Robot posture intelligent control method and system based on ad hoc network communication architecture

Abstract

The application relates to the technical field of robot posture control, and discloses a robot posture intelligent control method and system based on a self-organizing network communication architecture, which comprises the following steps: acquiring multi-dimensional environment initial data and robot current motion state data; performing multi-source data integration and key feature extraction operations according to the multi-dimensional environment initial data to obtain a fused environment feature set; performing risk identification and posture adjustment strategy generation operations according to the robot current motion state data and the fused environment feature set to obtain a preliminary posture adjustment strategy signal; and performing parameter optimization and secondary calibration operations on the preliminary posture adjustment strategy signal to obtain a final precise posture control instruction. The method can realize precise posture control of a robot in a complex dynamic environment and improve work safety and efficiency.

Description

Technical Field

[0001] This invention relates to the field of robot posture control technology, and in particular to a robot posture intelligent control method and system based on an ad hoc network communication architecture. Background Technology

[0002] Currently, in the field of modern intelligent manufacturing and automation, achieving accurate perception and intelligent posture control of complex dynamic environments is a core requirement for improving robot operation efficiency and safety, especially technological breakthroughs in multi-dimensional information processing and real-time response.

[0003] In one existing technology, environmental data is collected using one or a few sensors. This data is then input into a control algorithm for real-time analysis to determine the deviation between the robot's current posture and the desired posture. Based on the analysis results, corresponding control commands are generated to drive actuators such as motors or hydraulic devices to dynamically adjust the robot's posture. However, existing technologies have significant limitations when dealing with highly dynamic and complex real-world scenarios. Environmental perception methods struggle to effectively integrate visual data from multiple angles, failing to achieve real-time acquisition and in-depth analysis of comprehensive information. This directly limits the accurate judgment of the robot's own posture. The inadequacy of perception capabilities further restricts the real-time performance of posture control, making it difficult for the robot to quickly generate and execute effective posture adjustment commands when facing sudden obstacles or terrain changes.

[0004] In summary, existing technologies suffer from insufficient accuracy in robot posture adjustment commands under complex environments. Summary of the Invention

[0005] This invention relates to the field of robot posture control, and in particular to a robot posture intelligent control method and system based on an ad hoc network communication architecture, in order to solve the problem of insufficient accuracy of robot posture adjustment commands in complex dynamic environments in the prior art.

[0006] Firstly, to address the aforementioned technical problems, this invention provides a robot posture intelligent control method based on an ad hoc network communication architecture, comprising:

[0007] Obtain multi-dimensional initial environmental data;

[0008] Based on the initial multi-dimensional environmental data, multi-source data integration and key feature extraction operations are performed to obtain a fused environmental feature set.

[0009] Based on the set of environmental features, multi-angle visual data spatial distribution and change trend analysis are performed to obtain a panoramic information view of the dynamic environment.

[0010] The panoramic information view of the dynamic environment is divided into regions to obtain a set of divided view units, and operational risk points are identified based on the set of view units.

[0011] If there are potential operational risks, risk response triggering and preliminary attitude adjustment strategy signal generation operations are performed to obtain preliminary attitude adjustment strategy signals.

[0012] Based on the preliminary attitude adjustment strategy signal, the adjustment command parameters are optimized to obtain the final precise attitude control command.

[0013] In one optional implementation, acquiring multi-dimensional environmental initial data includes:

[0014] The sensor array acquires multi-angle data from the dynamic environment and preprocesses it to obtain an initial multi-dimensional data set.

[0015] The visual information in the initial multi-dimensional dataset is subjected to denoising and edge detection processing to extract clear visual feature data;

[0016] Based on the distance and terrain feature information in the initial multi-dimensional dataset, the spatial distribution in the environment is calculated, and the spatial structure data of the environment is determined.

[0017] If the matching degree between the spatial structure data and the visual feature data is lower than a preset matching degree threshold, then the two are calibrated to obtain a calibrated environmental information matrix.

[0018] The calibrated environmental information matrix is ​​interpolated, filled, and standardized to obtain a standardized environmental perception dataset.

[0019] Based on the environmental perception dataset, grouping is performed, and the distribution characteristics of each group are determined to obtain multi-dimensional initial environmental data.

[0020] In one optional implementation, based on the initial multi-dimensional environmental data, multi-source data integration and key feature extraction operations are performed to obtain a fused environmental feature set, including:

[0021] The initial data of the multi-dimensional environment is cleaned, and the cleaned data is time-stamped and missing values ​​are filled to obtain the complete dataset after filling.

[0022] Based on the completed dataset after filling in the gaps, the data in each dimension are filtered and processed to obtain significant feature data related to environmental perception.

[0023] The salient feature data is classified to obtain a grouped feature data set;

[0024] The data from different groups are integrated using the grouped feature data set. If the matching degree of the integrated data is lower than a preset threshold, consistency adjustment is performed to obtain the fused environmental feature set.

[0025] In one optional implementation, the step of performing multi-angle visual data spatial distribution and change trend analysis based on the environmental feature set to obtain a panoramic information view of the dynamic environment includes:

[0026] The environmental feature set is subjected to spatial coordinate transformation to obtain point data corresponding to the spatial distribution, thus obtaining the regional feature data before classification.

[0027] The data is transformed from the sensor's local coordinate system to the global coordinate system, generating point data containing three-dimensional coordinates.

[0028] Based on the regional feature data before classification, the regions are classified according to different regions to obtain the regional feature set after classification.

[0029] The classified regional feature data is processed to track changes in trends, and dynamic environmental change data over time is obtained to determine the completeness of the dynamic environmental change data.

[0030] If the completeness of the dynamic environment change data is lower than a preset completeness threshold, then the dynamic environment change data is filled with differences to obtain an adjusted dynamic environment data set.

[0031] Based on the adjusted dynamic environment data set, an all-round visual network model is constructed, and the spatial distribution and changing trend of multi-angle visual data are analyzed to obtain a panoramic information view of the dynamic environment.

[0032] In one optional implementation, the step of dividing the panoramic information view of the dynamic environment into regions to obtain a set of divided view units, and identifying operational risk points based on the set of view units, includes:

[0033] The terrain changes and obstacle areas in the panoramic information view are initially divided to obtain at least one independent terrain unit and obstacle unit, resulting in a set of divided view units.

[0034] Based on the set of view units, detailed features are captured from the terrain units and obstacle units, and the captured detailed features are classified in real time to determine the specific category and attribute status of each unit.

[0035] The specific category and the attribute status are matched with a pre-established threat database. If the attribute status matches the potential threat characteristics in the database, it is marked as a high-risk unit, thus obtaining a set of marked risk units.

[0036] Based on the marked set of risk units, the operational safety area is dynamically adjusted, and the panoramic information view of the dynamic environment is refreshed in real time in conjunction with a preset view update mechanism to determine the distribution of potential operational risk points.

[0037] In one optional implementation, if a potential operational risk exists, a risk response triggering and preliminary attitude adjustment strategy signal generation operation is performed to obtain a preliminary attitude adjustment strategy signal, including:

[0038] If there are potential operational risks, the panoramic information view of the dynamic environment is divided into regions to obtain at least one risk region unit, resulting in a set of divided risk regions.

[0039] The spatial location and risk category of the risk area units in the risk area set are captured in detail to determine the specific location coordinates and category labels of the units;

[0040] If the category label matches a high-risk feature in a pre-established threat assessment database, a corresponding real-time response instruction is generated, and the response signal triggered by the real-time response instruction is obtained.

[0041] Based on the response signal, the target operating area is dynamically planned, and a preliminary adjustment signal is generated in combination with the preset attitude adjustment logic;

[0042] Determine whether the preliminary adjustment signal meets the preset safety range. If it does, then determine the preliminary adjustment signal as the preliminary attitude adjustment strategy signal.

[0043] In one optional implementation, the step of optimizing the adjustment command parameters based on the preliminary attitude adjustment strategy signal to obtain the final precise attitude control command includes:

[0044] Based on the preliminary attitude adjustment strategy signal, the robot's current motion state data is obtained;

[0045] The current motion state data is compared with the pre-established task matching rules. If the deviation between the current motion state data and the target task exceeds the preset deviation threshold, a preliminary adjustment instruction is generated to obtain the basic parameter set for instruction adjustment.

[0046] The robot is dynamically adjusted based on the set of basic parameters to obtain dynamic adjustment data of the robot, and the parameters are optimized in combination with preset constraints to obtain the parameter optimization results.

[0047] Determine whether the parameterization result meets the preset precision control standard. If it does, obtain an intermediate parameter set applicable to the current scenario.

[0048] The robot is dynamically adjusted based on the intermediate parameter set, and the robot's status is monitored in real time to obtain the latest motion status information.

[0049] If the motion state information does not match the preset target task well enough, a second calibration is performed to obtain the final control parameter set.

[0050] The attitude is adjusted according to the final control parameter set, and it is determined whether the adjusted state meets the preset target task requirements. If it does, the final precise attitude control command is obtained.

[0051] In one optional embodiment, the sensor array includes a visual sensor, a distance sensor, and a terrain sensor, wherein the visual sensor is used to acquire visual information about the environment, the distance sensor is used to acquire distance information about the environment, and the terrain sensor is used to acquire terrain feature information about the environment.

[0052] Secondly, the present invention provides a robot posture intelligent control system based on a self-organizing network communication architecture, comprising:

[0053] The data acquisition module is used to acquire multi-dimensional initial environmental data;

[0054] The data fusion module is used to perform multi-source data integration and key feature extraction operations based on the multi-dimensional environmental initial data to obtain a fused environmental feature set.

[0055] The panoramic view construction module is used to perform multi-angle visual data spatial distribution and change trend analysis based on the set of environmental features to obtain a panoramic information view of the dynamic environment.

[0056] The risk identification module is used to divide the panoramic information view of the dynamic environment into regions to obtain a set of divided view units, and to identify operational risk points based on the set of view units.

[0057] The preliminary strategy generation module is used to trigger risk response and generate preliminary attitude adjustment strategy signals if there are potential operational risk points, thus obtaining preliminary attitude adjustment strategy signals.

[0058] The precise instruction determination module is used to optimize the instruction parameters based on the preliminary attitude adjustment strategy signal to obtain the final precise attitude control instruction.

[0059] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a robot posture intelligent control method based on an ad hoc network communication architecture as described in any one of the above.

[0060] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a power control method for subway emergency lighting as described in any one of the above.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] (1) This invention performs noise filtering, timestamp alignment, and missing value completion on multi-source heterogeneous data collected from dynamic environments. Then, through spatial coordinate transformation and regional classification modeling, it integrates and tracks the dynamic trends of environmental data. Ultimately, it achieves high-precision perception of panoramic information of dynamic environments, enabling robots to more accurately grasp the layout of the work scene and the motion state of elements. This solves the problems of incomplete data from traditional single sensors and delayed environmental response, laying a solid data foundation for risk identification and attitude adjustment.

[0063] (2) This invention uses a region growing algorithm to divide the dynamic environment panoramic view, extracts and classifies features to clarify unit attributes, links a threat database to match and mark high-risk units, and simultaneously adjusts the safe area and refreshes the view in real time. Ultimately, it achieves dynamic and accurate identification of potential operational risk points, improves the robot's ability to predict and avoid dangers, solves the problem of traditional risk identification relying on experience and lagging response, and builds a solid safety barrier for complex scene operations.

[0064] (3) This invention collects robot motion state data from the initial posture adjustment strategy signal, compares it with the task rules, and generates a basic parameter set if the deviation exceeds the limit. After optimization by PID algorithm, multiple rounds of calibration and safety verification, it finally achieves high-precision iteration of posture control commands, allowing the robot's working posture to adapt to task requirements, solving the problems of traditional commands being coarse and difficult to adapt to dynamic scenarios, and improving work efficiency and quality. Attached Figure Description

[0065] Figure 1 is a schematic flowchart of the robot posture intelligent control method based on self-organizing network communication architecture provided in the first embodiment of the present invention;

[0066] Figure 2 is a schematic diagram of the robot posture intelligent control system based on self-organizing network communication architecture provided in the second embodiment of the present invention. Detailed Implementation

[0067] 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, and 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.

[0068] Reference Figure 1 The first embodiment of the present invention provides a robot posture intelligent control method based on an ad hoc network communication architecture, comprising the following steps:

[0069] S101, acquire multi-dimensional initial environmental data;

[0070] S102, based on the initial multi-dimensional environmental data, perform multi-source data integration and key feature extraction operations to obtain a fused environmental feature set;

[0071] S103, Based on the set of environmental features, perform multi-angle visual data spatial distribution and change trend analysis to obtain a panoramic information view of the dynamic environment;

[0072] S104, the panoramic information view of the dynamic environment is divided into regions to obtain a set of divided view units, and operational risk points are identified based on the set of view units.

[0073] S105, if there are potential operational risk points, perform risk response triggering and preliminary attitude adjustment strategy signal generation operations to obtain preliminary attitude adjustment strategy signals.

[0074] S106, Based on the preliminary attitude adjustment strategy signal, perform adjustment command parameter optimization operation to obtain the final precise attitude control command.

[0075] In step S101, multi-dimensional environmental initial data is obtained.

[0076] In one feasible approach, acquiring the multi-dimensional environmental initial data includes:

[0077] The sensor array acquires multi-angle data from the dynamic environment and preprocesses it to obtain an initial multi-dimensional data set.

[0078] The visual information in the initial multi-dimensional dataset is subjected to denoising and edge detection processing to extract clear visual feature data;

[0079] Based on the distance and terrain feature information in the initial multi-dimensional dataset, the spatial distribution in the environment is calculated, and the spatial structure data of the environment is determined.

[0080] If the matching degree between the spatial structure data and the visual feature data is lower than a preset matching degree threshold, then the two are calibrated to obtain a calibrated environmental information matrix.

[0081] The calibrated environmental information matrix is ​​interpolated, filled, and standardized to obtain a standardized environmental perception dataset.

[0082] Based on the environmental perception dataset, grouping is performed, and the distribution characteristics of each group are determined to obtain multi-dimensional initial environmental data.

[0083] It should be noted that multi-angle data from a dynamic environment is synchronously collected through a sensor array, which includes visual sensors, distance sensors, and terrain feature sensors. After collection, visual information is categorized by resolution, distance information by measurement range, and terrain feature information by type, forming an initial multi-dimensional data set. For example, in an autonomous driving scenario, a camera captures road images, a lidar measures the distance to obstacles ahead (e.g., 10.5 meters), and an inertial measurement unit combines a digital map to obtain slope and road surface roughness, forming a structured data set after classification.

[0084] Then, the visual information in the initial dataset is denoised to remove noise interference caused by illumination fluctuations, and contour features are extracted using an edge detection algorithm to obtain clear visual feature data. For example, after Gaussian filtering is used to eliminate noise in road images, continuous lane line edge contours are extracted using the Canny operator.

[0085] Simultaneously, based on distance and terrain feature information, the spatial distribution of the environment is calculated using a triangulation algorithm. Specifically, by combining lidar ranging data with terrain slope information, the three-dimensional spatial coordinates of obstacles are calculated, generating spatial structure data of the environment. For example, if the lidar measures an obstacle at a distance of 10.5 meters, and combined with a 5-degree slope, its precise position in the vehicle coordinate system is calculated (X=10.5m, Y=2.3m, Z=0.2m).

[0086] If the matching degree between spatial structure data and visual feature data is lower than a preset threshold (e.g., positional deviation exceeds 5 meters), calibration is performed: First, the timestamps of the two types of data are aligned to ensure time synchronization; then, the positional deviation is corrected using a spatial coordinate transformation method. For example, when the visual detection of the obstacle's rectangular outline deviates from the lidar positioning by 5 meters, after aligning the data timestamps, the visual coordinates are mapped to the spatial coordinate system through an affine transformation to achieve positional calibration.

[0087] For missing areas in the calibrated environmental information matrix, linear interpolation is used to fill in the gaps based on surrounding valid data. Then, standardization is performed: visual data is converted into a grayscale matrix ranging from 0 to 255, distance data is uniformly converted to metric units, and terrain data is normalized to a slope coefficient in the 0-1 range. For example, if distance data for a certain area is missing, based on data of 12 meters in the previous frame and 9 meters in the next frame, the missing value is estimated to be 10.5 meters through interpolation; the terrain slope is normalized to 0.15 based on a 15% normalization.

[0088] Finally, the standardized dataset is grouped and processed according to feature dimensions. For example, the distance data is divided into near-field (0-5 meters), mid-field (5-20 meters), and other groups. The distribution characteristics of each group are analyzed (e.g., the near-field group detects 3 obstacles and they are densely distributed), and the final multi-dimensional environmental initial data is generated.

[0089] In summary, by employing multi-source data classification and acquisition, optimized visual feature extraction, precise spatial structure calculation, cross-modal data calibration, missing value interpolation and imputation, and standardized grouping processing, the completeness and reliability of environmental perception data are significantly improved. For example, in autonomous driving testing, obstacles on a 15% slope 10.5 meters ahead can be accurately identified, providing high-precision input for subsequent real-time attitude control and effectively solving the problem of fragmented perception data in dynamic environments. The standardization and grouping mechanisms further accelerate the system's efficiency in analyzing environmental features, supporting rapid decision-making in complex scenarios.

[0090] In step S102, based on the initial multi-dimensional environmental data, multi-source data integration and key feature extraction operations are performed to obtain a fused environmental feature set.

[0091] In one feasible approach, based on the initial multi-dimensional environmental data, multi-source data integration and key feature extraction operations are performed to obtain a fused environmental feature set, including:

[0092] The initial data of the multi-dimensional environment is cleaned, and the cleaned data is time-stamped and missing values ​​are filled to obtain the complete dataset after filling.

[0093] Based on the completed dataset after filling in the gaps, the data in each dimension are filtered and processed to obtain significant feature data related to environmental perception.

[0094] The salient feature data is classified to obtain a grouped feature data set;

[0095] The data from different groups are integrated using the grouped feature data set. If the matching degree of the integrated data is lower than a preset threshold, consistency adjustment is performed to obtain the fused environmental feature set.

[0096] It should be noted that the initial multi-dimensional environmental data undergoes data cleaning to filter out redundant noise caused by sensor malfunctions or environmental interference. Specifically, isolated points (such as invalid values ​​exceeding the effective range) in the distance data and blurred areas in the visual data are removed to form a cleaned dataset. For example, in autonomous driving scenarios, invalid ranging points exceeding 100 meters in the point cloud data collected by LiDAR are directly excluded; patch noise caused by strong light reflection in the image is removed through pixel threshold filtering.

[0097] The cleaned dataset undergoes timestamp synchronization to align sensor data from different acquisition frequencies to a unified time base. If any time points are missing, the missing values ​​are filled using linear interpolation based on valid data from adjacent time periods, generating a complete dataset with the missing data. For example, if the camera acquires data at 30 frames per second and the LiDAR at 10 times per second, the LiDAR data is matched to the timestamp of the most recent frame. If LiDAR data is missing at a certain moment, the distance data for that moment (10 meters) is interpolated based on the distance measurements of 9 meters in the previous frame and 11 meters in the next frame.

[0098] Based on the completed dataset after imputation, salient feature data directly related to environmental perception are selected: obstacle edge contours and lane line features are extracted from visual data; key obstacle information within a preset range (such as targets within 20 meters) is retained from distance data; and areas with slope changes exceeding a set threshold (such as slope difference ≥ 5%) are identified from terrain data. For example, in a road environment, pedestrian contour features in images, vehicle distance data within 15 meters detected by LiDAR, and information on road sections with slope abrupt changes exceeding 8% are extracted.

[0099] The filtered salient feature data are classified according to spatial distribution or functional attributes: distance data are divided into near distance group (0-5 meters) and medium distance group (5-20 meters), and visual features are divided into static object group (such as traffic signs) and dynamic object group (such as pedestrians), forming a set of grouped feature data. For example, pedestrians 3 meters ahead are classified into the "near distance dynamic object group", and stationary vehicles 20 meters away are classified into the "medium distance static object group".

[0100] The feature data from different groups are integrated and processed. If the matching degree of the integrated data is lower than a preset threshold (e.g., the deviation between visual and radar positioning exceeds 2 meters), dynamic consistency adjustment is performed: the data source is aligned through spatial coordinate transformation, and finally a fused set of environmental features is generated. For example, when the pedestrian position recognized by the image differs from the radar positioning by 2.5 meters, the visual coordinates are mapped to a unified spatial system through affine transformation based on the radar coordinates, forming a consistent "dynamic pedestrian 3 meters ahead" feature entry.

[0101] In summary, by improving data quality through noise filtering, ensuring temporal consistency through time synchronization, identifying key environmental elements through feature selection, optimizing data structure through classification processing, and resolving cross-sensor bias through dynamic calibration, the accuracy of environmental perception is significantly enhanced. For example, in autonomous driving, the integrated feature set can accurately describe complex scenarios such as "a stationary vehicle on a road section with an 8% abrupt change in slope 15 meters ahead," providing highly reliable input for subsequent risk assessment. The efficient fusion mechanism of multi-source data is particularly suitable for the rapid identification of sudden obstacles in dynamic environments, solving the decision-making lag problem caused by data fragmentation in traditional methods.

[0102] In step S103, based on the set of environmental features, a multi-angle visual data spatial distribution and change trend analysis operation is performed to obtain a panoramic information view of the dynamic environment.

[0103] In one optional implementation, the step of performing multi-angle visual data spatial distribution and change trend analysis based on the environmental feature set to obtain a panoramic information view of the dynamic environment includes:

[0104] Based on the set of environmental features, multi-angle visual data spatial distribution and change trend analysis are performed to obtain a panoramic information view of the dynamic environment.

[0105] The environmental feature set is subjected to spatial coordinate transformation to obtain point data corresponding to the spatial distribution, thus obtaining the regional feature data before classification.

[0106] Based on the regional feature data before classification, the regions are classified according to different regions to obtain the regional feature set after classification.

[0107] The classified regional feature data is processed to track changes in trends, and dynamic environmental change data over time is obtained to determine the completeness of the dynamic environmental change data.

[0108] If the completeness of the dynamic environment change data is lower than a preset completeness threshold, then the dynamic environment change data is filled with differences to obtain an adjusted dynamic environment data set.

[0109] It should be noted that when performing spatial coordinate transformation on the environmental feature set, the local coordinate system data from different sensors need to be uniformly transformed to the global coordinate system to generate point data containing three-dimensional coordinates (such as X, Y, and Z axis coordinates). Specifically, coordinate system alignment is achieved through a homogeneous coordinate transformation matrix, converting the two-dimensional pixel coordinates (u, v) collected by the camera, combined with depth information, into three-dimensional spatial coordinates (x, y, z) centered on the robot body. For example, in an autonomous driving scenario, if the camera detects an obstacle pixel with image coordinates (320, 240), combined with the 10.5-meter distance data measured by the LiDAR, and using the calibrated camera intrinsic and extrinsic parameter matrices, the actual spatial position of the obstacle in the vehicle coordinate system is calculated to be (10.5, -2.3, 0.2) meters, forming the region feature data before classification.

[0110] When classifying regions based on the pre-classification regional feature data, the regions are divided into forward, lateral, and backward regions according to spatial orientation. The forward region can be defined as the range of 0-30 meters directly in front of the robot, and the lateral regions are ranges of 5 meters to the left and right. Each region contains a subset of visual, distance, and terrain features. For example, the forward region subset may contain the visual outline of obstacles in front, distance measurement data of 10.5 meters, and terrain information with a slope of 5 degrees, thus forming the regional feature set after classification.

[0111] When tracking the changing trends of classified regional feature data, a time series model is established for the classified regional feature data, and the motion trajectory of feature points is tracked through Kalman filtering or recursive least squares method. The position and velocity vectors of obstacles in each region are collected at fixed time intervals (e.g., 100ms) to generate a dynamic environmental change data sequence. For example, if the coordinates of an obstacle in the foreground change from (20,0) → (18,0) → (16,0) meters for 5 consecutive frames, it is determined that the obstacle is approaching at a speed of 2 meters per second, and this trend data is recorded. An integrity threshold (e.g., 85%) is set, and the effective sampling rate of the dynamic environmental change data within a unit time window is calculated. If more than 2 out of 10 consecutive sampling points are missing (integrity 80% < 85%), cubic spline interpolation is used to reconstruct the missing values ​​based on the effective data before and after. For example, the obstacle positions at times t1-t3 are P1(10,2), P2(missing), and P3(8,2). Based on the coordinates and velocity directions of P1 and P3, the coordinates of time P2 are interpolated to be (9,2) meters, forming a complete dynamic environment data set.

[0112] Based on the adjusted dynamic environment dataset, a comprehensive visual network model is constructed by fusing feature data from various regions in the following specific ways: First, for the regional feature data (including images, distance, speed, etc.) collected by each node under the ad hoc network communication architecture, an adaptive weighted fusion algorithm is adopted. Dynamic weights are assigned to different regional data according to the reliability of the data source (such as node device accuracy and communication signal strength). For example, obstacle distance data collected by high-precision LiDAR nodes is given higher weights, while visual sensor data affected by occlusion is appropriately reduced in weight. Second, an attention mechanism module is introduced to focus on the feature data of key areas around the robot (such as within 10 meters in front), strengthening the feature extraction of dense dynamic target areas. Finally, data from different regions are integrated, and a multi-layer convolutional neural network (CNN) is used to extract spatial features of each region (such as obstacle shape and size). The CNN includes an input layer (receiving the fused feature map), three convolutional layers (using 32, 64, and 128 3×3 convolutional kernels to extract features, respectively), and two pooling layers (2×2 max pooling). Simultaneously, a two-layer LSTM of a recurrent neural network (RNN) is used. It can process temporal correlation information (such as position changes of a target in continuous motion) without additional training. It can directly use the preset network structure parameters to perform feature fusion and build a comprehensive visual network model with both spatial perception and temporal correlation capabilities.

[0113] When combining time-series analysis to label the motion trajectory of dynamic targets, the historical position data of the dynamic targets is smoothed using the Kalman filter algorithm: position and velocity are used as state variables, and a state transition matrix is ​​constructed based on the sensor sampling period (e.g., 100ms). A noise covariance matrix corresponding to the sensor measurement accuracy is set, and the predicted state and covariance of the next moment are calculated based on the state transition matrix during the prediction phase; during the update phase, the optimal estimate is calculated using Kalman gain based on the current position data collected by the sensor, and iterative optimization is performed to eliminate measurement noise. Based on the historical trajectory, the motion trend within the next 3 seconds is predicted, thereby generating a dynamic environmental panoramic information view containing spatial distribution and temporal trend. Taking the foreground area as an example, this view integrates the calibrated obstacle spatial distribution (e.g., coordinates (9,2) meters), motion trend (-1 m / s along the X-axis), and area classification labels (high-risk areas), and visualizes the 360-degree environmental state around the robot in a bird's-eye view projection, providing a comprehensive environmental basis for subsequent risk identification.

[0114] In summary, this process achieves unified calibration of environmental feature data in the spatial dimension and continuous tracking in the temporal dimension, effectively integrating the spatial distribution and changing trends of multi-angle visual data. This provides a comprehensive and accurate environmental basis for subsequent risk identification and posture adjustment, significantly improving the robot's perception capabilities and adaptability in complex dynamic environments.

[0115] In step S104, the panoramic information view of the dynamic environment is divided into regions to obtain a set of divided view units, and operational risk points are identified based on the set of view units.

[0116] In one feasible approach, the panoramic information view of the dynamic environment is divided into regions to obtain a set of divided view units, and operational risk points are identified based on the set of view units, including:

[0117] The terrain changes and obstacle areas in the panoramic information view are initially divided to obtain at least one independent terrain unit and obstacle unit, resulting in a set of divided view units.

[0118] Based on the set of view units, detailed features are captured from the terrain units and obstacle units, and the captured detailed features are classified in real time to determine the specific category and attribute status of each unit.

[0119] The specific category and the attribute status are matched with a pre-established threat database. If the attribute status matches the potential threat characteristics in the database, it is marked as a high-risk unit, thus obtaining a set of marked risk units.

[0120] Based on the marked set of risk units, the operational safety area is dynamically adjusted, and the panoramic information view of the dynamic environment is refreshed in real time in conjunction with a preset view update mechanism to determine the distribution of potential operational risk points.

[0121] It should be noted that, based on the pixel distribution features of the panoramic information view, a pre-trained semantic segmentation convolutional neural network model automatically divides terrain change areas and obstacle areas. This model aggregates spatially adjacent and feature-similar pixels into independent units based on pixel depth differences and texture continuity: terrain units (such as areas with continuous slope) and obstacle units (such as closed contour areas). For example, in an autonomous driving panoramic view, the model segments areas with abrupt changes in depth value 10 meters ahead as "gully terrain units" and identifies moving elliptical contours on the left as "pedestrian obstacle units," generating a set of view units containing multiple independent units.

[0122] It is worth noting that the core of the "pre-trained semantic segmentation convolutional neural network model" lies in using a deep convolutional neural network to perform semantic classification and region segmentation on the pixel features of the panoramic information view, so as to accurately divide the terrain change area and obstacle area.

[0123] Model Structure: The semantic segmentation convolutional neural network model used in this invention is an improved U-Net architecture. The input layer of this network receives pixel data (including RGB color channels, depth channels, and texture feature channels) from a panoramic view. Each channel corresponds to a normalized pixel feature parameter (such as pixel brightness value, depth coordinate value, and texture gradient value). The encoder part of the network contains four convolutional modules, each consisting of two 3×3 convolutional kernels, a BatchNormalization layer, and a ReLU activation function. It progressively extracts local features and global context information of pixels and achieves feature dimensionality reduction through a max pooling layer. The decoder part is symmetrical to the encoder. It restores the feature map size through upsampling operations and performs skip connections with the feature maps of the corresponding layers of the encoder to fuse feature information of different scales. Finally, the output layer with 1×1 convolutional kernels generates a semantic segmentation mask with the same size as the input view, where each pixel corresponds to a category label of terrain or obstacle (such as "flat ground", "slope", "pedestrian", "vehicle", etc.).

[0124] Model Training: The model is trained using supervised learning. Training data comes from panoramic environmental views collected by the robot in different scenarios and manually annotated semantic segmentation labels (including terrain type and obstacle category). Pixel data from the panoramic environmental views is used as input, and manually annotated semantic segmentation labels are used as the target output. The difference between the predicted and true labels is calculated using the cross-entropy loss function. Iterative training is performed on a large-scale dataset using the SGD optimizer (with an initial learning rate of 0.001, decaying to 0.1 every 10 epochs). Data augmentation techniques (such as random cropping, rotation, and brightness adjustment) are used to improve the model's generalization ability until the loss function converges and the intersection-over-union (IoU) on the validation set reaches a preset threshold (e.g., above 0.85). The trained model can automatically identify and segment terrain change areas and obstacle areas from new panoramic information views.

[0125] When extracting detailed features from the view unit set, terrain units are analyzed by extracting slope angle (calculated using the arctangent value from the depth coordinates of the unit boundary points) and surface curvature (based on the rate of change of the normal vectors of adjacent triangular faces); obstacle units are analyzed by extracting motion velocity (time difference between removing the centroid positions of two consecutive frames) and size parameters (volume of the smallest circumscribed cube). The feature vectors are input into a cascaded classifier. The first stage uses a support vector machine to distinguish major unit categories (terrain / obstacles), and the second stage uses a decision tree to further subdivide subcategories (e.g., slope / steps, pedestrians / vehicles). For example, if a unit with a slope angle of 15° and a curvature greater than 0.25 is detected, it is classified as a "steep slope terrain unit"; the moving unit size is 0.5m. It has a speed of 1.5 m / s and is classified as a "cyclist obstacle course".

[0126] When matching specific categories and attribute states with the threat database, the database stores quantified threat characteristic thresholds, such as "terrain units with a slope > 12° pose a risk of overturning" and "moving obstacles with a distance < 3m and a speed > 1.5m / s pose a risk of collision." If an obstacle unit is 2.5m away from the robot and has a speed of 2m / s, and its attribute state matches the "high-risk collision risk" feature in the database, it is marked as a high-risk unit, forming a set of risk units.

[0127] Based on the spatial distribution of risk unit sets, safe zones are dynamically divided: using the robot's position as the base point and the risk unit coordinates as the core, a hazardous impact zone (such as a circular restricted area with a radius equal to the unit velocity multiplied by the response time) is generated, and a polygonal safe zone with avoidance paths is generated within the reachable space. Simultaneously, incremental view updates are initiated: every 100 milliseconds, new point cloud data from the LiDAR is fused, and the panoramic view terrain model is updated through triangular mesh reconstruction.

[0128] The panoramic view terrain model is updated through the following specific process: First, for newly acquired LiDAR point cloud data, a voxel-based downsampling algorithm is used to simplify the data, setting the point cloud resolution to 0.05 meters to retain key feature points while removing redundant points to reduce data volume. Second, the Random Sample Consensus (RANSAC) algorithm is used to segment the ground and non-ground points in the point cloud, eliminating interference from dynamic targets and focusing on terrain structure data. Next, the new point cloud data is fused with the panoramic view terrain model based on an incremental Poisson surface reconstruction algorithm: First, a directed point set containing the vertices of the new point cloud and the historical model is constructed. The normal vector of each point is calculated and its direction is determined by fitting a plane using neighborhood points, and this is used as the gradient constraint condition for the Poisson equation. Then, a three-dimensional voxel mesh is established, and a Poisson equation system is constructed with the mesh nodes as unknowns. The source terms of the equation are determined by the projection of the normal vectors of the directed points onto the mesh. The solution method... The process yields the implicit signed distance function (negative values ​​indicate the interior of the model, and positive values ​​indicate the exterior). Then, combining the triangular mesh boundaries of the panoramic view terrain model, isosurface extraction is performed on the signed distance function to generate an initial triangular mesh that integrates new point cloud features. For overlapping areas between the initial mesh and the historical model, vertex distance is calculated for matching and alignment, preserving the more accurate parts of the historical model (such as recently updated terrain details). Only the mesh in the newly added point cloud coverage area is reconstructed, achieving seamless integration of old and new data. Then, triangular mesh topology optimization (such as edge shrinkage and vertex smoothing) is used to update terrain surface details, ensuring model continuity. Finally, a time decay factor is introduced to reduce the weight of point cloud data in areas not updated for more than 3 seconds, prioritizing the preservation of the most recently observed terrain features, thereby achieving dynamic updates to the panoramic view terrain model.

[0129] For example, when a new rolling stone unit (coordinates (5,7), speed 3m / s) is detected to the right front, a circular restricted area with a diameter of 6 meters is immediately generated in the safe area, and the predicted rolling stone trajectory area (orange gradient area) is displayed as a heat map in the updated panoramic view.

[0130] Through the above steps, a full-process quantitative processing from area division to risk positioning is achieved. This not only solves the one-sidedness of single feature identification, but also ensures the timeliness of risk identification through a real-time update mechanism, providing highly reliable risk coordinates and level basis for triggering precise attitude adjustment strategies in the future.

[0131] In step S105, if there are potential operational risk points, a risk response triggering and preliminary attitude adjustment strategy signal generation operation is performed to obtain a preliminary attitude adjustment strategy signal.

[0132] In one feasible approach, if potential operational risks exist, a risk response triggering and preliminary attitude adjustment strategy signal generation operation is performed to obtain a preliminary attitude adjustment strategy signal, including:

[0133] If there are potential operational risks, the panoramic information view of the dynamic environment is divided into regions to obtain at least one risk region unit, resulting in a set of divided risk regions.

[0134] The spatial location and risk category of the risk area units in the risk area set are captured in detail to determine the specific location coordinates and category labels of the units;

[0135] If the category label matches a high-risk feature in a pre-established threat assessment database, a corresponding real-time response instruction is generated, and the response signal triggered by the real-time response instruction is obtained.

[0136] Based on the response signal, the target operating area is dynamically planned, and a preliminary adjustment signal is generated in combination with the preset attitude adjustment logic;

[0137] Determine whether the preliminary adjustment signal meets the preset safety range. If it does, then determine the preliminary adjustment signal as the preliminary attitude adjustment strategy signal.

[0138] It should be noted that the panoramic information view based on the dynamic environment is used for region division. The specific execution process is as follows: First, the panoramic view is scanned at the pixel level, and pixel clusters of terrain abrupt changes or obstacle outlines are identified by edge detection operators. For the detected pixel clusters, the area of ​​their minimum bounding rectangle is calculated. If the area exceeds a preset threshold (e.g., 0.5 square meters), it is determined to be a valid risk area unit. Each risk area unit records its geometric center point coordinates, boundary vertex coordinates, and region type identifier. For example, in an autonomous driving scenario, when an irregular pothole is detected on the road ahead, the system extracts the edge pixels of the pothole using the Sobel operator, calculates its minimum bounding rectangle area to be 1.2 square meters (1.8 meters long × 0.7 meters wide), generates a risk unit numbered #R01, records the center point coordinates in the vehicle coordinate system (X=15.3m, Y=0.2m), and marks it as a "road surface defect" type.

[0139] Feature extraction is performed on the divided risk area set: For the spatial location of each risk unit, its absolute coordinates are converted into polar coordinates (distance d, azimuth angle θ) relative to the robot body using a coordinate transformation matrix. Simultaneously, the unit type identifier is parsed, and dynamic risk parameters are calculated based on real-time motion state data (such as the robot's current speed v and attitude angle φ). Taking a moving obstacle as an example, if the coordinates of the lateral area unit #R02 (type "moving object") are detected as (d=3.5m, θ=45°), combined with the robot's current speed v=1.5m / s and the object's trajectory, the minimum relative distance d_min=1.8m within the next 2 seconds is predicted. This parameter is matched against the threat assessment database: The database pre-stores the correspondence between various high-risk features and response commands. For example, the high-risk feature entry "d_min<2m and θ∈[30°,150°]" completely matches this entry, triggering the obstacle avoidance response command #Alert02. The database uses a tree index structure, and the matching response time is controlled within 10ms.

[0140] Dynamic path planning is performed based on real-time response commands: a two-dimensional gridded environment model (grid resolution 0.1m×0.1m) is established with the robot's current position as the starting point of the path and the target operation area as the ending point.

[0141] A two-dimensional gridded environment model is established through the following specific process: First, based on the segmented terrain and obstacle unit data in the panoramic information view, the 360-degree environment space around the robot is divided into uniform grid units of 0.1m × 0.1m, with each grid corresponding to a tiny area in the actual physical space. Second, each grid is assigned a multi-dimensional attribute label, including terrain attributes (such as "flat", "slope", "gully", determined based on the slope and flatness data of the terrain unit), obstacle attributes (such as "no obstacle", "static obstacle", "dynamic obstacle", labeled according to the type and motion state of the obstacle unit), and risk level (such as "low risk", "medium risk", "high risk", calculated by combining the influence range and motion trend of the risk unit). Finally, the updated environmental data of each node is synchronized in real time through an ad hoc network communication architecture, and the grid attributes are dynamically corrected. For example, when a new dynamic obstacle is detected in a grid, its obstacle attribute is immediately updated to "dynamic obstacle" and the risk level is increased, thereby constructing a two-dimensional gridded environment model that can reflect environmental changes in real time.

[0142] An improved A* algorithm is used for path search, in which a risk unit penalty term is added to the heuristic function: ( These are the spatial coordinates of the current path node. These are the spatial coordinates of the target endpoint. Let n be the distance from node n to the k-th risk unit. For unit risk coefficient, and These are the path length weighting coefficient and the risk avoidance strength coefficient, which are set based on the robot's motion efficiency requirements and safety level requirements, respectively (e.g., 0.8 and 0.7). For example, for the #R01 pit unit ( =0.8), a safety buffer zone of 0.5m from its boundary is automatically generated during path planning. After planning, the attitude parameters are calculated: for wheeled robots, the left and right wheel speed difference Δω=K·Δθ (K is the steering gain coefficient, Δθ is the path yaw angle) is calculated based on the differential drive model; for legged robots, the joint angle sequence is calculated through inverse kinematics. Finally, a preliminary adjustment signal containing the target pose (x,y,φ) and motion parameters (v,ω) is generated.

[0143] The generated preliminary adjustment signals undergo safety verification: First, a robot dynamics constraint model is established. For example, the maximum centripetal acceleration constraint for a wheeled robot is a_max ≤ μ·g (μ is the friction coefficient, and g is the gravitational acceleration). For the steering command ω, the verification condition is |v·ω| ≤ a_max; for the speed command v, the verification condition is |Δv / Δt| ≤ j_max (j_max is the maximum jerk). For example, when the planned path requires the vehicle to complete an 8° turn (ω = 0.14 rad / s) within 1 second and the speed v = 0.8 m / s, the centripetal acceleration is calculated as a = v·ω = 0.112 m / s². Under dry road surface conditions (μ=0.8, a_max=7.84m / s) The condition is that a ≤ a_max and the rate of change of acceleration is 0.4 m / s². ≤3m / s The signal is deemed valid based on the constraints. If the verification fails, iterative optimization of the instructions is initiated: ω and v are adjusted using gradient descent until all safety constraints are met.

[0144] In summary, this complete process, through hierarchical identification, precise matching, dynamic planning, and safety verification, enables the robot to quickly and reasonably generate initial risk response posture commands in complex dynamic scenarios. This lays a solid foundation for subsequent command optimization and precise control, effectively improving the timeliness and scientific nature of the robot's response to potential risks and ensuring the safe progress of operations.

[0145] In step S106, based on the preliminary attitude adjustment strategy signal, the adjustment command parameters are optimized to obtain the final precise attitude control command.

[0146] In one feasible approach, based on the preliminary attitude adjustment strategy signal, an adjustment command parameter optimization operation is performed to obtain the final precise attitude control command, including:

[0147] Based on the preliminary attitude adjustment strategy signal, the robot's current motion state data is obtained;

[0148] The current motion state data is compared with the pre-established task matching rules. If the deviation between the current motion state data and the target task exceeds the preset deviation threshold, a preliminary adjustment instruction is generated to obtain the basic parameter set for instruction adjustment.

[0149] The robot is dynamically adjusted based on the set of basic parameters to obtain dynamic adjustment data of the robot, and the parameters are optimized in combination with preset constraints to obtain the parameter optimization results.

[0150] Determine whether the parameterization result meets the preset precision control standard. If it does, obtain an intermediate parameter set applicable to the current scenario.

[0151] The robot is dynamically adjusted based on the intermediate parameter set, and the robot's status is monitored in real time to obtain the latest motion status information.

[0152] If the motion state information does not match the preset target task well enough, a second calibration is performed to obtain the final control parameter set.

[0153] The attitude is adjusted according to the final control parameter set, and it is determined whether the adjusted state meets the preset target task requirements. If it does, the final precise attitude control command is obtained.

[0154] It should be noted that the robot's state monitoring module acquires real-time motion state data. This data includes physical quantities such as joint angles, linear velocity, angular velocity, and acceleration. For example, in an industrial robotic arm scenario, a high-precision encoder collects the real-time angles of each joint (e.g., joint 1 angle is 35.2° ± 0.1°), while an inertial measurement unit (IMU) acquires the three-dimensional spatial velocity of the end effector (e.g., X-axis velocity 0.48 m / s) and attitude angles (e.g., pitch angle 2.3°). This data is updated at a sampling frequency of 100Hz, forming a dynamic state dataset.

[0155] The current motion state data is compared with preset task matching rules. These rules define key parameter thresholds for the target task, such as a position error threshold of ±0.05 meters and a velocity error threshold of ±0.1 meters per second for target trajectory tracking. If a deviation exceeds the threshold (e.g., the measured velocity of 0.48 meters per second is lower than the target velocity of 0.6 meters per second, or the deviation of 0.12 meters per second exceeds the allowable 0.1 meters per second), the built-in control algorithm is activated to generate preliminary adjustment instructions. Specifically, by establishing a robot dynamics model, a proportional-integral-derivative (PID) control algorithm is used to calculate the compensation amount: based on the velocity deviation of 0.12 meters per second, combined with the robot's mass and friction coefficient, the motor torque increment is calculated, generating a basic parameter set {torque increment: +6.4 N·m, duration: 0.3 seconds}.

[0156] After executing the basic parameter set instructions, dynamic adjustment data is collected in real time (e.g., the actual speed increases to 0.58 m / s after adjustment). Combined with preset constraints (e.g., maximum joint torque limit of 20 N·m, steering angular velocity limit of 10° / s), a quadratic programming (QP) optimization algorithm is used to optimize the parameters: The objective function min||v_actual - v_target|| is constructed. The constraint is |joint torque| ≤ 18 N·m (with a safety margin). For example, when a low road adhesion coefficient is detected, the theoretically calculated steering angular velocity of 15° / s is optimized to 8° / s, and the optimized parameters are output as follows: Steering angular velocity: 8° / s, Acceleration: 0.3 m / s². }

[0157] Determine whether the parameter optimization results meet the precision control standard (e.g., the speed stabilizes within ±0.05 m / s of the target value). If they do, determine the intermediate parameter set. For example, if the optimized speed stabilizes at 0.58 m / s, it meets the standard and is used as the intermediate parameter set for subsequent adjustments.

[0158] The robot's operation is controlled based on an intermediate parameter set, while the latest motion state is monitored in real time via a laser tracker or vision sensor (e.g., actual turning angle 4.6° vs. target 5.0°). If the matching degree is insufficient (deviation > 0.5°), a secondary calibration is initiated: an iterative learning control (ILC) algorithm is used, based on the historical error sequence [ The predicted compensation values ​​are calculated as [0.4°, 0.3°, 0.2°], generating the final control parameter set {steering angle compensation: +0.53°, torque compensation: +7.2 N·m}. A Kalman filter is introduced during the calibration process to eliminate sensor noise and ensure parameter reliability.

[0159] The final control parameter set is converted into PWM signals (e.g., 72% duty cycle) by the motion control card and sent to the servo motor. After execution, the status data is continuously monitored (500Hz sampling rate). When the position error is <0.01m and the attitude stabilization time is ≥1 second for 5 consecutive sampling cycles, the target task requirements are determined to be met, and a locked, precise attitude control command is output. For example, if the trajectory tracking error of the robotic arm end effector is stable within ±0.008m, the system records the final command parameters and stores them in the control register.

[0160] In summary, this process, through multi-stage parameter optimization and calibration, fully integrates the robot's motion state and task constraints, effectively improving the accuracy of posture control commands. This ensures that the robot can stably and efficiently complete target tasks in complex dynamic environments, providing strong support for the safety and reliability of robot operations.

[0161] Reference Figure 2 The second embodiment of the present invention provides a robot posture intelligent control system based on an ad hoc network communication architecture, comprising:

[0162] The data acquisition module is used to acquire multi-dimensional initial environmental data;

[0163] The data fusion module is used to perform multi-source data integration and key feature extraction operations based on the multi-dimensional environmental initial data to obtain a fused environmental feature set.

[0164] The panoramic view construction module is used to perform multi-angle visual data spatial distribution and change trend analysis based on the set of environmental features to obtain a panoramic information view of the dynamic environment.

[0165] The risk identification module is used to divide the panoramic information view of the dynamic environment into regions to obtain a set of divided view units, and to identify operational risk points based on the set of view units.

[0166] The preliminary strategy generation module is used to trigger risk response and generate preliminary attitude adjustment strategy signals if there are potential operational risk points, thus obtaining preliminary attitude adjustment strategy signals.

[0167] The precise instruction determination module is used to optimize the instruction parameters based on the preliminary attitude adjustment strategy signal to obtain the final precise attitude control instruction.

[0168] It should be noted that the robot posture intelligent control device based on self-organizing network communication architecture provided in this embodiment of the invention is used to execute all the process steps of the robot posture intelligent control method based on self-organizing network communication architecture in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0169] In summary, this invention provides a robot posture intelligent control method based on an ad hoc network communication architecture, comprising: acquiring multi-dimensional initial environmental data; performing multi-source data integration and key feature extraction operations based on the multi-dimensional initial environmental data to obtain a fused environmental feature set; performing multi-angle visual data spatial distribution and change trend analysis operations based on the environmental feature set to obtain a panoramic information view of the dynamic environment; dividing the panoramic information view of the dynamic environment into regions to obtain a set of divided view units, and identifying operational risk points based on the set of view units; if potential operational risk points exist, performing risk response triggering and preliminary posture adjustment strategy signal generation operations to obtain a preliminary posture adjustment strategy signal; and performing adjustment command parameter optimization operations based on the preliminary posture adjustment strategy signal to obtain the final precise posture control command. The method collects multi-dimensional environmental data through a sensor array, integrates multiple sources, constructs a panoramic view, identifies risks, and optimizes commands. This enables intelligent and precise control of robot posture in complex dynamic environments, effectively solving the problems of incomplete information from a single sensor and insufficient environmental perception that lead to low accuracy of posture adjustment commands. It improves the robot's adaptability to dynamic obstacles and terrain changes, reduces operational errors, and enhances work efficiency and operational safety.

[0170] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a robot posture intelligent control program based on an ad hoc network communication architecture. When the processor executes the computer program, it implements the steps in the various embodiments of the robot posture intelligent control method based on an ad hoc network communication architecture described above, for example... Figure 1 The step S101 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data acquisition module.

[0171] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0172] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0173] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0174] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0175] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0176] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0177] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A robot posture intelligent control method based on an ad hoc network communication architecture, characterized in that, include: Obtain multi-dimensional initial environmental data; Based on the initial multi-dimensional environmental data, multi-source data integration and key feature extraction operations are performed to obtain a fused environmental feature set. Based on the set of environmental features, multi-angle visual data spatial distribution and change trend analysis are performed to obtain a panoramic information view of the dynamic environment. The panoramic information view of the dynamic environment is divided into regions to obtain a set of divided view units, and operational risk points are identified based on the set of view units. If there are potential operational risks, risk response triggering and preliminary attitude adjustment strategy signal generation operations are performed to obtain preliminary attitude adjustment strategy signals. Based on the preliminary attitude adjustment strategy signal, the adjustment command parameters are optimized to obtain the final precise attitude control command. The step of performing multi-angle visual data spatial distribution and change trend analysis based on the environmental feature set to obtain a panoramic information view of the dynamic environment includes: performing spatial coordinate transformation on the environmental feature set to obtain point data corresponding to the spatial distribution, resulting in pre-classification regional feature data; classifying the pre-classification regional feature data according to different regions to obtain a classified regional feature set; performing change trend tracking on the classified regional feature data to obtain dynamic environmental change data over time, and determining the completeness of the dynamic environmental change data; if the completeness of the dynamic environmental change data is lower than a preset completeness threshold, then performing interpolation filling on the dynamic environmental change data to obtain an adjusted dynamic environmental data set; and constructing an all-round visual network model based on the adjusted dynamic environmental data set to analyze the spatial distribution and change trend of multi-angle visual data, thereby obtaining a panoramic information view of the dynamic environment. Based on the adjusted dynamic environment dataset, a comprehensive visual network model is constructed by fusing feature data from various regions in the following specific ways: For the regional feature data collected by each node under the self-organizing network communication architecture, an adaptive weighted fusion algorithm is used to assign dynamic weights to different regional data according to the credibility of the data source; an attention mechanism module is introduced to focus on the feature data of key areas around the robot and extract features from areas with dense dynamic targets; finally, data from different regions are integrated, and spatial features of each region are extracted by combining a multi-layer convolutional neural network. At the same time, temporal correlation information is processed by a recurrent neural network, and feature fusion is performed using preset network structure parameters to construct a comprehensive visual network model. The process of dividing the panoramic information view of the dynamic environment into regions to obtain a set of divided view units, and identifying operational risk points based on the view unit set, includes: initially dividing the terrain changes and obstacle areas in the panoramic information view to obtain at least one independent terrain unit and obstacle unit, resulting in a set of divided view units; extracting detailed features from the terrain units and obstacle units based on the view unit set, and classifying the extracted detailed features in real time to determine the specific category and attribute status of each unit; matching the specific category and attribute status with a pre-established threat database, and marking the attribute status as a high-risk unit if it matches the potential threat features in the database, resulting in a set of marked risk units; dynamically adjusting the operational safety area based on the set of marked risk units, and updating the panoramic information view of the dynamic environment in real time using a preset view update mechanism to determine the distribution of potential operational risk points.

2. The intelligent robot posture control method based on ad hoc network communication architecture according to claim 1, characterized in that, The acquisition of multi-dimensional environmental initial data includes: The sensor array acquires multi-angle data from the dynamic environment and preprocesses it to obtain an initial multi-dimensional data set, which includes visual information, distance information, and terrain feature information. The visual information in the initial multi-dimensional dataset is subjected to denoising and edge detection processing to extract clear visual feature data; Based on the distance and terrain feature information in the initial multi-dimensional dataset, the spatial distribution in the environment is calculated, and the spatial structure data of the environment is determined. If the matching degree between the spatial structure data and the visual feature data is lower than a preset matching degree threshold, then the two are calibrated to obtain a calibrated environmental information matrix. The calibrated environmental information matrix is ​​interpolated, filled, and standardized to obtain a standardized environmental perception dataset. Based on the environmental perception dataset, the data is grouped according to feature dimensions, and the distribution characteristics of each group are determined to obtain multi-dimensional initial environmental data; among which, visual information is grouped according to resolution, distance information is grouped according to measurement range, and terrain feature information is grouped according to type.

3. The intelligent robot posture control method based on ad hoc network communication architecture according to claim 1, characterized in that, Based on the initial multi-dimensional environmental data, multi-source data integration and key feature extraction operations are performed to obtain a fused environmental feature set, including: The initial data of the multi-dimensional environment is cleaned, and the cleaned data is time-stamped and missing values ​​are filled to obtain the complete dataset after filling. Based on the completed dataset after filling in the gaps, the data in each dimension are filtered and processed to obtain significant feature data related to environmental perception. The salient feature data is classified according to spatial distribution or functional attributes to obtain a grouped feature data set; among which, distance data is divided into near distance group and medium distance group, and visual features are divided into static object group and dynamic object group. The data from different groups are integrated using the grouped feature data set. If the matching degree of the integrated data is lower than a preset threshold, consistency adjustment is performed to obtain the fused environmental feature set.

4. The intelligent robot posture control method based on ad hoc network communication architecture according to claim 1, characterized in that, If potential operational risks exist, a risk response trigger and preliminary attitude adjustment strategy signal generation operation will be performed to obtain a preliminary attitude adjustment strategy signal, including: If there are potential operational risks, the panoramic information view of the dynamic environment is divided into regions to obtain at least one risk region unit, resulting in a set of divided risk regions. The spatial location and risk category of the risk area units in the risk area set are captured in detail to determine the specific location coordinates and category labels of the units; If the category label matches a high-risk feature in a pre-established threat assessment database, a corresponding real-time response instruction is generated, and the response signal triggered by the real-time response instruction is obtained. Based on the response signal, the target operating area is dynamically planned, and a preliminary adjustment signal is generated in combination with the preset attitude adjustment logic; Determine whether the preliminary adjustment signal meets the preset safety range. If it does, then determine the preliminary adjustment signal as the preliminary attitude adjustment strategy signal.

5. The intelligent robot posture control method based on ad hoc network communication architecture according to claim 1, characterized in that, The step of optimizing the adjustment command parameters based on the preliminary attitude adjustment strategy signal to obtain the final precise attitude control command includes: Based on the preliminary attitude adjustment strategy signal, the robot's current motion state data is obtained; The current motion state data is compared with the pre-established task matching rules. If the deviation between the current motion state data and the target task exceeds the preset deviation threshold, a preliminary adjustment instruction is generated to obtain the basic parameter set for instruction adjustment. The robot is dynamically adjusted based on the basic parameter set to obtain dynamic adjustment data of the robot, and the parameters are optimized in combination with preset constraints to obtain the parameter optimization results. Determine whether the parameter optimization result meets the preset precision control standard. If it does, obtain an intermediate parameter set suitable for the current scenario. The robot is dynamically adjusted based on the intermediate parameter set, and the robot's status is monitored in real time to obtain the latest motion status information. If the motion state information does not match the preset target task well enough, a second calibration is performed to obtain the final control parameter set; The attitude is adjusted according to the final control parameter set, and it is determined whether the adjusted state meets the preset target task requirements. If it does, the final precise attitude control command is obtained.

6. The intelligent robot posture control method based on ad hoc network communication architecture according to claim 2, characterized in that, The sensor array includes a visual sensor, a distance sensor, and a terrain sensor. The visual sensor is used to collect visual information, the distance sensor is used to collect distance information, and the terrain sensor is used to collect terrain feature information.

7. A robot posture intelligent control system based on a self-organizing network communication architecture, characterized in that, A robot posture intelligent control method based on an ad hoc network communication architecture as described in any one of claims 1 to 6 includes: The data acquisition module is used to acquire multi-dimensional initial environmental data, the robot's current motion state, and target task constraints; The data fusion module is used to perform multi-source data integration and key feature extraction operations based on the multi-dimensional environmental initial data to obtain a fused environmental feature set. The panoramic view construction module is used to perform multi-angle visual data spatial distribution and change trend analysis based on the set of environmental features to obtain a panoramic information view of the dynamic environment. The risk identification module is used to divide the panoramic information view of the dynamic environment into regions to obtain a set of divided view units, and to identify operational risk points based on the set of view units. The preliminary strategy generation module is used to trigger risk response and generate preliminary attitude adjustment strategy signals if there are potential operational risk points, thus obtaining preliminary attitude adjustment strategy signals. The precise instruction determination module is used to perform adjustment instruction parameter optimization operations based on the preliminary attitude adjustment strategy signal, the current motion state and the target task constraints, to obtain the final precise attitude control instruction.

8. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a robot posture intelligent control method based on an ad hoc network communication architecture as described in any one of claims 1 to 6.

Images