas, all tasks of the robot, task priorities, robot parameters, robot capabilities and constraints to obtain a simulation data set, and defining the simulation data set as a training set;

[0096] Step S63: Run the evaluation, use the training set to run the closed-loop detection algorithm and the greedy algorithm of deepening optimization, and record the closed-loop detection accuracy, missed detection rate, false detection rate, task completion time, task balance and robot utilization rate, and evaluate the accuracy of the closed-loop detection results and the effect of the task allocation results;

[0097] Step S64: Analyze the effects of the closed-loop detection algorithm and the greedy algorithm of the deepening optimization to obtain analysis results;

[0098] Step S65: Optimize the algorithm and parameters. According to the evaluation results and analysis, optimize the closed-loop detection algorithm and the greedy algorithm for deepening optimization. Use random learning rate update as the parameter setting and optimization method, and set the learning rate to 0.1%, and the random mutation rate to between -1% and 1%. Perform iterative experiments and evaluations to gradually adjust and optimize the algorithm.

[0099] Step S66: Optimization verification, verifying the performance of the optimized closed-loop detection algorithm and the deeply optimized greedy algorithm, using actual scenarios for verification, obtaining actual verification results, and ensuring that the closed-loop detection algorithm and the deeply optimized greedy algorithm can achieve the expected performance in different data sets and scenarios;

[0100] Step S67: Iterate in a loop, and continuously evaluate and optimize according to the verification results of actual closed-loop detection accuracy, missed detection rate, false detection rate, task completion time, task balance and robot utilization rate to improve the performance of the closed-loop detection algorithm and the greedy algorithm for deepening optimization.

[0101] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0102] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0103] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A robot intelligent group control method based on machine learning, characterized by: The following steps are involved: Step S1: Synchronization and communication, initializing the time steps and exploration positions of all robots, using the time synchronization protocol to assimilate the time steps of all robots; Step S2: Mapping, each robot uses SLAM technology to generate a dynamic map of the exploration area and its corresponding positions, and each robot explores the exploration area; Step S3: Duplicate area elimination and map merging: the exploration areas of all robots that have completed exploration are merged after removing duplicates to obtain the total exploration area; Step S4: Closed loop detection, using a closed loop detection algorithm based on machine learning to correct the inconsistencies in the total exploration area and obtain the final correction map, specifically including the following steps: Step S41: feature extraction, using edge feature extraction method to extract the features of the total explored area at each moment; Step S42: Initialize the machine learning algorithm, select the recurrent neural network as the feature extraction and matching model, and set the loop stop condition; Step S43: feature matching, calculating the similarity between the features of the total explored area at the current moment and the features of the total explored area at the previous moment to obtain a feature matching result; Step S44: loop detection, using the RANSAC method to perform closed-loop detection on the feature matching results to obtain closed-loop detection results; Step S45: closed-loop correction: when a closed loop exists in the closed-loop detection result, the closed loop is corrected using a nonlinear least squares method to obtain a correction map; Step S46: repositioning, repositioning the exploration position of each robot according to the correction map; Step S47: Machine learning, repeating steps S43 to S46 until the loop stop condition is reached to obtain the final corrected map; Step S5: Multi-robot task allocation, using the greedy algorithm with in-depth optimization to allocate tasks to multiple robots and obtain a task allocation list, specifically including the following steps: Step S51: Data collection, collecting data on all tasks, task priorities, robot parameters, robot capabilities and constraints; Step S52: Data initialization, setting all tasks and robots to unassigned state, and creating an empty task list; Step S53: setting the search depth limit and the maximum number of iterations; Step S54: Iterate the greedy algorithm of deepening optimization to obtain a task allocation list, which specifically includes the following steps: Step S541: Set the current search depth to 1 and start iterative deepening search; Step S542: Execute the greedy algorithm of deepening optimization to select tasks and robots according to the current search depth; Step S543: Assign the selected task to the selected robot, mark the task as assigned, increase the search depth by 1, and update the task list; Step S544: Check whether the current search depth reaches the set search depth limit. If so, jump to step S547; otherwise, continue to step S545; Step S545: Increase the current search depth, return to step S542, and continue to deepen the search; Step S546: If the search reaches the maximum number of iterations or the task cannot be assigned any further, jump to step S547; Step S547: Go back to the previous search depth and cancel the previous task assignment; Step S548: Check whether there are any unassigned tasks. If all tasks have been assigned, jump to step S549; otherwise, continue to step S545. Step S549: Output the result, output the final task list, and define it as a task allocation list; Step S6: Performance evaluation and optimization, evaluating the performance of the closed-loop detection algorithm and the greedy algorithm with deepening optimization, and optimizing them according to actual conditions.

Images