Cooperative positioning method for intelligent cluster robot

Abstract

The invention provides a cooperative positioning method for an intelligent cluster robot, and the method comprises the following steps: carrying out the real-time collection of environment data and self-state data through a corresponding sensor based on the cooperative positioning demands of the cluster robot, and generating a basic data collection report; based on the single positioning report, information sharing among clusters is carried out through a wireless communication network, data information obtained by all robots in the clusters is exchanged, distributed decision making is carried out in cooperation with an edge computing technology and a decentralized cooperative control strategy, and a cluster information sharing report is generated. Through decentration and edge calculation, the calculation pressure can be dispersed to each robot in the cluster, so that bottleneck of calculation resources of a central server caused by huge scale of data to be processed in a complex environment is avoided, and meanwhile, a robot group can more flexibly share tasks to improve the working capability of the cluster robot.

Description

Technical Field

[0001] The present invention relates to the technical field of robot positioning, and particularly to a collaborative positioning method for intelligent swarm robots. Background Art

[0002] Swarm robots are usually composed of multiple autonomous robots. They draw on the collective behaviors of flocks of birds, schools of fish, etc. in nature, and multiple relatively simple robots coordinate to complete complex tasks. In order to ensure the efficient completion of tasks, swarm robots also need to know each other's positions, so as to facilitate task allocation or dynamic adjustment according to their actual positions.

[0003] In reality, the collaborative positioning of swarm robots usually involves multiple robots. Its core idea is to enhance the environmental perception ability of each robot through the mutual cooperation and information exchange among the robots in the swarm, especially in complex environments or scenarios where GPS signals cannot effectively cover. However, for large-scale swarm robots in complex environments, although the traditional centralized collaborative positioning method can provide a globally optimal solution, it may also face problems such as excessive computational burden, communication bottlenecks, etc. At the same time, the transmission of a large amount of data information may also cause excessive channel load, which may lead to communication delays, increasing the accumulation of positioning errors. In addition, in a dynamic environment, affected by external moving objects, other robots, and obstacles, the positioning effect often cannot reach the level in a static environment, reducing the collaborative ability between robots. Therefore, a collaborative positioning method for intelligent swarm robots is proposed to solve the above problems. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a collaborative positioning method for intelligent swarm robots to solve at least the above problems.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A collaborative positioning method for intelligent swarm robots, the method comprising the following steps:

[0007] Step 1: Based on the collaborative positioning requirements of swarm robots, collect environmental data and its own state data in real time through corresponding sensors, and preprocess various types of data to generate a basic data collection report;

[0008] Step 2: Based on the basic data collection report, perform single-robot positioning one by one through GPS, initially determine the current position, and combine visual technology to match the image features of the scene, and deduce the relative position information of each robot to generate a single-robot positioning report;

[0009] Step 3: Based on the single-unit positioning report, information is shared among clusters through a wireless communication network, data information obtained by each robot in the cluster is exchanged, and distributed decision-making is performed in conjunction with edge computing technology and decentralized collaborative control strategies to generate a cluster information sharing report;

[0010] Step 4: Based on the cluster information sharing report, relative positioning and error correction are performed, the positioning data of different robots are integrated using a multi-sensor data fusion algorithm, a cluster robot constraint graph is constructed, nonlinear optimization is performed, and the cluster robots are synchronously positioned at different time steps in conjunction with a synchronization algorithm to generate a collaborative positioning report;

[0011] Step 5: Based on the collaborative positioning report, collaborate according to the overall goals and tasks of the cluster, assign independent travel directions, adjust the movement speed of each robot in the cluster, locally optimize the position through the set bounding box, combine path planning and dynamic obstacle avoidance algorithm, and generate a cluster collaborative movement report;

[0012] Step 6: Based on the cluster collaborative mobility report, provide real-time feedback, optimize the positioning strategy in a dynamic environment through machine learning, and generate a positioning feedback report.

[0013] As a further solution of the present invention, the environmental data and self-state data include lidar data, visual sensor data, GPS data and IMU data, wherein the lidar data is used to reflect the distance, shape and position information of surrounding objects, the visual sensor data is used to build maps and object recognition, the GPS data is used to preliminarily determine the robot's positioning information, and the IMU data is used to perceive the robot's angular velocity, acceleration and posture changes.

[0014] As a further solution of the present invention, the specific steps of preprocessing various types of data are:

[0015] Based on the various types of collected data, correlation analysis is performed to screen out features with high correlation. At the same time, outlier detection is performed through the Z-Score algorithm to identify points that exceed the set threshold, eliminate redundant data and repeatedly collected data, and generate a data cleaning set;

[0016] Based on the data cleaning set, the missing data is estimated and filled by interpolation, and low-frequency noise is removed by high-pass filtering, and high-frequency useful signals are retained to generate a data sorting set;

[0017] Based on the data collation set, the data is standardized and normalized, all features are mapped into a unified range, and the fairness of each feature in the calculation is ensured by adjusting the data range to generate a data processing set.

[0018] As a further solution of the present invention, the specific steps of combining visual technology to perform image feature matching are:

[0019] Based on the preprocessed image data, feature points of each image in the scene are extracted one by one using the ORB algorithm, including corner points and edge points, to generate a visual feature extraction set;

[0020] Based on the feature extraction set, the similarity of feature points in the two images is calculated to match the feature points, and the RANSAC algorithm randomly selects sample points to estimate the transformation model, and the accuracy of the transformation model is verified by the remaining points, and the wrong matching points are screened out to generate a visual feature matching set;

[0021] Based on the feature matching set, through triangulation and in combination with the positions of known reference objects, the corresponding positions of each robot in three-dimensional space are inferred from the image feature points from multiple perspectives to generate a robot spatial position estimation set.

[0022] As a further solution of the present invention, the information sharing between clusters through the wireless communication network includes spectrum scanning, channel quality perception and channel allocation, wherein spectrum scanning is used to perceive the usage of wireless spectrum, and the signal strength of a specific frequency band is determined by the signal peak in the spectrum through fast Fourier transform. Channel quality perception is used to monitor channel quality and judge the load level of each channel through the signal-to-noise ratio. Channel allocation is used to evenly distribute channel load, and low-load channels are identified and marked through load balancing algorithm analysis to avoid high-load time periods and frequency bands.

[0023] As a further solution of the present invention, the specific steps of the distributed decision-making are:

[0024] Based on the wireless communication network, the position information, speed and other sensor data between robots are shared, and the distributed consensus algorithm is used to coordinate positioning and tasks through information transmission to generate decentralized reports;

[0025] Based on the decentralized report, the local computing unit of each robot in the cluster aggregates and analyzes the received data, performs local decisions, and coordinates the overall decision to generate an edge computing report;

[0026] Based on the edge computing report, the common goal of the cluster robots is ensured through the maximum consensus algorithm, and data information is exchanged according to the set period, the timeliness of the data known by each robot is confirmed, and a distributed decision report is generated.

[0027] As a further solution of the present invention, the specific steps of the nonlinear optimization are:

[0028] Based on the data information corresponding to each robot within the cluster, multi-sensor data fusion is performed through a particle filter algorithm to correct the positioning error and generate a data fusion report;

[0029] Based on the data fusion report, according to the relative positioning information between each robot within the cluster, a constraint graph is constructed through a graph optimization method, and the constraint relationship between the positions of each robot is mapped as an edge in the graph to generate a position constraint report;

[0030] Based on the position constraint report, a non-linear optimization problem is constructed through the Gauss-Newton method, the state variables of each robot are represented as optimization variables, and by minimizing the error function in the constraint graph, the positions of each robot are calculated to generate a cluster robot position optimization report.

[0031] As a further solution of the present invention, the specific steps of the simultaneous localization are as follows:

[0032] Based on a clock synchronization algorithm, the robots within the cluster are initialized to obtain a consistent clock reference, and the received data information is marked with timestamps to generate a reference time set;

[0033] Based on the reference time set, through an adaptive time synchronization algorithm, according to the time step lengths required for different tasks, the synchronization frequency and step lengths of the robots are adjusted in real time to generate a step length adjustment set;

[0034] Based on the step length adjustment set, the positions of the robots are dynamically calibrated through a dynamic calibration algorithm, and global consistency verification is performed through a global optimization algorithm to generate a simultaneous localization report.

[0035] As a further solution of the present invention, the specific generation steps of the cluster collaborative movement report are as follows:

[0036] Based on the overall cluster goal and task, the cluster robots are coordinated through a game theory algorithm, and combined with a global path planning algorithm and a local path planning algorithm, the traveling directions of each robot are designed, and the robot speeds are dynamically adjusted by a differential drive model to generate a robot movement report;

[0037] Based on the robot movement report, combined with sensor data, environmental information and preliminary positioning, the bounding boxes of the robots are set one by one, and the positions of the robots are adjusted within the preset bounding boxes through a space constraint optimization algorithm and dynamically updated as the robots move to generate a bounding box adjustment report;

[0038] Based on the bounding box adjustment report, in cooperation with the sensor scan results, obstacles in the dynamic environment are identified through a dynamic obstacle avoidance algorithm, and the planned path is adjusted to generate a cluster collaborative movement report.

[0039] As a further solution of the present invention, the specific steps for generating the positioning feedback report are as follows:

[0040] Based on the data collected by the sensors, perform real-time positioning and coordinate data integration, and statistically analyze the coordinate information in chronological order to generate a coordinate data change report;

[0041] Based on the trained model, use the deep learning model to combine the robot motion data to infer the moving distance and direction, and generate a coordinate calculation report;

[0042] Based on the coordinate data change report and the coordinate calculation report, compare the coordinate information, analyze the error range through the root mean square error, and perform upload correction to generate a positioning feedback report.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] 1. In the present invention, through decentralization and edge computing, the computing pressure can be dispersed to each robot in the cluster, so as to avoid the bottleneck of the computing resources of the central server caused by the large scale of data to be processed in a complex environment. Furthermore, the network latency can be reduced, and the cluster robots can more flexibly share tasks, optimizing the overall execution efficiency and enhancing the working ability of the cluster robots.

[0045] 2. In the present invention, by screening the wireless network channels, the channels with high noise or serious interference can be identified and marked, so as to avoid continuing to use such channels and affecting the speed and reliability of data transmission. At the same time, the packet loss rate can be reduced. In addition, by optimizing the traffic allocation and balancing the load of each channel, it can be avoided that a certain channel resource in the network is over-occupied, effectively improving the data sharing speed among the robots in the cluster, and further enhancing the positioning accuracy of the robots.

[0046] 3. In the present invention, through the game theory algorithm, path planning and bounding boxes, the robots can be effectively constrained in a dynamic environment, avoiding collisions or conflicting operations between the robots, ensuring that the cluster robots can flexibly cope with the complex and changeable dynamic environment, and improving the completion efficiency of the overall task. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only the preferred embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 It is a schematic diagram of the overall structural flow of a collaborative positioning method for intelligent cluster robots proposed in an embodiment of the present invention.

[0049] Figure 2 It is a schematic diagram of the distributed decision-making step of a collaborative positioning method for intelligent swarm robots proposed in an embodiment of the present invention.

[0050] Figure 3 It is a schematic diagram of the swarm collaborative movement reporting step of a collaborative positioning method for intelligent swarm robots proposed in an embodiment of the present invention. Detailed implementation manners

[0051] The principles and features of the present invention will be described below with reference to the accompanying drawings. The listed embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0052] Refer to Figures 1 - 3 , the present invention provides a collaborative positioning method for intelligent swarm robots, and the method includes the following steps:

[0053] Step 1: Based on the collaborative positioning requirements of the swarm robots, the environmental data and their own state data are collected in real time through corresponding sensors. According to various types of environmental data and robot state data required for positioning, the corresponding sensors installed on the surfaces of each robot are used to collect the required information, so as to provide a data basis for subsequent positioning analysis, and preprocess various types of data. By preprocessing the collected various types of data, the data quality can be improved, and then it is convenient for subsequent algorithms or models to directly use, avoiding positioning deviations caused by low data quality, and generating a basic data collection report;

[0054] Step 2: Based on the basic data collection report, individual positioning is performed one by one through GPS to initially determine the current position. In an open or well-signaled area, the GPS receiving module inside each robot can receive signals from multiple satellites, and calculate the distances between the robot and each satellite according to the time difference of the signals. Then, the longitude, latitude coordinates and altitude data of the current position can be initially determined through the triangulation method, and the relative position information of each robot can be deduced by matching the image features of the scene in combination with vision technology, generating an individual positioning report. In an indoor area or an area with poor GPS signals, image recognition algorithms and feature matching algorithms can be used to extract features from the image data obtained by the camera, such as features of buildings, road signs, etc., to assist the robot in understanding the relative position and the surrounding environment, and then it can be matched with the collected scene images to achieve the purpose of determining the position and orientation of the robot. Moreover, through the individual positioning report, the information such as the position and orientation of the robot inferred by GPS and vision technology can be transmitted to the subsequent steps in the form of a map or coordinates;

[0055] Step 3: Based on the monomer positioning report, conduct inter-cluster information sharing through the wireless communication network. When various types of data are transmitted through wireless communication networks such as Wi-Fi, Bluetooth, or 5G, the low-load channels identified and screened in real time can improve the data transmission speed and reliability. Furthermore, by evenly dispersing the data across different channels, it is possible to prevent some channels from being overloaded due to a large amount of data to be shared in complex environments. Exchange the data information obtained by each robot in the cluster. By transmitting the images, lidar data, GPS positioning data, etc. known by each robot to other nearby robots, the perception ability of each robot in the cluster towards the overall scenario can be enhanced. In combination with edge computing technology and decentralized collaborative control strategies, conduct distributed decision-making to generate a cluster information sharing report. Through the task scheduling algorithm, allocate data storage, processing, and decision-making tasks to the local computing units of each robot in the cluster to prevent the computing resources of a certain local computing unit from being overloaded. Moreover, in combination with the reinforcement learning algorithm or distributed decision-making algorithm, after integrating the collected information, each local computing unit can, while executing the assigned tasks based on local information, maintain a certain degree of cooperation with the local computing units of other robots and coordinate through the consensus algorithm to ensure that subsequent global optimization can be carried out;

[0056] Step 4: Based on the cluster information sharing report, conduct relative positioning and error correction. Use the multi-sensor data fusion algorithm to integrate the positioning data of different robots. The particle filter algorithm can estimate the states of multiple robots through distributed samples to fuse the data information collected by multiple sensors. Furthermore, by combining the relative distances, angles, and other sensor data between robots, the relative positions between the current robot and other robots or a group of robots can be estimated. In combination with the Kalman filter algorithm, the errors in relative positioning originating from sensors, motion models, or communication delays, etc. can be corrected to a certain extent to further improve the positioning accuracy after multi-sensor data fusion. Construct a constraint graph for cluster robots and conduct non-linear optimization. At the same time, in combination with the synchronization algorithm, perform synchronous positioning of cluster robots at different time steps to generate a collaborative positioning report. The graphical model of the constraint graph can be established through data information such as the relative positions, time synchronization, and sensor observations between robots. In multi-robot collaborative tasks, each node in the graphical model can represent the position of a certain robot. Furthermore, in combination with the edges representing the constraint relationships between robots in the graphical model, the data between different robots can be integrated and optimized. In combination with the graph optimization algorithm, the optimal position configuration information of the robots can be obtained by minimizing the errors of all edges in the constraint graph. In addition, since each robot in the cluster usually collects data at different time steps, the time synchronization algorithm can be used to perform synchronous positioning of different robots to ensure that the robots can share consistent position information at different time steps and reduce the errors caused by time asynchronization;

[0057] Step 5: Based on the collaborative localization report, collaborate according to the overall cluster goals and tasks, assign independent traveling directions, adjust the moving speeds of each robot within the cluster. From the assigned tasks and goals, calculate the current path through the path planning algorithm in combination with the global goals of each robot, the positions of other robots, and obstacle information, and dynamically adjust the speed according to the distance of neighboring robots through the velocity-obstacle model to avoid collisions between robots, optimize the overall execution efficiency, perform local optimization on the position through the set bounding box, combine the path planning and dynamic obstacle avoidance algorithms to generate a cluster collaborative movement report, delimit a safety area for each robot through the bounding box, ensure that the robot is always within the feasible area during movement, and reduce the probability of collisions through real-time adjustment of the local path. Among them, the optimal movement window can be calculated through the dynamic window method in combination with the speed, acceleration, and obstacle position of the robot, and then cooperate with the path planning algorithm to ensure that the robot can complete the task safely and quickly in a changing environment to achieve the purpose of local optimization;

[0058] Step 6: Based on the cluster collaborative movement report, perform real-time feedback, optimize the localization strategy in a dynamic environment through machine learning, generate a localization feedback report, respectively infer and calculate the coordinates of the cluster robots after movement according to the localization data, movement direction, and distance feedback by the sensors, and then perform coordinate comparison to judge the existing localization error. Further, interact with the environment through machine learning so that the robot can optimize its own behavior according to the environmental feedback.

[0059] Please refer to Figure 1 , the environmental data and its own state data include lidar data, visual sensor data, GPS data, and IMU data. Among them, the lidar data is used to reflect the distance, shape, and position information of surrounding objects. The lidar can scan the surrounding environment with laser beams and measure the reflected echoes, and then the spatial information can be obtained through the time difference to achieve the purposes of obstacle detection, map construction, and positioning in cooperation with GPS. The visual sensor data is used for map construction and object recognition. The camera and other visual sensors can capture image data in the scene, thereby providing the robot with morphological information about surrounding objects, assisting in identifying object types, and cooperating in constructing a three-dimensional space. The GPS data is used to initially determine the positioning information of the robot. The actual coordinate position of the robot can be jointly inferred by combining the GPS data with other positioning data. The IMU data is used to sense the angular velocity, acceleration, and attitude changes of the robot. Since the IMU includes sensors such as accelerometers, gyroscopes, and magnetometers, the IMU data can be used to understand the dynamic information of the robot in a scene with weak GPS signals, and then can also be used for accurate attitude and motion inference after being fused with other data.

[0060] Please refer to Figure 1, the specific steps for preprocessing various types of data are:

[0061] Based on the various types of data collected, correlation analysis is performed to screen out features with high correlation. Since the collected data comes from different channels, the data may contain redundant data, outliers, and missing values, and mixed data will inevitably affect the accuracy of subsequent analysis results. Therefore, by calculating the correlation coefficient between each feature, features with high correlation with the target variable or other features can be screened out, thereby reducing feature redundancy and improving the efficiency and accuracy of subsequent algorithms or models. At the same time, the Z-Score algorithm is used to detect outliers, identify points that exceed the set threshold, eliminate redundant data and repeatedly collected data, and generate a data cleaning set. Since the Z-Score can determine whether the data is an outlier by measuring the degree of deviation between the data point and the data mean, the Z-Score algorithm can be used to discover and eliminate outliers that exceed the set standard, thereby further improving data quality and avoiding overfitting or deviation in subsequent analysis.

[0062] Based on the data cleaning set, the missing data is estimated and filled by interpolation. The interpolation method can be used to infer the missing data value based on the rules between known data, which can effectively reduce the impact of missing data on the overall analysis results and ensure the integrity of the data. The low-frequency noise is removed by high-pass filtering, and the high-frequency useful signal is retained to generate a data sorting set. The high-pass filtering technology can be used to remove the noise of the data, and by filtering the low-frequency noise in the data, more representative high-frequency signals can be retained, reducing the interference of irrelevant factors, so as to improve the accuracy of subsequent analysis;

[0063] Based on the data collection set, the data is standardized and normalized, all features are mapped to a unified range, and the fairness of each feature in the calculation is ensured by adjusting the data range to generate a data processing set. Standardization can eliminate the differences between different feature dimensions, making the data more suitable for statistical analysis and modeling, while normalization can make each feature within the same scale range. Therefore, standardization and normalization can ensure the uniformity of the data and the fairness of each feature, thereby providing a reliable basic condition for subsequent further data processing, machine learning model training, etc.

[0064] See also Figure 1 ,The specific steps of combining visual technology for image feature matching are:

[0065] Based on the preprocessed image data, the ORB algorithm is used to extract feature points from each image in the scene one by one, including corner points and edge points, to generate a visual feature extraction set. The ORB algorithm can extract corner points that are easily distinguishable and matchable in the image through the FAST corner detector, and then each point can be described by the BRIEF descriptor to form a corresponding feature description, preparing for subsequent matching;

[0066] Based on the feature extraction set, by calculating the similarity of feature points in two images, feature point matching is performed. At the same time, the RANSAC algorithm randomly selects sample points to estimate the transformation model, and verifies the accuracy of the transformation model through the remaining points, screening out incorrect matching points to generate a visual feature matching set. The RANSAC algorithm can randomly select a certain number of feature point pairs in two images, and then the transformation model such as the homography matrix or the essential matrix can be calculated through the feature point pairs, and the transformation error of all feature points can be calculated reversely through the model to judge the correctness of the matched feature points according to the error size, providing an accurate data basis for estimating the robot position based on feature points in the subsequent process;

[0067] Based on the feature matching set, through triangulation, combined with the positions of known references, the corresponding positions of each robot in the three-dimensional space are deduced from the image feature points of multiple perspectives to generate a robot spatial position speculation set. According to the coordinates of the feature points in the image, they can be converted into actual spatial coordinates, and combined with the known camera parameters, the positions of each robot in the three-dimensional space can be deduced through the geometric triangulation method to make up for the weak GPS signal.

[0068] Please refer to Figure 1, the inter-cluster information sharing via a wireless communication network includes spectrum scanning, channel quality perception, and channel allocation. Among them, spectrum scanning is used to sense the usage of the wireless spectrum. The signal strength of a specific frequency band is determined by the signal peak in the spectrum through fast Fourier transform. After obtaining the signal data for a period of time and performing fast Fourier transform on the signal, the corresponding frequency-domain data can be obtained. Furthermore, by analyzing the signal strength of each frequency band in the frequency-domain data, the signal peak and its corresponding frequency band can be determined. Further, based on the strength of the signal peak, the occupancy of each frequency band can be identified and marked to obtain free channels. Quality perception is used to monitor the channel quality. The load level of each channel is judged by the signal-to-noise ratio. The corresponding signal-to-noise ratio can be calculated based on the received signal power and noise power. Furthermore, the quality of each channel can be evaluated according to the size of the signal-to-noise ratio. At the same time, the Kalman filtering algorithm can be used to reduce noise interference to improve the accuracy of perception. Channel allocation is used to evenly distribute the channel load. The low-load channels are analyzed and identified through the load balancing algorithm and marked, avoiding high-load time periods and frequency bands. Through the neural network, the load and quality of each channel can be monitored in real time, and the channels with lower load can be marked according to indicators such as the SNR and throughput of the channels. At the same time, according to the real-time load situation, dynamic allocation can be performed through the load balancing algorithm, avoiding allocating new tasks to channels that are already close to full load, so as to achieve the purpose of improving the channel utilization rate, ensuring the transmission speed and reliability, and preventing the positioning between cluster robots from being affected due to packet loss.

[0069] Please refer to Figure 2 , the specific steps of distributed decision-making are as follows:

[0070] Based on the wireless communication network, share the position information, speed, and other sensor data among robots. Through the wireless communication network, the data collected by the sensors equipped with the current robot can be uploaded to the local network in real time and exchanged with other robots for various data such as position, speed, acceleration, attitude angle, and obstacles. In cooperation with the distributed consensus algorithm, positioning and task coordination are carried out through information transmission to generate a decentralized report. A consistent decision-making basis is established among each robot through the distributed consensus algorithm to ensure that the robot can correctly understand and integrate the data information from other robots, and further ensure that each robot can share tasks more flexibly in the follow-up;

[0071] Based on decentralized reports, the data received is aggregated and analyzed by the local computing units of each robot in the cluster, local decisions are executed, and the local computing units coordinate on the overall decision to generate edge computing reports. The local computing units of each robot aggregate data from other robots and combine local sensor information for data processing and decision-making, such as judging whether to adjust the path, execute tasks, or avoid obstacles based on real-time position, speed, and environmental data. This can disperse the computing pressure and reduce network latency to a certain extent. At the same time, in cooperation with algorithms such as Dijkstra's algorithm and Bayesian decision theory, each local computing unit can comprehensively consider local decisions and the goals and tasks of other robots in the cluster to ensure the subsequent overall coordination and working ability of the cluster;

[0072] Based on the edge computing reports, the common goals of the cluster robots are ensured through the maximum consensus algorithm. Through the maximum consensus algorithm, the unified recognition of a certain goal by each robot in the cluster can be confirmed, and on this basis, appropriate adjustments to their respective behaviors can be made to prevent local decisions from interfering with the achievement of the final goal. Data information is exchanged according to the set period to confirm the timeliness of the data known by each robot, and a distributed decision report is generated. Through data information exchange, each robot in the cluster can regularly update its understanding of the environment and tasks, avoiding decision-making errors caused by some robots due to delays or packet losses, and improving the overall execution efficiency.

[0073] Please refer to Figure 1 , and the specific steps of non-linear optimization are as follows:

[0074] Based on the data information corresponding to each robot in the cluster, multi-sensor data fusion is performed through the particle filter algorithm to correct the positioning error and generate a data fusion report. Since there are various types of sensors on the cluster robots, such as lidar, cameras, and IMUs, and there are inevitably certain noises or errors in the data of each sensor, the probability distribution of the robot state can be represented through the particle filter algorithm, and then iterative correction can be performed through prediction, update, and resampling to obtain the state estimation values, weight distributions, and data consistency analysis of each sensor after data fusion;

[0075] Based on the data fusion report, according to the relative positioning information between each robot in the cluster, a constraint graph is constructed through the graph optimization method, and the constraint relationship between the positions of each robot is mapped as an edge in the graph to generate a position constraint report. Through graph optimization, the constraint relationship between each robot in the cluster can be represented by a graph model, enabling each robot in the cluster to correct its own position based on the relative position estimation of other robots;

[0076] Based on the position constraint report, a nonlinear optimization problem is constructed by the Gauss-Newton method. The state variables of each robot are represented as optimization variables. By minimizing the error function in the constraint graph, the positions of each robot are calculated, and a cluster robot position optimization report is generated. By the Gauss-Newton method, the error function in the position constraint graph can be minimized in the graph optimization problem. Furthermore, by adjusting the positions of the robots, all constraint relationships can be best fitted to achieve the position optimization of each robot within the cluster, thereby improving the collaborative efficiency and accuracy of the entire cluster of robots.

[0077] Please refer to Figure 1 , and the specific steps of synchronous positioning are as follows:

[0078] Based on the clock synchronization algorithm, the robots within the cluster are initialized to obtain a consistent clock reference. During initialization, each robot within the cluster can obtain the time information of other robots by sending time request packets and receiving response packets. Then, the clock deviation is corrected by comparing the received timestamps with its own time, and the clock deviation is further corrected by algorithms such as the Kalman filter algorithm or the average synchronization method to obtain a consistent clock reference. The data information received is marked with timestamps to generate a reference time set. By marking the data information received by each robot with timestamps, the data generation time of all robots within the cluster is obtained;

[0079] Based on the reference time set, through the adaptive time synchronization algorithm, according to the time step sizes required for different tasks, the synchronization frequency and step size of the robots are adjusted in real time to generate a step size adjustment set. The adaptive time synchronization algorithm collects the status information of each robot in real time according to the current task progress and synchronization error of each robot, and determines the synchronization frequency according to the required time step size and the computing power of each robot. Among them, for tasks requiring higher precision, the synchronization frequency is increased, and vice versa. At the same time, automatic adjustment can also be made according to the actual step size requirements at different stages of the task to ensure that the time synchronization accuracy of the robots can adapt to the task requirements;

[0080] Based on the step size adjustment set, the positions of the robots are dynamically calibrated by the dynamic calibration algorithm, and global consistency verification is performed by the global optimization algorithm to generate a synchronous positioning report. By the Kalman filter algorithm, each robot can calculate its own position, speed and other status information based on the current timestamp and sensor data, and perform calibration according to the synchronized clock information to accurately understand its relative position to itself and other robots during the task. At the same time, the least squares method can verify whether the position relationships and time information among the robots are consistent.

[0081] Please refer to Figure 3 , and the specific steps for generating the cluster collaborative movement report are as follows:

[0082] Based on the overall goals and tasks of the cluster, the swarm robots are coordinated through game theory algorithms. The game theory algorithms can design corresponding strategies for each robot, and determine the action plan through the game decisions among the robots, so as to maximize the overall benefit when executing tasks, avoid resource conflicts or task conflicts. Combining the global path planning algorithm and the local path planning algorithm, the traveling direction of each robot is designed. The optimal path from the starting point to the target point can be determined through the Dijkstra algorithm. Furthermore, in cooperation with the dynamic window method, the robot can handle the avoidance of obstacles and local adjustments during the task execution process, so as to dynamically complete the path planning. The differential drive model dynamically adjusts the robot's speed and generates a robot motion report. The differential drive model can assist the robot in dynamically adjusting the speed and steering angle according to the path planning results to achieve smooth movement;

[0083] Based on the robot motion report, combined with sensor data, environmental information and preliminary positioning, the bounding box of the robot is set one by one. The position of the robot is adjusted within the preset bounding box through the spatial constraint optimization algorithm and dynamically updated as the robot moves, generating a bounding box adjustment report. According to the current position, speed, size and other information of the robot, the motion range of the robot is calculated, and the bounding box is set for planning and obstacle avoidance. At the same time, the bounding box can be adjusted through genetic algorithms or simulated annealing algorithms, etc., to meet the minimum distance requirement between robots and prevent the robot from entering the safety area of other robots and causing collisions;

[0084] Based on the bounding box adjustment report, in cooperation with the sensor scanning results, obstacles in the dynamic environment are identified through the dynamic obstacle avoidance algorithm, and the planned path is adjusted, generating a swarm collaborative movement report. The repulsive force is calculated in real time through the artificial potential field method to guide the robot to avoid obstacles in the dynamic environment, ensuring that the robot always stays within the safe range, and further achieving the purpose of flexibly adjusting the future movement path to ensure that the swarm robots can flexibly respond to the complex and changeable dynamic environment.

[0085] Please refer to Figure 1 , and the specific generation steps of the positioning feedback report are as follows:

[0086] Based on the data collected by the sensor, real-time positioning and coordinate data integration are carried out, and the coordinate information is statistically analyzed in chronological order, generating a coordinate data change report. Through various positioning data continuously collected and the corresponding integration speculation results, each time point can be matched with the corresponding coordinates;

[0087] Based on the trained model, the moving distance and direction are speculated through the deep learning model combined with the robot motion data, generating a coordinate calculation report. The deep learning model can use the sensor data as input to train a model that can speculate and calculate the moving distance and direction of the robot, and then the coordinate information, displacement and direction change data of each robot at each moment can be obtained;

[0088] Based on the coordinate data change report and the coordinate calculation report, compare the coordinate information, analyze the error range through the root mean square error, and perform upload correction to generate a positioning feedback report. By comparing the coordinates integrated by the sensor and the coordinates calculated by the deep learning model, possible sensor errors, environmental interferences, etc. can be judged by error identification, and then the accuracy of positioning can be assisted in determination, and corresponding positioning optimization can be carried out through feedback.

[0089] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A collaborative positioning method for intelligent cluster robots, characterized in that: The method comprises the following steps: Step 1: Based on the collaborative positioning requirements of the swarm robots, the corresponding sensors are used to collect environmental data and their own status data in real time, and various types of data are pre-processed to generate a basic data collection report; Step 2: Based on the basic data collection report, the individual units are positioned one by one through GPS to preliminarily determine the current position, and the image features of the scene are matched with visual technology to calculate the relative position information of each robot and generate an individual positioning report; Step 3: Based on the single-unit positioning report, information is shared among clusters through a wireless communication network, data information obtained by each robot in the cluster is exchanged, and distributed decision-making is performed in conjunction with edge computing technology and decentralized collaborative control strategies to generate a cluster information sharing report; Step 4: Based on the cluster information sharing report, relative positioning and error correction are performed, the positioning data of different robots are integrated using a multi-sensor data fusion algorithm, a cluster robot constraint graph is constructed, nonlinear optimization is performed, and the cluster robots are synchronously positioned at different time steps in conjunction with a synchronization algorithm to generate a collaborative positioning report; Step 5: Based on the collaborative positioning report, collaborate according to the overall goals and tasks of the cluster, assign independent travel directions, adjust the movement speed of each robot in the cluster, locally optimize the position through the set bounding box, combine path planning and dynamic obstacle avoidance algorithm, and generate a cluster collaborative movement report; Step 6: Based on the cluster collaborative mobility report, provide real-time feedback, optimize the positioning strategy in a dynamic environment through machine learning, and generate a positioning feedback report.

2. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The environmental data and self-state data include lidar data, visual sensor data, GPS data and IMU data, wherein the lidar data is used to reflect the distance, shape and position information of surrounding objects, the visual sensor data is used to build maps and object recognition, the GPS data is used to preliminarily determine the robot's positioning information, and the IMU data is used to perceive the robot's angular velocity, acceleration and posture changes.

3. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The specific steps of preprocessing various types of data are as follows: Based on the various types of collected data, correlation analysis is performed to screen out features with high correlation. At the same time, outlier detection is performed through the Z-Score algorithm to identify points that exceed the set threshold, eliminate redundant data and repeatedly collected data, and generate a data cleaning set; Based on the data cleaning set, the missing data is estimated and filled by interpolation, and low-frequency noise is removed by high-pass filtering, and high-frequency useful signals are retained to generate a data sorting set; Based on the data collation set, the data is standardized and normalized, all features are mapped into a unified range, and the fairness of each feature in the calculation is ensured by adjusting the data range to generate a data processing set.

4. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The specific steps of combining visual technology to perform image feature matching are: Based on the preprocessed image data, feature points of each image in the scene are extracted one by one using the ORB algorithm, including corner points and edge points, to generate a visual feature extraction set; Based on the feature extraction set, the feature points are matched by calculating the similarity of the feature points in the two images, and the transformation model is estimated by randomly selecting sample points by the RANSAC algorithm, and the accuracy of the transformation model is verified by the remaining points, and the wrong matching points are screened out to generate a visual feature matching set; Based on the feature matching set, through triangulation and in combination with the positions of known reference objects, the corresponding positions of each robot in three-dimensional space are inferred from the image feature points from multiple perspectives to generate a robot spatial position estimation set.

5. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The information sharing between clusters through the wireless communication network includes spectrum scanning, channel quality perception and channel allocation, wherein spectrum scanning is used to perceive the use of wireless spectrum, and the signal strength of a specific frequency band is determined by the signal peak in the spectrum through fast Fourier transform. Channel quality perception is used to monitor channel quality and judge the load level of each channel through the signal-to-noise ratio. Channel allocation is used to evenly distribute channel loads, and low-load channels are identified and marked through load balancing algorithm analysis to avoid high-load time periods and frequency bands.

6. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The specific steps of the distributed decision-making are: Based on the wireless communication network, the position information, speed and other sensor data between robots are shared, and the distributed consensus algorithm is used to coordinate positioning and tasks through information transmission to generate decentralized reports; Based on the decentralized report, the local computing unit of each robot in the cluster aggregates and analyzes the received data, performs local decisions, and coordinates on the overall decision to generate an edge computing report; Based on the edge computing report, the common goal of the cluster robots is ensured through the maximum consensus algorithm, and data information is exchanged according to the set period, the timeliness of the data known by each robot is confirmed, and a distributed decision report is generated.

7. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The specific steps of the nonlinear optimization are: Based on the data information corresponding to each robot in the cluster, the particle filter algorithm is used to perform multi-sensor data fusion, correct the positioning error, and generate a data fusion report; Based on the data fusion report, according to the relative positioning information between the robots in the cluster, a constraint graph is constructed by a graph optimization method, and the constraint relationship between the positions of the robots is mapped to the edges in the graph to generate a position constraint report; Based on the position constraint report, a nonlinear optimization problem is constructed using the Gauss-Newton method, the state variables of each robot are represented as optimization variables, and the position of each robot is calculated by minimizing the error function in the constraint graph to generate a cluster robot position optimization report.

8. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The specific steps of the synchronous positioning are: Based on the clock synchronization algorithm, the robots in the cluster are initialized to obtain a consistent clock reference, and the received data information is marked with a timestamp to generate a reference time set; Based on the reference time set, an adaptive time synchronization algorithm is used to adjust the frequency and step length of robot synchronization in real time according to the time steps required by different tasks, thereby generating a step length adjustment set; Based on the step size adjustment set, the robot position is dynamically calibrated through a dynamic calibration algorithm, and global consistency verification is performed through a global optimization algorithm to generate a synchronous positioning report.

9. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The specific steps of generating the cluster coordinated mobility report are as follows: Based on the overall goals and tasks of the cluster, the cluster robots are coordinated through the game theory algorithm, and the travel direction of each robot is designed by combining the global path planning algorithm and the local path planning algorithm. The robot speed is dynamically adjusted by the differential drive model, and the robot motion report is generated; Based on the robot motion report, combined with sensor data, environmental information and preliminary positioning, the robot's bounding box is set one by one, and the robot position is adjusted within the preset bounding box through a spatial constraint optimization algorithm, and dynamically updated as the robot moves, to generate a bounding box adjustment report; Based on the bounding box adjustment report and in conjunction with the sensor scanning results, the dynamic obstacle avoidance algorithm is used to identify obstacles in the dynamic environment, adjust the planned path, and generate a cluster collaborative movement report.

10. The collaborative positioning method for intelligent cluster robots according to claim 1, characterized in that: The specific steps for generating the positioning feedback report are as follows: Based on the data collected by the sensor, real-time positioning and coordinate data integration are performed, and the coordinate information is counted in chronological order to generate a coordinate data change report; Based on the trained model, the deep learning model is combined with the robot motion data to infer the moving distance and direction, and a coordinate calculation report is generated; Based on the coordinate data change report and the coordinate calculation report, the coordinate information is compared, the error range is analyzed through the root mean square error, and the correction is uploaded to generate a positioning feedback report.

Images