A robot pose data processing method based on a distributed collaborative network

Abstract

The application discloses a kind of robot posture data processing methods based on distributed cooperative network, and specifically relates to data processing technical field;By constructing dynamic coupling matrix, the posture data of robot is coupled and time synchronization processing using nonlinear adjacent weight function, the dynamic posture coordination index of each node is extracted, the posture abnormal node is identified and data correction is carried out;By breaking the chain processing to key propagation node, posture repair is carried out using local differential reconstruction model, finally generates posture correction set, uses recursive self-calculus model to the posture data after correction Short-term evolution prediction and reverse check are carried out, generate consistent posture fusion field, as the input of unified action planning and control of multi-robot system;The application can improve the cooperativity and stability of system by fusing the posture data between multiple robots, and realize high-precision cooperative task execution in complex environment.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to a method for processing robot posture data based on a distributed collaborative network. Background Technology

[0002] With the widespread deployment of multi-agent robot systems in complex environments, particularly in disaster relief, deep-sea exploration, and complex industrial control scenarios, higher demands are being placed on the real-time performance and consistency of attitude information among robots. Currently, attitude data is primarily acquired independently by the inertial measurement units (IMUs) of each robot and processed centrally to generate system-level collaborative decisions. However, in practical deployments, factors such as network node instability, channel latency, and local sensor drift in distributed environments can cause inconsistencies, loss of synchronization, or distortion in attitude data. In severe cases, this can lead to disordered collaborative behavior and mission failure.

[0003] Especially in ad hoc networks with low bandwidth or high packet loss rates (such as non-line-of-sight scenarios like mountains, underground utility tunnels, and the seabed), traditional centralized attitude processing methods are difficult to operate effectively due to the lack of stable central processing nodes. Furthermore, while some existing distributed methods can mitigate synchronization failures to some extent, they cannot address the fundamental flaws of abnormal attitude drift and the amplification of local errors, easily leading to high-frequency misleading feedback in collaborative control, severely impacting task stability and system security. Summary of the Invention

[0004] The purpose of this invention is to provide a robot posture data processing method based on a distributed cooperative network to address the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a robot posture data processing method based on a distributed cooperative network, comprising:

[0006] S100. Obtain the raw posture data Pi of each collaborative robot and its corresponding timestamp Ti, and construct a time-calibrated posture data set. ;

[0007] S200. Based on the attitude data set P and combined with the collaborative communication topology graph between robots, extract the attitude difference ΔPij and time offset ΔTij between any two robots, and use a nonlinear adjacency weighting function. Construct the dynamic coupling matrix M(t);

[0008] S300. Use M(t) to extract the dynamic posture coordination index value of each robot node, and combine it with the instantaneous connectivity of the node in the topology graph to calculate the node stability factor, identify the posture anomaly propagation chain and the key propagation node set Vcrit.

[0009] S400. The original pose data of the nodes in Vcrit is broken, and local differential reconstruction estimation is performed from the highly cooperative neighborhood of non-abnormal nodes to obtain the pose repair estimate, which constitutes the pose correction set.

[0010] The S500 uses a recursive self-calculation model to perform evolution prediction and reverse verification of the posture correction set in the time domain, and outputs a posture fusion consistency field to drive the unified motion planning of the collaborative robot.

[0011] Preferably, S100 includes:

[0012] S101. Attitude data is collected by the inertial measurement unit built into each collaborative robot. The attitude data includes three-axis acceleration, three-axis angular velocity and three-axis magnetic field strength.

[0013] S102. Denoise the original attitude data based on the Kalman filter algorithm;

[0014] S103. Perform unified calibration on the timestamps collected from multiple collaborative robots to generate a globally aligned timestamp set;

[0015] S104. Map the denoised attitude data one-to-one with the corresponding global alignment timestamps to construct a time-calibrated attitude data set P.

[0016] Preferably, wherein S200 includes:

[0017] S201. Based on the time-calibrated attitude data in the attitude data set P, calculate the attitude difference ΔPij between any two collaborative robots at the same time point, where the attitude difference is a nine-dimensional Euclidean distance.

[0018] S202. Calculate the time synchronization offset ΔTij between the two corresponding robots, where the time synchronization offset is the absolute difference between the timestamps corresponding to their respective attitude data.

[0019] S203. Construct a nonlinear adjacency weight function f(ΔPij,ΔTij) based on the attitude difference ΔPij and the time offset ΔTij;

[0020] S204. The calculation results of the nonlinear adjacency weight function are combined according to the connection relationship between each robot in the cooperative communication topology to construct a time-varying dynamic coupling matrix M(t).

[0021] Preferably, the dynamic posture coordination index value of each robot node is extracted using M(t), including:

[0022] S301. Extract the set of coupling weights Wi={f(ΔPij,ΔTij)} between each robot node i and its communicating neighbor node j at the current time based on the dynamic coupling matrix M(t);

[0023] S302. Calculate the coupling mean μi and standard deviation σi of node i to measure the degree of cooperative fluctuation between it and its neighboring nodes;

[0024] S303, Construct dynamic attitude coordination index values ;

[0025] S304. Based on the dynamic attitude coordination index value Ci, when the coordination index value Ci of node i is lower than the set dynamic threshold θc, node i is determined to be a coordination abnormal node.

[0026] Preferably, the node stability factor is calculated by combining the instantaneous connectivity of nodes in the topology graph, and the attitude anomaly propagation chain and the key propagation node set Vcrit are identified, including:

[0027] S305. Based on the dynamic coupling matrix M(t) and the cooperative communication topology graph, calculate the instantaneous connectivity Di of each robot node at the current moment, where the connectivity is the sum of the coupling weights of node i and all its neighboring nodes.

[0028] S306. Perform joint normalization on the attitude coordination index value Ci and the instantaneous connectivity Di to construct a node stability factor. ;

[0029] S307. Construct a propagation path tracing graph for nodes whose stability factor Si is lower than the dynamic threshold θs, and mark their shortest path transmission chain in the topology graph to identify potential attitude anomaly propagation chains.

[0030] S308. Extract the key propagation node set Vcrit based on the node with the maximum path distribution degree or the lowest stability factor in the propagation path.

[0031] Preferably, the S400 includes:

[0032] S401. For each node in the critical propagation node set Vcrit, remove its original attitude data during the abnormal propagation period and mark the data of that period as null.

[0033] S402. Identify a set of highly cooperative neighbor nodes in the cooperative communication topology graph. The neighbor nodes must satisfy the following conditions: the cooperative index value is greater than the threshold θc, and the stability factor is higher than the threshold θs.

[0034] S403. In a highly collaborative neighborhood, a differential interpolation model with a time window of τ is constructed based on the rate of change of attitude difference between nodes, and the attitude estimate of the target node is reconstructed by weighted averaging.

[0035] S404. Fill in the reconstructed attitude estimates with the original abnormal data segments one by one to generate an attitude repair estimation sequence, and summarize them to form an attitude correction set.

[0036] Preferably, the S500 includes:

[0037] S501. Construct a set of attitude input subsequences based on a time series window, with each subsequence having a length of τ, and extract the attitude vector sequence from consecutive time segments in the attitude correction set.

[0038] S502. Perform short-term forward prediction on each subsequence to generate a predicted attitude sequence, representing the attitude evolution trend within the next τ steps.

[0039] S503. Compare the predicted sequence with the actual observations in the attitude correction set in reverse, and construct a consistency confidence score by calculating the mean of the predicted residuals within the time window.

[0040] S504. Perform weighted fusion processing on the attitude sequences of nodes with confidence scores higher than the threshold θf to generate a temporally continuous and spatially consistent attitude fusion consistent field.

[0041] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0042] 1. This invention successfully achieves efficient collaboration among multiple robot systems in dynamic environments by constructing a posture fusion consistency field. Through a recursive self-calculation model, short-term prediction and back-end verification of posture data are performed, combined with a weighted fusion mechanism, ensuring the consistency and accuracy of posture data among multiple robots. This consistency field provides a stable and accurate global posture reference for subsequent unified motion planning, significantly improving the task execution efficiency of collaborative robots in complex environments.

[0043] 2. This invention significantly enhances the adaptability of robot systems to environmental changes and ensures high coordination among robots by using posture prediction and back-verification based on a recursive model. Secondly, by utilizing a posture fusion consistent field for motion planning, it avoids the reliance on single-robot control found in traditional methods, enabling multi-robot systems to respond more flexibly and efficiently to dynamic tasks and environmental changes. Furthermore, the system's closed-loop feedback mechanism ensures that robots can adjust their actions in real time during task execution, improving system stability. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] For examples, please refer to Figure 1 As shown in this embodiment, a robot posture data processing method based on a distributed cooperative network includes:

[0048] S100. Obtain the raw posture data Pi of each collaborative robot and its corresponding timestamp Ti, and construct a time-calibrated posture data set. .

[0049] In this embodiment, attitude data is acquired through an inertial measurement unit (IMU) installed on the collaborative robot body. This IMU can output the robot's dynamic information in three-dimensional space in real time. Specifically, this includes:

[0050] Three-axis acceleration data is used to reflect the linear acceleration changes of the robot in the X, Y, and Z directions;

[0051] Three-axis angular velocity data are used to reflect the robot's angular velocity rotation information around the X, Y, and Z axes;

[0052] Three-axis magnetic field strength data are used to assist in correcting attitude deviations and determining absolute orientation.

[0053] The acquisition frequency was set to 100 Hz to ensure that the attitude data had sufficient time resolution to support subsequent high-precision processing. Each data point contains a 9-dimensional vector, denoted as the original body attitude data Pi=[ax,ay,az,ωx,ωy,ωz,mx,my,mz], where a represents acceleration, ω represents angular velocity, and m represents magnetic field strength.

[0054] To reduce high-frequency noise and occasional abnormal fluctuations in the raw sensor data, this implementation uses the Kalman filter algorithm to preprocess the attitude data for each dimension.

[0055] The Kalman filter algorithm is based on a discrete-time state-space model. It dynamically adjusts the estimated covariance matrix by iteratively updating the error between the prior state estimate and the observed data, thereby filtering out noise components. Specifically, it includes:

[0056] Initialize the state estimates and the error covariance matrix;

[0057] Calculate the residual between the predicted value and the observed value based on the sensor observations input at the current moment;

[0058] The state estimate is updated based on the minimum mean square error criterion.

[0059] Input the updated original body pose data Pi into the next processing step.

[0060] This algorithm removes interference components with frequencies exceeding 20 Hz and smoothly replaces outliers with instantaneous changes greater than a set threshold (such as 2g or 500 degrees / second), effectively improving the stability and continuity of attitude data.

[0061] Because multiple collaborative robots collect data independently, their internal clocks may have slight deviations, resulting in inconsistencies in the timestamps of the attitude data at the nanosecond or millisecond level. To ensure the accuracy of subsequent data synchronization processing, global alignment of the timestamps is required. The specific implementation method is as follows:

[0062] Each robot records a local timestamp Ti, in milliseconds, when collecting each piece of posture data;

[0063] By broadcasting synchronization signal frames, a reference robot is selected within the communication range, and its local clock is denoted as Tref. The offset of the synchronization signal time received by other robots is calculated. This ultimately forms a globally aligned timestamp set. n is the total number of timestamps, used to ensure that all attitude data are processed on the same time reference.

[0064] After denoising and time alignment, each pose data point is matched one-to-one with its corresponding global alignment timestamp to form a complete time-calibrated pose data set. This set is defined as follows: , where Pi is the original pose data of the body and Ti is the unified timestamp after time synchronization processing.

[0065] S200. Based on the attitude data set P and combined with the collaborative communication topology graph between robots, extract the attitude difference ΔPij and time offset ΔTij between any two robots, and use a nonlinear adjacency weighting function. Construct the dynamic coupling matrix M(t).

[0066] In this embodiment, in order to quantify the differences in posture states between collaborative robots, it is necessary to calculate the degree of difference in posture data of any two robots at the same point in time.

[0067] Suppose we extract the attitude data of robots i and j at a certain alignment time T from the attitude data set P, denoted as Pi(T) and Pj(T) respectively. Each attitude data is a 9-dimensional vector containing three-axis acceleration, three-axis angular velocity, and three-axis magnetic field strength. The attitude difference ΔPij is calculated using the Euclidean distance function, specifically by summing the squared differences of the three-axis acceleration, three-axis angular velocity, and three-axis magnetic field strength in each dimension, and then taking the square root. The attitude difference ΔPij is used to measure the overall difference in attitude between the two robot nodes.

[0068] To further characterize the collaborative offset caused by time synchronization errors between collaborative robots, this embodiment introduces a time synchronization offset parameter ΔTij. ΔTij is defined as the absolute difference between the timestamps of robot i and robot j when they are sampled at the same time, and is calculated as follows: , where Ti and Tj are the timestamps of the corresponding posture data of the two robots in set P, in milliseconds.

[0069] After obtaining the posture difference ΔPij and time offset ΔTij, a nonlinear adjacency weight function f(ΔPij, ΔTij) is constructed to characterize the cooperative tightness between the two robots. The design goal of this function is that the smaller the posture difference between the two robots and the more accurate the time synchronization, the stronger their cooperative coupling. This implementation uses a function of the following form for modeling: α and β are adjustment parameters used to control the influence weights of attitude difference and time offset on coupling strength. Preferably, α is 0.05 and β is 0.001.

[0070] After calculating the nonlinear adjacency weight values ​​between each pair of collaborative robots, the results of all weight functions are summarized and combined according to the connection relationships defined in the robot collaborative communication topology graph to construct the time-varying dynamic coupling matrix M(t).

[0071] The collaborative communication topology is represented as a graph structure G(V,E), where V is the set of robot nodes and E is the set of edges. For any robot pair (i,j) with a direct communication connection, the corresponding weight function value f(ΔPij,ΔTij) is used as the element in the i-th row and j-th column of M(t). If there is no direct communication connection between robots i and j, the corresponding element in M(t) is assigned a value of 0.

[0072] The final generated dynamic coupling matrix M(t) is an N-row, N-column real-valued matrix, where N is the number of robots participating in the collaboration. M(t) is used to quantify the attitude coupling strength of the entire multi-robot network at a certain moment.

[0073] S300: Extract the dynamic attitude coordination index value of each robot node using M(t), and calculate the node stability factor by combining the instantaneous connectivity of the node in the topology graph, and identify the attitude anomaly propagation chain and the set of key propagation nodes Vcrit.

[0074] S301: Based on the attitude differences and time synchronization offsets between each pair of cooperative robots in the dynamic coupling matrix M(t), a nonlinear adjacency weight function f(ΔPij, ΔTij) is calculated to extract the coupling weight set Wi between each robot node i and all its communicating neighbor nodes j at the current time. Specifically, for each pair of nodes i and j, at time t, the nonlinear adjacency weight function f(ΔPij, ΔTij) is calculated based on their attitude data differences ΔPij (i.e., the differences in acceleration, angular velocity, and magnetic field between the two robots in three-dimensional space) and the time offset ΔTij. This weight set Wi = {f(ΔPij, ΔTij)} is used to measure the cooperative relationship between node i and its neighbor nodes at time t.

[0075] S302: To further analyze the degree of cooperative fluctuation between node i and its neighboring nodes, the coupling average μi and standard deviation σi of node i are obtained by calculating the statistical characteristics of the coupling weight set Wi of node i. Specifically, μi represents the average coupling strength between node i and its neighboring nodes, and the calculation formula is: Σf(ΔPij,ΔTij) represents the sum of coupling weights over all neighboring nodes. σi is the coupling volatility of node i, calculated using the following formula: The standard deviation is calculated by summing the squares of the difference between the coupling weight of each neighbor node and the average value μi, and then dividing by the number of neighbor nodes n.

[0076] S303: Based on the coupling average μi and standard deviation σi of node i, a dynamic attitude coordination index value Ci is constructed for node i, which measures the degree of coordination between node i and its neighbors. The specific calculation formula is as follows: , where μi is the average coupling of node i, and σi is the standard deviation of coupling of node i. The index value Ci represents the degree of cooperation of node i. The larger the Ci value, the stronger the cooperation consistency between node i and its neighbors, and the smaller the Ci value, the weaker the cooperation of node i.

[0077] S304: Based on the dynamic attitude coordination index value Ci, if the coordination index value of node i is lower than the set dynamic threshold θc, then node i is marked as a candidate node with coordination abnormality.

[0078] The specific implementation method is as follows: compare the magnitude of Ci with the dynamic threshold θc. If Ci < θc, then node i is determined to be a collaborative abnormal node. This threshold θc is a dynamic value set according to the actual application scenario, which can be automatically adjusted according to environmental changes to ensure that the model captures abnormal nodes in real time.

[0079] S305: To further characterize the network connectivity of node i at the current moment, the instantaneous connectivity Di of node i is obtained by calculating the sum of the coupling weights between node i and all its neighboring nodes. Specifically, the set of coupling weights Wi = {f(ΔPij, ΔTij)} between node i and all its neighboring nodes j is calculated, and then these weights are summed to obtain the instantaneous connectivity of node i: Di ​​= Σf(ΔPij, ΔTij), where Σ represents the summation of the coupling weights over all neighboring nodes. Di represents the instantaneous connectivity strength of node i in the network; a larger value indicates a stronger connection between node i and other nodes.

[0080] S306: Based on the attitude coordination index Ci and the instantaneous connectivity Di, a node stability factor Si is constructed to evaluate the stability contribution of node i in the network. Specifically, the node stability factor Si = Ci × Di is calculated, where Ci is the attitude coordination index value of node i, and Di is the instantaneous connectivity of node i. This factor Si integrates the coordination consistency and connection strength of node i, and is used to measure the overall stability of node i.

[0081] S307: When the node stability factor Si is lower than the dynamic threshold θs, a propagation path tracing graph is constructed, and node i and its surrounding propagation paths are marked in the graph to identify potential attitude anomaly propagation chains. Specifically, based on the node stability factor Si and the network topology graph G(V,E), a breadth-first search (BFS) or depth-first search (DFS) algorithm is used to trace the propagation path originating from node i. Within the propagation path, each propagation path is marked according to the node stability factor along the path, and potential anomaly propagation chains are identified based on the stability of the path.

[0082] S308: In the propagation path, extract the set of key propagation nodes, Vcrit, based on the low values ​​of the path distribution degree or node stability factor. Specifically, based on the low values ​​of the distribution degree (i.e., the number of paths passing through a node) or stability factor in the path, select key nodes with significant influence; these nodes are called key propagation nodes. The set of key propagation nodes, Vcrit, is the core node set of the anomaly propagation chain, and its identification is helpful for subsequent anomaly posture repair and control.

[0083] S400. The original pose data of the nodes in Vcrit is broken, and local differential reconstruction estimation is performed from the high cooperative neighborhood of non-abnormal nodes to obtain the pose repair estimate, which constitutes the pose correction set.

[0084] After identifying the key propagation node set Vcrit, in order to avoid the misleading effect of abnormal data in subsequent processing, it is necessary to break the chain of its original attitude data.

[0085] In this step, for each node in the set of critical propagation nodes, its corresponding abnormal propagation period is determined. This period is determined by the start and end times marked on the propagation path tracing graph, denoted as [tstart,tend]. During this period, the original pose data Pi(t) is considered invalid and is not directly used for pose fusion processing. Specifically, Pi(t) within the time interval [tstart,tend] is replaced with null values ​​or "NaN" identifiers to indicate that the data segment is unusable; the pose data set after the chain break is used for subsequent interpolation and reconstruction operations to avoid abnormal propagation or misleading subsequent consensus estimation.

[0086] After the chain break is completed, in order to achieve reliable data reconstruction, it is necessary to identify a set of neighboring nodes that are in the neighborhood of the target node and have high cooperation.

[0087] This step relies on the cooperative communication topology graph G(V,E) and the calculated cooperative index value Ci and stability factor Si for screening. The screening rules are as follows: Set a cooperative threshold θc, preferably between 0.6 and 0.8, which represents the minimum acceptable level of cooperative consistency; set a stability factor threshold θs, preferably the lower percentile of the average stability value of the node group, usually the 25th percentile; for all neighboring nodes j that have direct communication connections with the target node i, if Cj>θc and Sj>θs are satisfied at the same time, they are included in the high cooperative neighborhood set.

[0088] To reconstruct the attitude of anomalous data segments, this implementation method utilizes the local consistency of the attitude change rate within a highly collaborative neighborhood set to construct a differential interpolation model for estimation and reconstruction. The specific steps are as follows: A time window τ is set, where τ is a positive integer representing the number of frames before and after the interpolation calculation, preferably 5 to 10; for any time t within the anomalous time period [tstart,tend] of target node i, a time window is extracted for each node j in its neighborhood. The pose data Pj(t') within the range;

[0089] For each neighboring node j, calculate the sequence of attitude change rates within that window, i.e. And average it within the window to obtain the local attitude differential vector Dj(t) of node j;

[0090] Using the coordination index value Cj of node j as the weight, a weighted average is performed on all Dj(t) to obtain the attitude estimate Pi(t) of target node i at time t; the reconstruction formula is expressed as: .

[0091] After calculating the attitude estimate Pi(t), it is filled into the corresponding abnormal time period data positions in chronological order to form the attitude repair estimation sequence Pi[tstart,tend]. This reconstructed sequence is merged with the original attitude data of node i to form the complete attitude correction result Picorrected(t), which is defined as follows: when t∈[tstart,tend], Picorrected(t)=Pi(t); when t does not belong to [tstart,tend], Picorrected(t)=original data Pi(t).

[0092] Finally, the data from all key propagation nodes after attitude correction are integrated to form an attitude correction set Pcorrected={Picorrected(t)}, which serves as the input basis for subsequent consistency fusion.

[0093] The S500 uses a recursive self-calculation model to perform evolution prediction and reverse verification of the posture correction set in the time domain, and outputs a posture fusion consistency field to drive the unified motion planning of the collaborative robot.

[0094] To achieve dynamic consistency evaluation of the attitude correction set in the time domain, it is first necessary to extract continuous time segments from the attitude correction set to construct an input subsequence set. The specific steps are as follows:

[0095] Set the time series window length τ, where τ is a positive integer, preferably between 5 and 15, and adjust it according to task requirements;

[0096] For each robot node i, in its attitude correction data sequence Picorrected(t), a sliding window method is used to extract an attitude subsequence Si(t) of length τ; each subsequence Si(t) = [Picorrected(t)]. [τ+1),...,Picorrected(t)], where each P is a 9-dimensional attitude vector (including triaxial acceleration, triaxial angular velocity, and triaxial magnetic field strength). The constructed attitude input subsequence set will be used as input to the recursive model for subsequent forward evolution prediction.

[0097] After constructing the subsequences, a short-term forward prediction is performed on each subsequence using a gated recurrent unit (GRU) model in a recurrent neural network to characterize the attitude evolution trend within the next τ steps. The implementation steps are as follows:

[0098] Each pose subsequence Si(t) is used as a time series input and fed into a pre-trained GRU neural network model for recursive computation. The network structure includes an input layer, a single-layer GRU recurrent unit, and an output mapping layer. Mean squared error is used as the loss function, and the model is pre-trained on multiple real robot sequences. The model output is a predicted pose sequence Pi(t+1), Pi(t+2),...,Pi(t+τ), each of which is a 9-dimensional prediction vector representing the pose estimate of node i in the next τ steps. This predicted sequence serves as the time-domain fitting result of the pose evolution trend of the current node, providing a reference for subsequent residual verification.

[0099] To evaluate the consistency between the predicted sequence and the actual observations, a reverse verification mechanism is constructed. By comparing the differences between the actual and predicted values, the collaborative credibility of nodes within a time window is extracted. The specific steps are as follows:

[0100] Extract the true observation sequence Picorrected(t+1),...,Picorrected(t+τ) of node i in the attitude correction set within the prediction interval; calculate the residual for each time step, where the residual is the Euclidean distance between the predicted and actual values;

[0101] The mean of the residuals over τ steps is denoted as Ri; the consistency confidence score Fi is calculated based on Ri, defined as: γ is the attenuation coefficient, with a preferred value between 0.1 and 0.5. A higher confidence score indicates a more reliable prediction and a more stable node state.

[0102] After obtaining the confidence score Fi for each node, the pose sequences of nodes with confidence scores higher than a set threshold θf are subjected to weighted fusion processing to output the final pose fusion consistency field. The specific operations are as follows: the threshold θf is an empirically set confidence lower limit, preferably 0.8; select all nodes that satisfy Fi≥θf, and extract their pose data Picorrected(t) at time t; calculate the fused pose vector Pfused(t) by weighting these data according to the confidence score Fi. , where Σ represents summing over the set of confidence nodes; perform the above calculation over all time points t to finally construct a complete temporally continuous and spatially consistent attitude fusion consistent field W={Pfused(t)}.

[0103] After generating the pose fusion consistency field W, the next step is to apply it to unified motion planning and control in a multi-robot system. The unified pose fusion consistency field provides a consistent and synchronized pose reference for all robots, facilitating collaborative actions and task allocation among them. Specifically:

[0104] The attitude fusion consistency field W={Pfused(t)} consists of a temporally continuous and spatially consistent sequence of attitude data. Pfused(t) at each time point t is a weighted average of the attitudes of multiple robots, reflecting the globally consistent attitude information in the multi-robot system.

[0105] After obtaining the posture fusion consistent field W, it is used as input to the unified motion planning module in the collaborative robot system. Based on the current fused posture information, combined with task requirements (such as path planning, obstacle avoidance, formation control, etc.) and constraints between robots, this module generates the globally optimal motion sequence.

[0106] Based on the optimal motion sequence obtained from the planning, the system sends specific motion commands (such as speed, angle, and acceleration) to each robot through a control algorithm. After receiving the commands, the robot control system adjusts the robot's motion state through its local execution control module (such as a PID controller, LQR controller, etc.).

[0107] Each robot adjusts its movements based on the globally consistent information provided by the posture fusion consistency field, ensuring coordination and synchronization among multiple robots.

[0108] Due to dynamic changes in the environment, the robot system may encounter new obstacles, path changes, or sensor errors, thus requiring real-time feedback and correction. In each control cycle, the robot continuously monitors environmental changes, updates its posture data, and exchanges information with other robots, forming a closed-loop control system.

[0109] If a robot deviates from its intended path or its actions become abnormal, the coordinated actions among multiple robots can be adjusted in real time by updating the posture fusion consistency field and recalculating the planned path to ensure the successful completion of the task.

[0110] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A robot posture data processing method based on a distributed cooperative network, characterized in that: include: S100. Obtain the raw posture data Pi of each collaborative robot and its corresponding timestamp Ti, and construct a time-calibrated posture data set. ; S200. Based on the attitude data set P and combined with the collaborative communication topology graph between robots, extract the attitude difference ΔPij and time offset ΔTij between any two robots, and use a nonlinear adjacency weighting function. Construct the dynamic coupling matrix M(t); S300. Use M(t) to extract the dynamic posture coordination index value of each robot node, and combine it with the instantaneous connectivity of the node in the topology graph to calculate the node stability factor, identify the posture anomaly propagation chain and the key propagation node set Vcrit. S400. The original pose data of the nodes in Vcrit is broken, and local differential reconstruction estimation is performed from the highly cooperative neighborhood of non-abnormal nodes to obtain the pose repair estimate, which constitutes the pose correction set. S500 uses a recursive self-calculation model to perform evolution prediction and reverse verification of the posture correction set in the time domain, and outputs a posture fusion consistency field to drive the unified motion planning of the collaborative robot. The dynamic posture coordination index value of each robot node is extracted using M(t), including: S301. Extract the set of coupling weights Wi={f(ΔPij,ΔTij)} between each robot node i and its communicating neighbor node j at the current time based on the dynamic coupling matrix M(t); S302. Calculate the coupling mean μi and standard deviation σi of node i to measure the degree of cooperative fluctuation between it and its neighboring nodes; S303, Construct dynamic attitude coordination index values S304. Based on the dynamic attitude coordination index value Ci, when the coordination index value Ci of node i is lower than the set dynamic threshold θc, node i is determined to be a coordination abnormal node. This involves combining the instantaneous connectivity of nodes in the topology graph to calculate node stability factors, identifying attitude anomaly propagation chains and the key propagation node set Vcrit, including: S305. Based on the dynamic coupling matrix M(t) and the cooperative communication topology graph, calculate the instantaneous connectivity Di of each robot node at the current moment, where the connectivity is the sum of the coupling weights of node i and all its neighboring nodes. S306. Perform joint normalization on the attitude coordination index value Ci and the instantaneous connectivity Di to construct a node stability factor. ; S307. Construct a propagation path tracing graph for nodes whose stability factor Si is lower than the dynamic threshold θs, and mark their shortest path transmission chain in the topology graph to identify potential attitude anomaly propagation chains. S308. Extract the key propagation node set Vcrit based on the node with the maximum path distribution degree or the lowest stability factor in the propagation path.

2. The robot posture data processing method based on a distributed cooperative network according to claim 1, characterized in that: The S100 includes: S101. Attitude data is collected by the inertial measurement unit built into each collaborative robot. The attitude data includes three-axis acceleration, three-axis angular velocity and three-axis magnetic field strength. S102. Denoise the original attitude data based on the Kalman filter algorithm; S103. Perform unified calibration on the timestamps collected from multiple collaborative robots to generate a globally aligned timestamp set; S104. Map the denoised attitude data one-to-one with the corresponding global alignment timestamps to construct a time-calibrated attitude data set P.

3. The robot posture data processing method based on a distributed cooperative network according to claim 1, characterized in that: The S200 includes: S201. Based on the time-calibrated attitude data in the attitude data set P, calculate the attitude difference ΔPij between any two collaborative robots at the same time point, where the attitude difference is a nine-dimensional Euclidean distance. S202. Calculate the time synchronization offset ΔTij between the two robots, where the time synchronization offset is the absolute difference between the timestamps corresponding to their respective attitude data. S203. Construct a nonlinear adjacency weight function f(ΔPij,ΔTij) based on the attitude difference ΔPij and the time offset ΔTij; S204. The calculation results of the nonlinear adjacency weight function are combined according to the connection relationship between each robot in the cooperative communication topology to construct a time-varying dynamic coupling matrix M(t).

4. The robot posture data processing method based on a distributed cooperative network according to claim 1, characterized in that: The S400 includes: S401. For each node in the critical propagation node set Vcrit, remove its original attitude data during the abnormal propagation period and mark the data of that period as null. S402. Identify a set of highly cooperative neighbor nodes in the cooperative communication topology graph. The neighbor nodes must satisfy the following conditions: the cooperative index value is greater than the threshold θc, and the stability factor is higher than the threshold θs. S403. In a highly collaborative neighborhood, a differential interpolation model with a time window of τ is constructed based on the rate of change of attitude difference between nodes, and the attitude estimate of the target node is reconstructed by weighted averaging. S404. Fill in the reconstructed attitude estimates with the original abnormal data segments one by one to generate an attitude repair estimation sequence, and summarize them to form an attitude correction set.

5. The robot posture data processing method based on a distributed cooperative network according to claim 4, characterized in that: The S500 includes: S501. Construct a set of attitude input subsequences based on a time series window, with each subsequence having a length of τ, and extract the attitude vector sequence from consecutive time segments in the attitude correction set. S502. Perform short-term forward prediction on each subsequence to generate a predicted attitude sequence, representing the attitude evolution trend within the next τ steps. S503. Compare the predicted sequence with the actual observations in the attitude correction set in reverse, and construct a consistency confidence score by calculating the mean of the predicted residuals within the time window. S504. Perform weighted fusion processing on the attitude sequences of nodes with confidence scores higher than the threshold θf to generate a temporally continuous and spatially consistent attitude fusion consistent field.

Images