An agv cross-warehouse navigation seamless switching method and system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 南京欧米麦克机器人科技有限公司
- Filing Date
- 2026-05-08
- Publication Date
- 2026-06-02
Smart Images
Figure CN122130101A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of automated guided vehicles, specifically to a method and system for seamless switching of AGV navigation across workshops. Background Technology
[0002] In the field of industrial automation, navigation technology for Automated Guided Vehicles (AGVs) has developed rapidly, playing a vital role in improving production efficiency and reducing labor costs. With the expansion of industrial production scale and the increasing complexity of production environments, AGVs need to achieve seamless navigation switching between different workshops and even different environments to adapt to diverse production demands.
[0003] In traditional AGV navigation technology, a single navigation algorithm is often used to solve the navigation problem. For example, traditional 3D laser SLAM algorithms mainly rely on point cloud matching to determine the AGV's pose, which can build relatively accurate maps and achieve navigation in indoor environments; while pure RTK algorithms obtain the AGV's absolute coordinates by receiving satellite signals, making them suitable for open outdoor environments. In addition, some switching algorithms use "hard switching" modes, directly switching between different navigation modes.
[0004] However, these traditional navigation technologies and switching algorithms have significant drawbacks. Traditional 3D laser SLAM algorithms are prone to pose drift due to the accumulation of point cloud matching errors in outdoor, feature-sparse scenes; while pure RTK algorithms fail indoors or in occluded areas due to satellite signal loss. Furthermore, existing "hard switching" modes are prone to causing abrupt position changes in AGVs during transition areas, affecting the stability and accuracy of navigation. Summary of the Invention
[0005] To address the issues of positioning jumps and control instability during hard switching between single or multiple navigation algorithms, as well as the insufficient robustness of algorithms in complex scenarios, this application provides a seamless switching method and system for AGV navigation across workshops.
[0006] Firstly, this application provides a method for seamless switching of AGV navigation across workshops, including: Collect environmental point cloud data, absolute coordinates, and inertial data; Based on the environmental point cloud data, the current environment type is determined and a navigation mode switching command is output to guide the switching of navigation modes; the output process of the navigation mode switching command includes: Calculate the environmental feature entropy value based on the collected environmental point cloud data; Based on the environmental feature entropy value, it is determined whether a first triggering condition for outputting a navigation mode switching command is met. If the first triggering condition is met, a navigation mode switching command is output. The first triggering condition includes: when the environmental feature entropy value is less than a first feature entropy value threshold and lasts for a preset time, a SLAM to RTK navigation mode switching is triggered; when the environmental feature entropy value is greater than or equal to a second feature entropy value threshold and lasts for a preset time, and the indoor map matching confidence reaches a preset confidence level, an RTK to SLAM navigation mode switching is triggered; the map matching confidence is the matching confidence between the navigation position corresponding to the current mode and the indoor map position. During the switching process, positioning data from different navigation modes are merged to eliminate positioning jumps; AGV motion control commands are generated based on the fused positioning data.
[0007] By adopting the above scheme, the collection of environmental point cloud data provides a foundation for subsequent navigation mode switching and positioning data fusion; by judging the environment type based on the environmental point cloud data and outputting switching instructions, the switching of different navigation modes can be realized, breaking through the environmental limitations of single navigation technology; by fusing positioning data from different navigation modes, positioning jumps can be eliminated, making positioning more accurate and improving the robustness of the algorithm in complex scenarios; motion control instructions are generated based on the fused positioning data to ensure the normal operation of the AGV.
[0008] Preferably, this includes: acquiring environmental point cloud data, absolute coordinates, and inertial data through 3D lidar, RTK-GNSS devices, and IMU sensors.
[0009] By adopting the above scheme, 3D lidar, RTK-GNSS devices and IMU sensors are used to accurately collect environmental point cloud data, absolute coordinates and inertial data, providing a reliable data foundation for subsequent operations.
[0010] Preferably, the process of eliminating the positioning jump includes: When the RTK signal is interrupted, the absolute coordinates collected by the RTK-GNSS device and the inertial data collected by the IMU sensor are fused by extended Kalman filtering, and the fusion weight is dynamically adjusted according to the PDOP value of the RTK positioning accuracy of the RTK-GNSS device. When the PDOP value is less than or equal to the PDOP threshold, the weight of the absolute coordinates is greater than 80%; when the PDOP value is greater than the PDOP threshold, the weights of the absolute coordinates and the inertial data are each half. The process of eliminating the positioning jump also includes: An adaptive Kalman filter algorithm is used to fuse positioning data from SLAM navigation mode and RTK navigation mode to obtain positioning results. The fusion steps include: real-time calculation of positioning errors of the two navigation modes; real-time updating of process noise covariance and observation noise covariance based on the positioning errors of the two navigation modes; and fusing positioning data of the two modes based on the covariance matrix to output positioning results.
[0011] By adopting the above scheme, when the RTK signal is interrupted, the extended Kalman filter can be used to fuse absolute coordinates and inertial data, and the weights can be dynamically adjusted according to the PDOP value to ensure positioning stability. The adaptive Kalman filter algorithm is used to fuse positioning data of different navigation modes and update the noise covariance matrix in real time, effectively eliminating positioning jumps and improving positioning accuracy and navigation reliability.
[0012] Preferably, the AGV motion control command generation step includes: interpolating the linear velocity and angular velocity commands of the AGV at the moment of switching, and limiting the corresponding velocity change rate to be less than a preset velocity change rate.
[0013] By adopting the above scheme, the linear velocity and angular velocity commands of the AGV are interpolated and the rate of velocity change is limited at the moment of navigation mode switching, which can prevent the AGV from accelerating or turning sharply and ensure the smoothness of the AGV's driving.
[0014] Preferably, the output process of the navigation mode switching command further includes: The steps for calculating the entropy value of environmental features include: dividing the collected environmental point cloud data into local neighborhoods, extracting the geometric feature types of each neighborhood, calculating the probability of each type of feature, and calculating the information entropy value; The sliding window adaptive threshold method is adopted to calculate the mean and standard deviation of the historical entropy value queue within a sliding window of a preset length, and to update the first feature entropy value threshold and the second feature entropy value threshold based on the calculated mean and standard deviation. The rate of change of environmental feature entropy is calculated and the preset duration is updated according to the rate of change of environmental feature entropy, so that when the rate of change of environmental feature entropy increases, the preset duration is shortened, and vice versa when the rate of change of environmental feature entropy decreases, the preset duration is extended.
[0015] By adopting the above scheme, the collected environmental point cloud data is divided into local neighborhoods, geometric feature types are extracted, the probabilities of various features are statistically analyzed, and information entropy values are calculated. This accurately quantifies the environmental complexity, thereby more precisely calculating the environmental feature entropy values. Furthermore, by adaptively adjusting the threshold and preset duration in conjunction with environmental changes, the accuracy and timeliness of navigation mode switching command output are improved, and unnecessary switching jitter is reduced.
[0016] Preferably, the output process of the navigation mode switching command further includes: By combining the collected environmental point cloud data, absolute coordinates, inertial data, and relevant signal data involved in the acquisition process, the noise index, RTK signal quality, and laser SLAM covariance are calculated, and the AGV motion status is acquired simultaneously. The comprehensive environment adaptability is defined as the degree of support of the current environment for SLAM navigation modal positioning. The environmental feature entropy value, noise index value, RTK signal quality, and laser SLAM covariance are used as the influencing factors of the environment adaptability. The comprehensive environment adaptability is calculated by a weighted product model algorithm. The overall environmental adaptability is dynamically corrected by utilizing the AGV's motion state, so that the support for SLAM navigation modal positioning is reduced when the AGV's motion state is in a preset high-speed state or a sharp turn state. To obtain the final overall environmental adaptability; Based on the final comprehensive environment adaptability, determine whether the second triggering condition for outputting the navigation mode switching command is met; the second triggering condition is: when the final comprehensive environment adaptability is less than the first environment adaptability threshold and continues for a preset time, trigger the SLAM to RTK navigation mode switching; when the comprehensive environment adaptability is greater than or equal to the second environment adaptability threshold and continues for a preset time, and the indoor map matching confidence reaches a preset confidence level, trigger the RTK to SLAM navigation mode switching.
[0017] By adopting the above scheme, and combining information from multiple aspects such as environmental characteristic entropy, noise index, RTK signal quality, laser SLAM covariance, and AGV motion status, the timing of navigation mode switching can be determined more accurately, thereby improving the accuracy and stability of navigation mode switching and avoiding frequent and unnecessary switching.
[0018] Preferably, the output process of the navigation mode switching command further includes: Input environmental point cloud data into a pre-built scene recognition model, and output the current scene category and the confidence level of each category; Select scene categories with a category confidence level greater than the category confidence level threshold, and calculate the comprehensive environment adaptability by matching the combination form and weight coefficient of the environment adaptability influencing factors according to the selected scene categories.
[0019] By adopting the above scheme, the combination and weight of environmental adaptability influencing factors can be flexibly adjusted according to the scenario category, the comprehensive environmental adaptability can be calculated more accurately, and the accuracy of navigation mode switching command output and the environmental adaptability of navigation can be improved.
[0020] Preferably, the process of eliminating the positioning jump further includes: In the process of fusing positioning data from SLAM navigation mode and RTK navigation mode using an adaptive Kalman filter algorithm, the final environment adaptability is mapped to the fusion weights to obtain the corresponding weights of positioning data under different modes. Based on the corresponding weights of positioning data under different modes, and combined with the positioning errors of the two navigation modes, the process noise covariance and observation noise covariance are updated in real time. Furthermore, a fading factor calculated based on the covariance matching of the residual sequence between prediction and observation is introduced to update the process noise covariance in real time.
[0021] By adopting the above scheme, when fusing positioning data using the adaptive Kalman filter algorithm, the weights are determined by combining the final environmental adaptability, the noise covariance matrix is updated, and a fading factor is introduced to amplify the noise covariance during the process. This can further improve the accuracy and stability of positioning data fusion, better eliminate positioning jumps, and improve the reliability of AGV navigation.
[0022] Secondly, this application provides an AGV cross-workshop navigation seamless switching system, comprising: The perception module is used to collect environmental point cloud data, absolute coordinates, and inertial data; The decision module is used to determine the current environment type based on the environmental point cloud data and output a navigation mode switching command to guide the switching of navigation modes. The output process of the navigation mode switching command includes: calculating the environmental feature entropy value based on the collected environmental point cloud data; determining whether the first triggering condition for outputting the navigation mode switching command is met based on the environmental feature entropy value; and outputting the navigation mode switching command when the first triggering condition is met. The first triggering condition includes: triggering the SLAM to RTK navigation mode switching when the environmental feature entropy value is less than a first feature entropy value threshold and lasts for a preset time; and triggering the RTK to SLAM navigation mode switching when the environmental feature entropy value is greater than or equal to a second feature entropy value threshold and lasts for a preset time, and the indoor map matching confidence reaches a preset confidence level. The map matching confidence level is the matching confidence between the navigation position corresponding to the current mode and the indoor map position. The fusion module is used to merge positioning data from different navigation modes during the switching process to eliminate positioning jumps; The control module is used to generate AGV motion control commands based on the fused positioning data.
[0023] By adopting the above scheme, the current environment type is determined based on the collected multi-type data and navigation mode switching instructions are given. Positioning data from different navigation modes are integrated to eliminate positioning jumps. AGV motion control instructions are generated based on the integrated positioning data, breaking through the environmental limitations of single navigation technology and realizing seamless switching of AGV navigation across workshops.
[0024] In summary, this application has the following beneficial effects: 1. By designing dynamic mode switching and data fusion, the system determines the environment type based on environmental point cloud data and outputs switching commands to achieve switching between different navigation modes. An adaptive Kalman filter algorithm is used to fuse positioning data from different navigation modes, enabling continuous output of positioning data and control commands. This solves the problem of complementary environmental blind zones between traditional 3D laser SLAM algorithms and pure RTK algorithms, overcomes the environmental limitations of single navigation technologies, and improves positioning reliability in complex environments. 2. By comprehensively considering environmental characteristic entropy values, noise index values, RTK signal quality, laser SLAM covariance, and AGV motion state, the system more accurately judges the environment's support for SLAM navigation mode positioning, thereby optimizing navigation mode switching decisions and improving the accuracy and reliability of navigation mode switching. When fusing positioning data under different navigation modes, the system uses the final environmental adaptability to determine the fusion weights, combines them with the real-time updated noise covariance matrix, and introduces a fading factor to amplify the noise covariance during the process, further improving the rationality and accuracy of positioning data fusion, effectively reducing positioning jumps during switching, and enhancing the stability of navigation positioning. Attached Figure Description
[0025] Figure 1 This is a flowchart of the AGV cross-workshop navigation seamless switching method described in a specific embodiment; Figure 2 This is a schematic diagram of the AGV cross-workshop navigation seamless switching system described in a specific embodiment. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0027] like Figure 1 As shown in the figure, this application discloses a method for seamless switching of AGV navigation across workshops, including: data acquisition, environment judgment, data fusion, and instruction generation steps; through the collaborative work of multiple steps, the advantages of different navigation modes can be comprehensively utilized to adapt to different environments, achieving the effect of overcoming the environmental limitations of single navigation technology and realizing seamless switching of AGV navigation across workshops. The following is a further detailed description of this application.
[0028] S1. Collect environmental point cloud data, absolute coordinates, and inertial data.
[0029] Specifically, for the operating AGV, environmental point cloud data, absolute coordinates, and inertial data are collected by installed 3D LiDAR, RTK-GNSS devices, and IMU sensors, respectively. Correspondingly, the acquisition process involves relevant signal data, such as RTK signal data and LiDAR signal data.
[0030] Among them, 3D LiDAR can be either mechanical or solid-state. Mechanical LiDAR uses rotating scanning to acquire point cloud data of the surrounding environment, while solid-state LiDAR acquires point cloud data through electronic scanning. RTK-GNSS devices receive satellite signals to obtain absolute coordinates, and their working principle is to improve positioning accuracy using carrier phase differential technology. IMU sensors measure inertial data such as the angular velocity and acceleration of the AGV. Common IMU sensors include microelectromechanical inertial measurement units based on MEMS technology.
[0031] S2. Determine the current environment type and output a navigation mode switching command to guide the switching of navigation modes.
[0032] Specifically, the output process of the navigation mode switching command is quite complex, including the following steps: First, the environmental feature entropy value is calculated using environmental point cloud data.
[0033] The "environmental characteristic entropy" index is defined to quantify environmental complexity, and the calculation formula is as follows: In the formula, This represents the environmental characteristic entropy value at time t. This represents the proportion of the i-th type of feature points, such as the proportion of corner points and planes.
[0034] Secondly, based on the environmental feature entropy value, it is determined whether the first triggering condition for outputting the navigation mode switching command is met. If the first triggering condition is met, the navigation mode switching command is output.
[0035] Among them, when hour, If the value is small, it indicates sparse indoor features, and is therefore identified as an RTK navigation mode; when... hour, A larger value indicates richer indoor features, thus classifying it as a SLAM navigation mode; therefore, the environmental feature entropy value can be used as a basis for further analysis. This corresponds to the switching of navigation modes, but to avoid frequent switching, a switching lag condition is set, thus establishing a first trigger condition. The first trigger condition includes: when the environmental feature entropy value... And continue for a preset time T (e.g.) At 200ms, a switch from SLAM to RTK navigation mode is triggered; when the environmental feature entropy value... Greater than the second feature entropy value If the indoor map matching confidence reaches a preset level (e.g., 90%), the RTK to SLAM navigation mode switch is triggered.
[0036] In this context, SLAM navigation mode refers to positioning and guiding AGV operation using 3D LiDAR. This embodiment employs an indoor enhanced 3D SLAM navigation method, specifically a multi-feature constrained point cloud matching approach using a joint corner and planar feature matching strategy. This includes: preprocessing the LiDAR point cloud data (denoising and downsampling); extracting planar features (such as walls and ground) using the RANSAC algorithm; extracting corner features (such as column corners) using the Harris3D algorithm; constructing feature descriptors and performing coarse matching to calculate the initial pose estimate; and using the initial pose as input, performing ICP fine matching to output the indoor SLAM pose. The specific strategy execution unit, including the point cloud preprocessing unit, feature extraction unit, and pose calculation unit, is integrated into the AGV main control chip. In the indoor enhanced 3D SLAM navigation modal method, point cloud data of moving targets can also be identified and removed through temporal point cloud comparison. The removal criterion is that the displacement velocity of the moving target is less than a preset displacement velocity threshold, thereby avoiding interference from dynamic targets (such as pedestrians in the workshop) in map construction.
[0037] The map matching confidence score is the matching confidence score between the navigation location and the indoor map location in the current mode, that is, the probability that the current navigation location is within the range of the indoor map location.
[0038] RTK navigation mode refers to positioning and guiding AGV operation using an RTK-GNSS device. To address the issue of short-term RTK signal obstruction, an outdoor RTK-IMU tightly coupled navigation method is selected, and a tightly coupled model based on Extended Kalman Filter (EKF) can also be designed. The specific steps are as follows: Based on the pseudorange and carrier phase observations output by the RTK-GNSS device, and the angular velocity and acceleration output by the IMU; establish state equations (including position, velocity, attitude, and IMU bias) and observation equations (using RTK observations as the measurement); through EKF prediction and update steps, fuse the two types of data to output the outdoor positioning result, and accordingly execute the RTK data parsing unit, IMU data processing unit, and EKF fusion unit to achieve real-time data interaction and calculation, all integrated into the AGV main control chip.
[0039] S3. During the switching process, the positioning data of different navigation modes are merged to eliminate positioning jumps.
[0040] Specifically, when When AGV operation is in the RTK navigation mode, it is dominated by RTK navigation. hour, If the value is too large, the AGV will operate primarily in SLAM navigation mode; when... Alternatively, if the first trigger condition is met, the corresponding modality switch will be executed. During the transition and switching process, if the positioning is directly based on the data before and after the switch, positioning jumps will occur. Therefore, a multi-data fusion approach is chosen to eliminate positioning jumps during the transition and switching process. The elimination process includes: First, check if the RTK signal is interrupted. When the RTK signal is interrupted, fuse the absolute coordinates collected by the RTK-GNSS device and the inertial data collected by the IMU sensor through extended Kalman filtering, and dynamically adjust the fusion weight according to the RTK positioning accuracy PDOP value of the RTK-GNSS device. When fusing the two types of data, the specific weight settings are as follows: when the PDOP value is less than or equal to the PDOP threshold (e.g., 3, indicating good signal), the weight of the absolute coordinates is greater than 80%; when the PDOP value is greater than the PDOP threshold (e.g., 3, indicating moderate signal), the weights of the absolute coordinates and the inertial data are each 50%.
[0041] Secondly, an adaptive Kalman filter algorithm is used to fuse positioning data from SLAM navigation mode and RTK navigation mode to obtain the positioning result. The specific fusion steps include: The positioning error for two navigation modes is calculated in real time. This includes the acquired laser SLAM pose. and its covariance RTK pose obtained and its covariance During the measurement update phase of AKF, the observation vector is constructed as follows: And based on the observation matrix The state is mapped to the observation, and the difference is calculated to obtain the defined error, including SLAM positioning error and RTK differential accuracy. Based on the positioning errors of the two navigation modes, the process noise covariance is updated in real time. and observation noise covariance It also fuses positioning data from two modes based on the covariance matrix and outputs positioning results.
[0042] S4. Generate AGV motion control instructions based on the fused positioning data.
[0043] Specifically, the motion control command generation step includes: interpolating the linear and angular velocity commands of the AGV at the moment of switching, limiting the rate of speed change to less than a preset rate of speed change. Interpolation can be performed using methods such as cubic spline interpolation to smooth the speed commands, preventing the AGV from accelerating or turning sharply and ensuring smooth operation.
[0044] Through multi-step collaborative work, comprehensively utilizing data from different sensors and multiple algorithms, the environmental limitations of single navigation technologies have been overcome. It can accurately determine and switch navigation modes in different environments, and eliminate positioning jumps through data fusion, ensuring the stability and accuracy of AGV navigation. Simultaneously, the processing of motion control commands ensures the smoothness of AGV movement, improving the practicality and reliability of the entire navigation system, representing a significant improvement and enhancement compared to traditional navigation technologies.
[0045] A specific embodiment, differing from the above embodiments, involves more accurately calculating the environmental feature entropy value and adaptively adjusting the threshold and preset duration according to environmental changes, thereby improving the accuracy and timeliness of navigation mode switching command output and reducing unnecessary switching jitter; the method further includes: First, a continuous step-by-step method is used to calculate the "environmental feature entropy". Considering that laser SLAM systems typically operate with tens of thousands of point clouds per frame, extracting feature values for each point would be computationally expensive. Therefore, the feature values of each point are combined and treated as a "soft category" using fuzzy membership, and then the entropy is calculated. The calculation steps include: The collected point cloud data of the environment is divided into local neighborhoods; for example, for each point... Take the neighborhood point set within radius r, calculate the covariance matrix; solve for the eigenvalues and normalize them, and use the combination of normalized eigenvalues to define the feature type, such as line feature, surface feature, sphere feature, and others; Extract the geometric feature type of each neighborhood and calculate the probability of each feature type; that is, for all valid points in the current frame, calculate the frequency of occurrence of the four feature types mentioned above to obtain the probability distribution. For example: In the formula, These represent the frequency of occurrence of the four feature types, respectively. Calculate the entropy value, such as: ;in, Avoid log(0) for extremely small positive numbers.
[0046] Secondly, a dynamic environmental feature entropy threshold design is implemented. Considering that traditional fixed thresholds are prone to boundary oscillations when the environment changes slowly, a sliding window adaptive threshold method is adopted, combined with sliding statistics of historical entropy values, to achieve dynamic adjustment of the environmental feature entropy threshold. The specific steps are as follows: Calculate the historical entropy queue within a sliding window of a preset length L (e.g., 30 frames). mean and standard deviation The first and second feature entropy thresholds are updated based on the calculated mean and standard deviation. The feature entropy threshold is defined as follows: , Where k is the sensitivity coefficient, which is usually taken as 2-3.
[0047] Additionally, introducing entropy filtering can reduce frequent switching caused by noise and improve real-time response. After calculating the environmental feature entropy threshold, an exponential smoothing filter is added, with the following formula: In the formula, This is a smoothing factor.
[0048] Finally, the rate of change of environmental feature entropy is calculated, and the preset duration is updated based on the rate of change of environmental feature entropy. The formula is as follows: In the formula, The sensitivity coefficient is used to shorten the preset duration when the rate of change of the environmental characteristic entropy increases, and conversely, to extend the preset duration when the rate of change of the environmental characteristic entropy decreases.
[0049] A specific embodiment, differing from the above embodiments, comprehensively considers multiple factors such as noise index values, RTK signal quality, laser SLAM covariance, and AGV motion state to determine the overall environmental adaptability, accurately judging the timing of navigation mode switching, reducing erroneous switching caused by inaccurate judgment of a single factor, and improving the accuracy and reliability of navigation mode switching. The method further includes: the output process of the navigation mode switching command further includes: First, by combining the collected environmental point cloud data, absolute coordinates, inertial data, and relevant signal data involved in the acquisition process, the noise index, RTK signal quality, and laser SLAM covariance are calculated, and the AGV motion status is acquired simultaneously.
[0050] Among them, the noise index value refers to the noise quantization index of lidar. Or RTK environmental noise quantification index The noise quantification metrics for lidar include: local roughness of the lidar point cloud (calculating the root mean square error of the distance from each point in the point cloud to the fitted plane in its neighborhood), echo intensity signal-to-noise ratio (analyzing the intensity consistency of points at the same location across consecutive frames based on relevant signal data involved in lidar operation, or calculating the ratio of effective echo to noise echo), and multipath reflection detection rate (detecting anomalies caused by multipath reflection by analyzing the spatiotemporal consistency of the point cloud, defining the proportion of anomalies as a noise metric), etc. The environmental noise quantification metrics for RTK include: carrier-to-noise ratio (analyzing and calculating the ratio of signal power to noise power density based on relevant signal data involved in RTK-GNSS device operation), and multipath metrics (residual analysis of pseudorange and carrier phase observations), etc. This embodiment uses lidar noise quantification metrics... For example.
[0051] The RTK signal quality can be obtained by weighted comprehensive calculation based on indicators such as positioning status (fixed solution, floating-point solution, single-point solution, or no solution), PDOP value, and number of satellites, as shown in the formula: In the formula, Values are taken only for a fixed solution (or a floating-point solution under specific conditions); where, laser SLAM covariance It can be calculated based on laser SLAM positioning data; where motion states include linear velocity v and angular velocity w.
[0052] Then, the overall environment fit A is defined as the support of the current environment for SLAM navigation modal positioning (the value ranges from 0 to 1, with higher values indicating better suitability for laser SLAM). The environmental feature entropy, noise index, RTK signal quality, and laser SLAM covariance are used as influencing factors for environment fit, expressed as follows: The weighted product model algorithm is used for calculation, and the formula is as follows: In the formula, , This represents the measured maximum entropy value. , This is the calibrated maximum noise value; , The degradation threshold, such as 0.05. ; The index weights reflect the importance of each influencing factor.
[0053] Then, the overall environment adaptability is dynamically corrected using the AGV's motion state. This is done so that when the AGV's motion state is a preset high-speed state or a sharp turn state, the support for SLAM navigation modal positioning is reduced, thus obtaining the final overall environment adaptability. The specific formula is: ;in, This is an empirical coefficient.
[0054] Finally, based on the final overall environment adaptability, it is determined whether the second triggering condition for outputting the navigation mode switching command is met.
[0055] Specifically, compared to the first triggering condition, this embodiment further considers multiple factors affecting positioning, using the final comprehensive adaptability as the switching standard. To avoid false interference, a second triggering condition is set: when the final comprehensive environmental adaptability is less than the first environmental adaptability threshold and continues for a preset time, the SLAM to RTK navigation mode switching is triggered; when the comprehensive environmental adaptability is greater than or equal to the second environmental adaptability threshold and continues for a preset time, and the indoor map matching confidence reaches a preset confidence level, the RTK to SLAM navigation mode switching is triggered.
[0056] Furthermore, considering that the contributions of various influencing factors to positioning reliability differ under different environments, a globally fixed fusion function may fail in certain scenarios. Therefore, neural network technology is used to initially identify the current AGV's scenario classification, and the contributions of different influencing factors to positioning reliability are adaptively adjusted for different scenario types. Thus, the output process of the navigation mode switching command also includes: Environmental point cloud data is input into a pre-built scene recognition model, which outputs the current scene category and the confidence score for each category. The pre-built scene recognition model uses a neural network model. The model's input includes collected environmental point cloud data, absolute coordinates, and inertial data. The model's output is the scene type and its confidence score. For example, scene types include: open space, urban canyon, tunnel / corridor, and indoor structure.
[0057] Select scene categories with a category confidence level greater than a category confidence threshold (e.g., 90%), and calculate the comprehensive environment fit score by matching the combination form and weight coefficients of environmental fit influencing factors to the selected scene categories. Different scene categories each have their own matching combination form and weight coefficients of environmental fit influencing factors. For example, in open, undeveloped areas, only... , All are 0, meaning the higher the RTK quality, the lower the A. In the urban canyon area, all fit influence factors exist, with weights of 0.5, 1.5, 1.2, and 0.8 respectively.
[0058] A specific embodiment, differing from the above embodiments, maps the final environment adaptability to the fusion weights. This allows for the reasonable allocation of weights for different navigation modal positioning data based on the environment adaptability. Combined with a real-time updated noise covariance matrix and the introduction of a fading factor to amplify the noise covariance during the process, it enables more accurate fusion of positioning data from SLAM and RTK navigation modes, further reducing positioning jumps and improving positioning stability and accuracy. The method also includes: the elimination process of positioning jumps further includes: First, in the process of fusing positioning data from SLAM navigation mode and RTK navigation mode using an adaptive Kalman filter algorithm, the final environmental adaptability is mapped to the fusion weights to obtain the corresponding weights of positioning data under different modes.
[0059] Specifically, measure the noise covariance matrix. Designed as diagonally divided blocks: ;in, ; ;in, and Final environmental adaptability The decision is made using the following formula: , Then, based on the corresponding weights of the positioning data under different modes, and combined with the positioning errors of the two navigation modes, the process noise covariance and observation noise covariance are updated in real time.
[0060] Furthermore, a fading factor based on covariance matching of the residual sequences between predictions and observations is introduced to update the process noise covariance in real time, as shown in the formula: In the formula, For process noise covariance; , , In the formula, For state covariance, This represents the theoretical variance.
[0061] Furthermore, considering that when multiple final integrated environment adaptability scores are calculated according to the above embodiment, there are multiple scenario type confidence scores greater than the preset type confidence threshold; therefore, multiple final integrated environment adaptability scores are obtained, and multiple switching command decisions are acquired. The final integrated environment adaptability scores with the most consistent output navigation mode switching command results are selected. Adaptive Kalman filtering algorithms are used to fuse the positioning data under SLAM navigation mode and RTK navigation mode respectively, obtaining multiple positioning results. Finally, a weighted voting method or DS evidence theory and other decision fusion methods are used to perform decision fusion on the multiple positioning results, ultimately obtaining the fused positioning data.
[0062] like Figure 2 As shown in the illustration, this application also discloses a seamless switching system for AGV cross-workshop navigation, specifically including: The sensing module 100 is used to collect environmental point cloud data, absolute coordinates, and inertial data; Decision module 200 is used to determine the current environment type based on the environmental point cloud data and output a navigation mode switching command to guide the switching of navigation modes; The fusion module 300 is used to fuse positioning data from different navigation modes during the switching process to eliminate positioning jumps; The control module 400 is used to generate AGV motion control commands based on the fused positioning data.
[0063] In one specific embodiment, the decision module 200 in the system is further configured to use a sliding window adaptive threshold method to calculate the mean and standard deviation of the historical entropy value queue within a sliding window of a preset length, and update the first feature entropy value threshold and the second feature entropy value threshold according to the calculated mean and standard deviation; calculate the rate of change of environmental feature entropy value and update the preset duration according to the rate of change of environmental feature entropy value, so that when the rate of change of environmental feature entropy value increases, the preset duration is shortened, and conversely, when the rate of change of environmental feature entropy value decreases, the preset duration is extended.
[0064] In a specific embodiment, the decision module 200 in the system is further configured to combine the collected environmental point cloud data, absolute coordinates, inertial data, and relevant signal data involved in the acquisition process to calculate noise index values, RTK signal quality, and laser SLAM covariance, and simultaneously acquire the AGV motion state; define the comprehensive environment adaptability as the support of the current environment for SLAM navigation modal positioning, and use environmental feature entropy values, noise index values, RTK signal quality, and laser SLAM covariance as environmental adaptability influencing factors, and calculate the comprehensive environment adaptability using a weighted product model algorithm; and then use the AGV motion state to evaluate the comprehensive environment adaptability. Dynamic correction is implemented to reduce the support of SLAM navigation mode positioning when the AGV is in a preset high-speed state or a sharp turn state, thereby obtaining the final comprehensive environment adaptability. Based on the final comprehensive environment adaptability, it is determined whether the second triggering condition for outputting the navigation mode switching command is met. The second triggering condition is as follows: when the final comprehensive environment adaptability is less than the first environment adaptability threshold and lasts for a preset time, the SLAM to RTK navigation mode switching is triggered; when the comprehensive environment adaptability is greater than or equal to the second environment adaptability threshold and lasts for a preset time, and the indoor map matching confidence reaches a preset confidence level, the RTK to SLAM navigation mode switching is triggered.
[0065] The decision module 200 in the system is also used to input environmental point cloud data into a pre-built scene recognition model, output the current scene category and the confidence level of each category; select scene categories with a category confidence level greater than the category confidence level threshold, and calculate the comprehensive environment adaptability by matching the combination form and weight coefficient of the environment adaptability influencing factors according to the selected scene category.
[0066] In a specific embodiment, the fusion module 300 in the system is further configured to map the final environment adaptability to the fusion weights in the process of fusing positioning data under SLAM navigation mode and positioning data under RTK navigation mode using an adaptive Kalman filter algorithm, thereby obtaining the corresponding weights of positioning data under different modes; and based on the corresponding weights of positioning data under different modes, combined with the positioning errors of the two navigation modes, update the process noise covariance and observation noise covariance in real time; and introduce a fading factor calculated based on the covariance matching of the residual sequence between prediction and observation to update the process noise covariance in real time.
[0067] This application also discloses a computer-readable storage medium.
[0068] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed, such as the AGV cross-workshop navigation seamless switching method described above. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0069] This application also discloses a computer device.
[0070] Specifically, the computer device includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and executed to perform the aforementioned seamless switching method for AGV cross-workshop navigation.
[0071] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.
Claims
1. A method for seamless switching of AGV navigation across workshops, characterized in that, Includes the following steps: Collect environmental point cloud data, absolute coordinates, and inertial data; Based on the environmental point cloud data, the current environment type is determined and a navigation mode switching command is output to guide the switching of navigation modes; The output process of the navigation mode switching command includes: Calculate the environmental feature entropy value based on the collected environmental point cloud data; Based on the environmental feature entropy value, it is determined whether a first triggering condition for outputting a navigation mode switching command is met. If the first triggering condition is met, a navigation mode switching command is output. The first triggering condition includes: when the environmental feature entropy value is less than a first feature entropy value threshold and lasts for a preset time, a SLAM to RTK navigation mode switching is triggered; when the environmental feature entropy value is greater than or equal to a second feature entropy value threshold and lasts for a preset time, and the indoor map matching confidence reaches a preset confidence level, an RTK to SLAM navigation mode switching is triggered; the map matching confidence is the matching confidence between the navigation position corresponding to the current mode and the indoor map position. During the switching process, positioning data from different navigation modes are merged to eliminate positioning jumps; AGV motion control commands are generated based on the fused positioning data.
2. The seamless switching method for AGV cross-workshop navigation according to claim 1, characterized in that, include: Environmental point cloud data, absolute coordinates, and inertial data are collected using 3D LiDAR, RTK-GNSS devices, and IMU sensors.
3. The seamless switching method for AGV cross-workshop navigation according to claim 2, characterized in that, The process of eliminating the positioning jump includes: When the RTK signal is interrupted, the absolute coordinates collected by the RTK-GNSS device and the inertial data collected by the IMU sensor are fused by extended Kalman filtering, and the fusion weight is dynamically adjusted according to the PDOP value of the RTK positioning accuracy of the RTK-GNSS device. When the PDOP value is less than or equal to the PDOP threshold, the weight of the absolute coordinates is greater than 80%; when the PDOP value is greater than the PDOP threshold, the weights of the absolute coordinates and the inertial data are each half. The process of eliminating the positioning jump also includes: An adaptive Kalman filter algorithm is used to fuse positioning data from SLAM navigation mode and RTK navigation mode to obtain positioning results. The fusion steps include: real-time calculation of positioning errors of the two navigation modes; real-time updating of process noise covariance and observation noise covariance based on the positioning errors of the two navigation modes; and fusing positioning data of the two modes based on the covariance matrix to output positioning results.
4. The seamless switching method for AGV cross-workshop navigation according to claim 2, characterized in that, The AGV motion control command generation step includes: interpolating the linear velocity and angular velocity commands of the AGV at the moment of switching, and limiting the corresponding velocity change rate to less than the preset velocity change rate.
5. The seamless switching method for AGV cross-workshop navigation according to claim 1, characterized in that, The process of outputting the navigation mode switching command also includes: The steps for calculating the entropy value of environmental features include: dividing the collected environmental point cloud data into local neighborhoods, extracting the geometric feature types of each neighborhood, calculating the probability of each type of feature, and calculating the information entropy value; The sliding window adaptive threshold method is adopted to calculate the mean and standard deviation of the historical entropy value queue within a sliding window of a preset length, and to update the first feature entropy value threshold and the second feature entropy value threshold based on the calculated mean and standard deviation. The rate of change of environmental feature entropy is calculated and the preset duration is updated according to the rate of change of environmental feature entropy, so that when the rate of change of environmental feature entropy increases, the preset duration is shortened; when the rate of change of environmental feature entropy decreases, the preset duration is extended.
6. The seamless switching method for AGV cross-workshop navigation according to claim 1, characterized in that, The process of outputting the navigation mode switching command also includes: By combining the collected environmental point cloud data, absolute coordinates, inertial data, and relevant signal data involved in the acquisition process, the noise index, RTK signal quality, and laser SLAM covariance are calculated, and the AGV motion status is acquired simultaneously. The comprehensive environment adaptability is defined as the degree of support of the current environment for SLAM navigation modal positioning. The environmental feature entropy value, noise index value, RTK signal quality, and laser SLAM covariance are used as the influencing factors of the environment adaptability. The comprehensive environment adaptability is calculated by a weighted product model algorithm. The overall environmental adaptability is dynamically corrected by utilizing the AGV's motion state, so that the support for SLAM navigation modal positioning is reduced when the AGV's motion state is in a preset high-speed state or a sharp turn state. To obtain the final overall environmental adaptability; Based on the final comprehensive environment adaptability, determine whether the second triggering condition for outputting the navigation mode switching command is met; the second triggering condition is: when the final comprehensive environment adaptability is less than the first environment adaptability threshold and continues for a preset time, trigger the SLAM to RTK navigation mode switching; when the comprehensive environment adaptability is greater than or equal to the second environment adaptability threshold and continues for a preset time, and the indoor map matching confidence reaches a preset confidence level, trigger the RTK to SLAM navigation mode switching.
7. The seamless switching method for AGV cross-workshop navigation according to claim 6, characterized in that, The process of outputting the navigation mode switching command also includes: Input environmental point cloud data into a pre-built scene recognition model, and output the current scene category and the confidence level of each category; Select scene categories with a category confidence level greater than the category confidence level threshold, and calculate the comprehensive environment adaptability by matching the combination form and weight coefficient of the environment adaptability influencing factors according to the selected scene categories.
8. The seamless switching method for AGV cross-workshop navigation according to claim 6, characterized in that, The process of eliminating the positioning jump also includes: In the process of fusing positioning data from SLAM navigation mode and RTK navigation mode using an adaptive Kalman filter algorithm, the final environment adaptability is mapped to the fusion weights to obtain the corresponding weights of positioning data under different modes. Based on the corresponding weights of positioning data under different modes, and combined with the positioning errors of the two navigation modes, the process noise covariance and observation noise covariance are updated in real time. Furthermore, a fading factor calculated based on the covariance matching of the residual sequence between prediction and observation is introduced to update the process noise covariance in real time.
9. A seamless switching system for AGV cross-workshop navigation, characterized in that, include: The perception module is used to collect environmental point cloud data, absolute coordinates, and inertial data; The decision module is used to determine the current environment type based on the environmental point cloud data and output a navigation mode switching command to guide the switching of navigation modes; The process of outputting the navigation mode switching command includes: calculating the environmental feature entropy value based on the collected environmental point cloud data; determining whether the first triggering condition for outputting the navigation mode switching command is met based on the environmental feature entropy value; and outputting the navigation mode switching command when the first triggering condition is met. The first triggering condition includes: triggering SLAM to RTK navigation mode switching when the environmental feature entropy value is less than a first feature entropy value threshold and lasts for a preset time; and triggering RTK to SLAM navigation mode switching when the environmental feature entropy value is greater than or equal to a second feature entropy value threshold and lasts for a preset time, and the indoor map matching confidence reaches a preset confidence level. The map matching confidence level is the matching confidence between the navigation position corresponding to the current mode and the indoor map position. The fusion module is used to merge positioning data from different navigation modes during the switching process to eliminate positioning jumps; The control module is used to generate AGV motion control commands based on the fused positioning data.