Global initialization method and system for autonomous vehicle based on 6g leading constraints in complex environment

Abstract

The application relates to a kind of complex environment-based global initialization method of 6G leading constraint of autonomous vehicle, comprising: obtaining the multi-source position information of autonomous vehicle;With 6G positioning information as the main global initialization constraint source, candidate pose set is constructed in combination with the GNSS positioning information and historical position information;According to the accuracy and reliability of each candidate pose, the corresponding particle initialization distribution range is adaptively set, and the initial particle set constrained by 6G positioning information space is generated;The matching score of each particle in the initial particle set is obtained and the weight is updated by matching laser radar real-time point cloud and high-precision map;High-weight particle set is screened out through resampling and iterative convergence judgment;Statistical analysis and weighted fusion are carried out on the high-weight particle set, and the global initialization pose containing the position and attitude of the vehicle is generated.The application realizes the rapid and accurate convergence of global position and attitude in complex environment with 6G as the leading constraint.

Description

Technical Field

[0001] This invention belongs to the field of intelligent connected vehicle positioning and navigation technology, and relates to a global initialization method and system for autonomous vehicles based on 6G dominant constraints in complex environments. Background Technology

[0002] With the rapid development of industrial automation and intelligent transportation technologies, autonomous vehicles are increasingly being used in complex environments such as factories, ports, and logistics warehouses. In these scenarios, vehicles need to perform high-precision loading and unloading operations, transport goods, or park at designated berths, thus placing extremely high demands on the accuracy and reliability of their global pose.

[0003] Currently, global positioning of autonomous vehicles typically relies on Global Navigation Satellite Systems (GNSS) to obtain initial position information, which is then used as the initial reference for LiDAR positioning. Global pose localization is achieved through matching online point clouds with high-precision maps. However, existing global initialization technologies (i.e., rapidly determining the vehicle's global pose under conditions of no prior knowledge or unreliable prior knowledge) have long faced severe challenges in practical applications, specifically in the following aspects: (1) GNSS signals are prone to failure or insufficient accuracy in complex environments. In enclosed or semi-enclosed operating environments such as factories and ports, there are usually a large number of steel structures and buildings that block the signal, making GNSS signals extremely susceptible to multipath interference or complete failure. This makes it impossible for traditional positioning methods based on GNSS or lidar to obtain accurate global prior information during the initialization phase, which in turn leads to the inability to quickly converge to the correct global pose, or even positioning jumps or initialization failures, failing to provide a stable initialization reference, seriously affecting the safe operation of vehicles and operational efficiency.

[0004] (2) Initialization methods based on historical location information or prior pose prediction have poor robustness. To address the GNSS failure problem, existing technologies typically use locally stored historical location information of the vehicle as the initial reference for particle filtering, such as CN121067843A and CN106707315A. However, these methods have significant limitations: when the vehicle undergoes cross-regional scheduling, manual towing, or location changes between the last power-off and the current power-on, the historical location information becomes unreliable or completely invalid. In this situation, the system lacks effective alternatives, resulting in a very high probability of initialization failure.

[0005] (3) Pure laser point cloud global search is inefficient. In the absence of effective prior information, pure laser point cloud matching methods (such as ICP, GICP, NDT, etc.) usually need to perform brute-force search or large-scale grid search across the entire map, such as CN114779264B, CN113268697B, and CN115100274B. This method has a huge search space, and the computational load increases linearly or even exponentially with the map size. In large-scale environments with repetitive structures, such as factories and ports, dynamic objects such as cranes and hoists can easily interfere with the matching results. Initialization takes a very long time and is very prone to falling into local optima, resulting in an unstable success rate.

[0006] (4) Single external positioning sources (such as UWB / geomagnetic) have limitations. Technologies such as ultra-wideband (UWB), geomagnetic, or reflective landmarks are used for assisted positioning, such as CN115932926A and CN119689533A. These solutions require additional deployment of expensive infrastructure and have high maintenance costs. Moreover, these single external positioning sources often do not have wide-area coverage capabilities. When the signal is lost or the accuracy degrades, the system lacks redundancy and backup, making it difficult to meet the high reliability requirements of complex industrial environments.

[0007] (5) The global initialization of particle filtering is unstable. In the absence of effective prior information, traditional particle filtering global initialization methods usually need to seed a large number of particles across the entire scene and rely entirely on sensor observations to iteratively converge to the global pose, such as CN109492769A and CN120342804A. This method requires the generation of a large number of particles to cover the entire state space, resulting in huge computational resource consumption; in complex environments with repetitive structures or dynamic disturbances, the particle set is prone to degradation, resulting in slow convergence speed. Due to the lack of external spatial constraints, the success rate of this method in the initialization stage is extremely unstable, making it difficult to meet the real-time and reliability requirements under complex working conditions.

[0008] Therefore, existing technologies lack a global initialization method that can still achieve fast, high-precision, and highly robust operation in complex environments where GNSS signals are unavailable and historical location information is invalid. Summary of the Invention

[0009] The purpose of this invention is to address the problems existing in the prior art by proposing a global initialization method and system for autonomous vehicles based on 6G dominant constraints in complex environments. With 6G positioning as the core constraint and combined with laser point cloud matching, it can quickly and accurately establish the global position and attitude of autonomous vehicles in complex environments, and is applicable to complex environments such as factories, ports, and logistics warehouses.

[0010] This invention achieves fast and robust global initialization by introducing 6G-assisted positioning and constructing a multi-source candidate location system. First, 6G positioning is introduced as a core global prior constraint in complex environments. When GNSS signals are unavailable or unstable, the coverage and positioning capabilities of the 6G network in complex environments such as indoors or semi-enclosed spaces are utilized to make 6G positioning information a high-priority candidate location information, providing a reliable global initialization area constraint for laser positioning, thus overcoming the problem of inability to initialize in traditional GNSS-restricted scenarios. Second, based on the 6G prior constraint, the initialization speed and accuracy are improved. Locally stored historical location information and available GNSS information are fused, and an initial particle set is generated through an accuracy-adaptive particle filtering strategy, enabling the particles to converge quickly and stably to the actual vehicle pose, significantly improving the efficiency and initialization accuracy of laser point cloud matching with high-precision maps. Finally, a multi-source candidate location information fusion mechanism is constructed to enhance system robustness. By unifying 6G positioning information, GNSS positioning information, and historical location information into the candidate location information system, and dynamically adjusting the weight of each information source according to its availability and accuracy, the impact of a single location information source failure on the global initialization process is reduced, improving the reliability and stability of global positioning initialization for autonomous vehicles in complex environments.

[0011] To achieve the above objectives, this invention provides a global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments, comprising the following steps: Step 1: Obtain multi-source location information of the autonomous vehicle. The multi-source location information includes at least 6G positioning information, GNSS positioning information, and historical location information stored locally by the vehicle. Step 2: Using the 6G positioning information as the main global initialization constraint source, construct a candidate pose set by combining the GNSS positioning information and historical location information; based on the accuracy and reliability of each candidate pose, adaptively set the corresponding particle initialization distribution range to generate an initial particle set spatially constrained by the 6G positioning information. Step 3: Match the real-time point cloud of the LiDAR with the high-precision map to obtain the matching score of each particle in the initial particle set and update the weight; through resampling and iterative convergence judgment, select the high-weight particle set. Step 4: Perform statistical analysis and weighted fusion on the high-weight particle set to generate a global initial pose containing the vehicle's position and attitude, and use it as the initial input of the laser positioning system to achieve global initialization of autonomous vehicles in complex environments.

[0012] This invention introduces 6G positioning technology. 6G networks offer wide coverage, high update frequency, and sub-meter or even centimeter-level positioning accuracy in indoor and semi-enclosed environments, providing reliable prior information for global positioning of lidar. In cases of GNSS failure or unavailable historical location information, 6G positioning can serve as core candidate location information to assist particle filtering in global initialization. The specific process includes dynamically adjusting the particle distribution range based on the accuracy of different candidate location information; higher-accuracy 6G location information corresponds to a smaller distribution range, while lower-accuracy information corresponds to a larger distribution range, thus enabling particles to quickly converge to the actual vehicle location, significantly improving the initialization success rate and convergence speed. By using 6G positioning information as the primary source, combined with GNSS information and historical location information, this invention can achieve highly reliable and rapid global positioning initialization of autonomous vehicles in complex environments such as factories, ports, and logistics warehouses, providing accurate initial poses for online point cloud matching and high-precision map matching of lidar, significantly improving global positioning reliability and operational efficiency.

[0013] The present invention further adopts the following technical solution: In step 2, the 6G positioning information is used as the primary global initialization constraint source, and a candidate pose set is constructed by combining the GNSS positioning information and historical location information, including: Receive raw location information sent from the 6G network side, the information including at least spatial coordinates, timestamp and location quality parameters; The validity and stability of the raw 6G positioning information are judged, and positioning data with timed-out timestamps or abnormal consecutive frame jumps are removed. The effective 6G positioning information is transformed from its original coordinate system to a coordinate system compatible with the high-precision geolocation system. Figure 1 A unified global coordinate system is established, and its positioning accuracy and reliability are evaluated based on channel conditions or signal angle of arrival parameters; When 6G positioning information is available and stable, it is set as the highest priority candidate pose information to dominate the spatial distribution center of subsequent particles. The GNSS positioning information is used to determine the solution state. If it is a fixed solution, it is used as a medium variance constraint source to generate a medium-range particle cloud; if it is a floating-point solution, it is used as a large variance constraint source to generate a large-range particle cloud. The historical location information is used to determine consistency. If the vehicle does not move significantly, it is used as a medium variance constraint source to generate a medium-range particle cloud; if the vehicle moves slightly, it is used as a large variance constraint source to generate a large-range particle cloud. The processed 6G positioning information, GNSS positioning information, and historical location information are sorted according to priority, with 6G positioning information having the highest priority and used to guide the distribution of particles within a small spatial range. GNSS positioning information and historical location information are used as supplementary candidates to enhance global search capabilities.

[0014] Furthermore, in step 2, the corresponding particle initialization distribution range is adaptively set, including: Establish a mapping relationship between candidate pose accuracy and particle distribution variance. For candidate poses with higher accuracy, set a smaller particle distribution range to reduce the search space and improve the convergence speed. For candidate poses with lower accuracy, set a larger particle distribution range to enhance the global search capability and avoid getting trapped in local optima. When the candidate pose is high-precision 6G positioning information, a low-variance particle cloud is generated, with the smallest particle distribution range, which is used to accurately constrain the global initialization search space. When the candidate pose is a fixed GNSS solution or historical location information that has not moved, a particle cloud with moderate variance is generated, with a moderate particle distribution range, to help verify the accuracy of 6G positioning information. When the candidate pose is a GNSS floating-point solution or historical position information with a small range of movement, a particle cloud with large variance is generated, which has the largest particle distribution range. This is used to enhance the global search capability and avoid getting trapped in local optima.

[0015] In step 3, the matching score calculation for each particle in the initial particle set includes: Based on the candidate poses represented by particles, the real-time LiDAR point cloud is transformed from the vehicle system to the global map coordinate system; The algorithm employs NDT (Normal Distribution Transform), ICP (Iterative Closest Point), or GICP (Generalized Iterative Closest Point) to calculate the degree of matching between the transformed point cloud and fixed structural features in the high-precision map. The matching degree is normalized into particle weights, so that particles with high consistency with the map structure receive high weights, while particles that match dynamic interference objects or deviate from the map structure receive low weights.

[0016] Furthermore, in step 3, the iterative convergence determination includes: Monitor the spatial distribution variance of the particle set after resampling. When the variance is less than a preset threshold, it is determined that convergence has been achieved. Alternatively, monitor the stability of particle matching scores across multiple consecutive frames, and determine that the particles have converged when the fluctuation in matching scores is less than a preset range. If any of the above convergence conditions are met, the iteration stops and the current set of high-weight particles is output; otherwise, the next round of particle scoring, weight updating, and resampling iterations continues.

[0017] Step 4 involves statistical analysis and weighted fusion of the high-weight particle set, including: The converged particles are sorted according to their weights, and the top-ranked particles are selected to form a high-weight candidate set. The spatial mean and covariance of the high-weight candidate set are calculated, and smoothing is performed in combination with the neighborhood distribution characteristics. The spatial mean and covariance matrix are used to characterize the concentration and stability of the pose. When the statistical confidence of the high-weight particle set is insufficient or the pose distribution dispersion exceeds a preset threshold, the second-highest weight particle is introduced to participate in the statistical calculation or the initialization is judged to have failed, so as to improve the stability and robustness of the output pose.

[0018] Furthermore, in step 4, its global initial pose is used as the initial input to the laser positioning system, including: Receive global initial pose information output by the high-weighted particle processing module, wherein the pose includes the vehicle's position and attitude information in the map coordinate system; The global initial pose is subjected to basic validity and confidence verification. When the pose meets the preset consistency condition, it is injected into the laser positioning module as the initial guess value of the point cloud matching algorithm. Under the initial pose constraint, the laser positioning module performs continuous point cloud matching and pose update within a local area to achieve stable positioning of the vehicle in the factory environment.

[0019] This invention also provides a global initialization system for autonomous vehicles based on 6G dominant constraints in complex environments, including: a candidate position information acquisition and filtering module, a point cloud and map matching module, a particle filter pose inference module, a high-weight particle processing module, and a laser positioning initialization module. The candidate position information acquisition and filtering module is used to acquire multi-source position information, which includes at least 6G positioning information, GNSS positioning information, and historical position information, and to perform availability assessment and priority ranking on the multi-source position information to generate a candidate pose set subject to 6G dominant constraints. The point cloud and map matching module is used to receive real-time LiDAR point cloud data of the vehicle and match and score it with a high-precision map. The particle filtering pose inference module is used to generate initial particles based on the candidate pose set, iteratively update the particle weights based on the matching results of the lidar point cloud and the high-precision map, and filter out high-weight particles. The high-weighted particle processing module is used to perform statistical analysis and weighted fusion on the converged high-weighted particle set to generate a global initial pose. The laser positioning initialization module is used to receive the global initial pose and input it as an initial value into the laser positioning system to achieve rapid global positioning of the vehicle. The present invention further provides a device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described global initialization method for autonomous vehicles based on 6G dominant constraints under complex environments.

[0020] The present invention further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the global initialization method for autonomous vehicles based on 6G dominant constraints in the complex environment.

[0021] This invention is the first to use 6G positioning information as a reliable spatial constraint for global initialization of laser positioning, directly limiting the particle search area to a trusted subspace, realizing the transformation from full-map search to regional search, and upgrading from traditional algorithm pruning to prior constraints, significantly reducing computational complexity and achieving second-level fast global initialization; based on the accuracy of 6G positioning, an accuracy-adaptive particle distribution strategy is adopted, which can dynamically adjust the number and distribution range of particles, greatly reducing the calculation of invalid particles and invalid matches while ensuring coverage of uncertainties, thus reducing computational overhead. This invention eliminates the need for historical poses as a prerequisite, using them only as candidate information. When historical poses are unavailable, unreliable, or the vehicle has been moved, 6G positioning can be used directly as the core alternative constraint to complete global initialization, improving system robustness. Simultaneously, 6G positioning is introduced as an external coarse positioning method for complex industrial scenarios such as factories, ports, and logistics warehouses. It constructs a redundant candidate location system with GNSS and historical poses, supporting dynamic switching and fusion of multi-source information. This avoids initialization failure due to the failure of a single external positioning source, adapting to discontinuous usage scenarios. Ultimately, it achieves rapid, stable, low-cost, and high-success-rate global initialization in complex industrial environments. Attached Figure Description

[0022] The invention will now be further described with reference to the accompanying drawings.

[0023] Figure 1 This is a system block diagram of the global initialization method in this invention.

[0024] Figure 2 This is a schematic diagram of the candidate location information acquisition and filtering module in this invention.

[0025] Figure 3 This is a schematic diagram of the particle filter inference module in this invention.

[0026] Figure 4This is a framework diagram of the global initialization system based on 6G dominant constraints in complex environments according to the present invention.

[0027] Figure 5 This is a flowchart of the 6G positioning auxiliary information processing of the present invention.

[0028] Figure 6 This is a diagram showing the relationship between candidate position information and particle distribution in this invention. Detailed Implementation

[0029] The technical solutions of this application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments provided herein are merely illustrative of this application and are not intended to limit this application. Furthermore, the embodiments provided below are some embodiments for implementing this application, and not all embodiments for implementing this application. Unless otherwise specified, the technical solutions described in the embodiments of this application can be implemented in any combination.

[0030] This invention discloses a global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments, comprising the following steps: Step 1: Obtain multi-source location information of the autonomous vehicle. The multi-source location information includes at least 6G positioning information, GNSS positioning information, and historical location information stored locally by the vehicle.

[0031] Step 2: Using 6G positioning information as the primary global initialization constraint source, and combining it with the vehicle's locally stored historical location information and GNSS positioning information, a candidate pose set is constructed. Based on the accuracy and reliability of each candidate pose, the spatial distribution range and quantity of the initial particles are constrained and guided primarily by the 6G positioning information.

[0032] (1) Using 6G positioning information as the main global initialization constraint source, this step specifically includes: receiving the raw positioning information sent by the 6G network side, the raw positioning information including at least spatial coordinates, timestamps and positioning quality parameters; judging the validity and stability of the raw 6G positioning information, and removing positioning data with timestamp timeouts or abnormal consecutive frame jumps; transforming the valid 6G positioning information from its original coordinate system to a coordinate system compatible with high precision geolocation. Figure 1 A global coordinate system is established, and its positioning accuracy and reliability are evaluated based on channel status or signal angle of arrival parameters. When 6G positioning information is available and stable, it is set as the highest priority candidate pose information to dominate the spatial distribution center of subsequent particles.

[0033] (2) Construct a candidate pose set by combining GNSS positioning information and historical location information; this step specifically includes: judging the solution state of GNSS positioning information. If it is a fixed solution, it is used as a medium variance constraint source to generate a medium-range particle cloud; if it is a floating-point solution, it is used as a large variance constraint source to generate a large-range particle cloud; judging the consistency of historical location information. If the vehicle has not moved significantly, it is used as a medium variance constraint source to generate a medium-range particle cloud; if the vehicle has moved slightly, it is used as a large variance constraint source to generate a large-range particle cloud; sort the processed 6G positioning information, GNSS positioning information and historical location information according to priority. Among them, 6G positioning information has the highest priority and is used to guide the distribution of particles in a smaller spatial range. GNSS positioning information and historical location information are used as supplementary candidates to enhance the global search capability.

[0034] (3) Based on the accuracy and reliability of each candidate pose, the corresponding particle initialization distribution range is adaptively set to generate an initial particle set constrained by 6G positioning information space. This step specifically includes: establishing a mapping relationship between candidate pose accuracy and particle dispersion variance. For candidate poses with higher accuracy, a smaller particle dispersion range is set, and high-density particle dispersion is performed to reduce the search space and improve convergence speed. For candidate poses with lower accuracy, a larger particle dispersion range is set, and low-density particle dispersion is performed to enhance global search capability and avoid getting trapped in local optima, thus achieving an accuracy-adaptive particle generation strategy. When the candidate pose is high-precision 6G positioning information, a low-variance particle cloud is generated with the smallest particle distribution range to accurately constrain the global initialization search space. When the candidate pose is a fixed GNSS solution or historical location information that has not moved, a medium-variance particle cloud is generated with a moderate particle distribution range to assist in verifying the accuracy of the 6G positioning information. When the candidate pose is a floating-point GNSS solution or historical location information with small-scale movement, a large-variance particle cloud is generated with the largest particle distribution range to enhance global search capability and avoid getting trapped in local optima.

[0035] Step 3: The particle filter pose inference module is used to iteratively update the particles, normalize their weights, and resample them, and a stable set of high-weight particles is obtained based on the convergence criterion.

[0036] (1) Matching the real-time point cloud of the LiDAR with the high-precision map to obtain the matching score of each particle in the initial particle set and update the weight; the matching score calculation in this step includes: based on the candidate pose represented by the particle, converting the real-time LiDAR point cloud from the vehicle system to the global map coordinate system; using the Normal Distribution Transform (NDT), Iterative Closest Point (ICP) or Generalized Iterative Closest Point (GICP) algorithm to calculate the matching degree between the converted point cloud and the fixed structural features in the high-precision map; the point cloud matching process can be implemented using existing mature point cloud registration methods in the field. The above NDT, ICP or GICP algorithms are only examples, and the present invention does not limit them. Those skilled in the art can choose other equivalent methods to implement them according to specific application scenarios; normalize the matching degree into particle weights so that particles with high consistency with the map structure get high weights, while particles that match dynamic interference objects or deviate from the map structure get low weights; the weight calculation process can be implemented using the weight update method in existing particle filtering, and the present invention does not limit it.

[0037] (2) By resampling and iterative convergence judgment, a high-weight particle set is selected; the iterative convergence judgment in this step includes: monitoring the spatial distribution variance of the particle set after resampling. When the variance is less than the preset threshold, it is determined that the particles are highly concentrated in space, indicating that convergence has been achieved; or, monitoring the stability of particle matching scores in multiple consecutive frames. When the fluctuation of the matching score is less than the preset range, it is determined that the matching state is stable, indicating that the particles have converged; if any of the above convergence conditions are met, the iteration is stopped and the current high-weight particle set is output; otherwise, the next round of particle scoring, weight update and resampling iteration continues.

[0038] Step 4 involves statistical analysis and weighted fusion of the high-weight particle set to generate a globally initialized pose containing vehicle position parameters and quaternion attitude parameters. This pose is then used as the initial input to the laser positioning system to achieve global initialization of autonomous vehicles in complex environments. The statistical analysis and weighted fusion in this step include: sorting the converged particles according to their weights and selecting the top-ranked particles to form a high-weight candidate set; calculating the spatial mean and covariance of the high-weight candidate set and smoothing it using neighborhood distribution characteristics; the spatial mean and covariance matrices are used to characterize the concentration and stability of the pose; when the statistical confidence of the high-weight particle set is insufficient or the pose distribution dispersion exceeds a preset threshold, the second-highest weighted particle is introduced to participate in the statistical calculation or the initialization is deemed failed, thereby improving the stability and robustness of the output pose. In this step, the global initial pose as the initial input includes: receiving the global initial pose information output by the high-weight particle processing module, the pose including the vehicle's position and attitude information in the map coordinate system; performing basic validity and confidence checks on the global initial pose, and when the pose meets the preset consistency conditions, injecting it into the laser positioning module as the initial guess value for the point cloud matching algorithm; under the constraints of the initial pose, the laser positioning module performs continuous point cloud matching and pose updates in a local area to achieve stable positioning of the vehicle in the factory environment.

[0039] In this invention, a global initialization system for autonomous vehicles based on 6G dominant constraints in complex environments is described, with reference to... Figure 1 The system is divided into three main functional areas based on the workflow: an input information processing area, a particle filtering processing area, and a global initialization pose access area. The input information processing area acquires and processes multi-source candidate position information, using 6G positioning information as a high-priority global prior constraint to limit the initialization search area. The particle filtering processing area, based on the 6G prior constraints and combined with available GNSS information and historical position information, employs a precision-adaptive particle generation and iteration strategy to evaluate and converge candidate poses. The global initialization pose access area receives the highly reliable initialization pose output from the particle filtering processing area and provides it to the laser positioning module as the initial pose input. This invention significantly reduces the particle search space by introducing 6G positioning as the dominant global initialization constraint in complex environments, and achieves high-precision pose verification by combining it with lidar point cloud matching, thereby enabling fast and robust global positioning initialization in complex environments.

[0040] The input information processing area includes a candidate location information acquisition and filtering module and a point cloud and map matching module. (Reference) Figure 2The candidate location information acquisition and filtering module is used to acquire multi-source location information, evaluate its availability, and prioritize it to generate a set of candidate poses dominated by 6G, which is used for subsequent particle initialization. The multi-source location information includes at least 6G positioning information, GNSS positioning information, and historical location information. This module receives 6G positioning information in real time as the main global pose prior information, and loads and reads historical location information stored locally on the vehicle, as well as real-time received GNSS positioning information, to form multi-source candidate pose input. The point cloud and map matching module is used to receive real-time LiDAR point cloud data from the vehicle and perform matching and scoring with a high-precision map. This module calculates the matching score between the point cloud and the map for each candidate pose, providing a basis for particle filtering weights. Particles with higher matching scores will receive higher weights, guiding the particles to converge to the true pose.

[0041] The particle filtering processing area includes a particle filtering pose inference module and a high-weight particle processing module. The particle filtering pose inference module generates initial particles based on the candidate pose set, iteratively updates particle weights based on the matching results of the LiDAR point cloud and the high-precision map, and selects high-weight particles. For example... Figure 3 As shown, each round of particle filtering iteration includes the following steps: (1) Initial Particle Generation: Initial particles are generated based on the candidate pose set, with the candidate pose corresponding to the 6G positioning information serving as the main particle seeding constraint source, supplemented by historical location information and GNSS positioning information stored locally on the vehicle. For candidate poses from different sources, this module sets corresponding particle seeding weights and perturbation ranges according to their positioning accuracy and reliability indicators, and randomly samples particles in the position and planar orientation dimensions. Specifically, for 6G positioning information with high positioning accuracy and high confidence, a smaller spatial perturbation range and a higher particle generation weight are set to concentrate the initial particles in their neighborhood area; for GNSS positioning information and historical pose information with relatively low accuracy or decreased reliability, a larger perturbation range is set to enhance the global search capability. The initial particles form a non-uniformly distributed particle set constrained by the 6G positioning information in the global space, providing good initial conditions for subsequent particle weight calculation and iterative convergence.

[0042] (2) Point cloud matching score: For each particle, the LiDAR point cloud is matched with the high-precision map to calculate the matching score of the particle at its current position. The matching score reflects the rationality of the particle pose, and a high score indicates that the particle pose is closer to the real global position.

[0043] (3) Weight normalization: The matching score of each particle is converted into a weight and normalized. The resulting weight is used to reflect the importance of the particle and plays a key role in resampling and convergence judgment.

[0044] (4) Resampling: Resampling is performed based on particle weights, copying high-weight particles and removing low-weight particles to keep the number of particles constant, while increasing the proportion of high-weight particles to ensure that the search direction is concentrated in the most likely pose region.

[0045] (5) Convergence judgment: For the resampled particle set, calculate the variance of the particle distribution or the distance change, etc., to determine whether the particles have converged to the stable region. If the convergence condition is met, stop the iteration; otherwise, proceed to the next round of perturbation, scoring, normalization and resampling iteration.

[0046] It should be noted that the above-mentioned particle weight update, normalization, resampling, and convergence judgment processes can be implemented using existing particle filtering methods in the field, such as the sequential Monte Carlo method based on importance sampling; convergence criteria such as particle distribution variance and matching score stability can also be implemented using conventional statistical analysis methods, and this invention does not limit these methods. The technical implementation of this invention does not depend on a specific particle filtering algorithm.

[0047] The aforementioned particle filter pose inference module implements iterative inference of particle filtering under 6G positioning information constraints. It determines the initial spatial distribution center and disturbance range of particles and generates initial particles under 6G positioning information constraints. It calculates particle matching scores by matching LiDAR point clouds and high-precision maps. After particle weight normalization, it resamples high-weight particles according to the normalized weights to generate the particle set for the next round of iteration. At the same time, it judges whether the iteration has ended according to the particle weight distribution convergence criterion until the convergence condition is met or the maximum number of iterations is reached. Finally, it outputs the highest weight particle set for subsequent fusion. Its core protection points include the particle filter iterative inference and convergence judgment mechanism under 6G positioning constraints, the particle matching scoring function design based on LiDAR point clouds and high-precision maps, and the particle weight update and resampling strategy combined with the 6G constraints.

[0048] The high-weighted particle processing module performs statistical analysis and weighted fusion on the converged high-weighted particle set to generate a global initial pose containing the vehicle's position and attitude, which serves as the input for subsequent laser positioning initialization. The high-weighted particle processing module also performs statistical analysis and weighted fusion on the high-weighted particle set converged by the particle filter pose inference module under 6G guided particle distribution conditions to generate stable and reliable global initialization candidate pose results. These candidate poses include the vehicle's position parameters in the high-precision map coordinate system and attitude parameters represented by quaternions, and serve as the initialization input for the laser positioning system to shorten the convergence time of the subsequent positioning system.

[0049] The global initialization pose access area includes a laser positioning initialization module, which is used to receive the global initialization pose and input it as an initial value into the laser positioning system to achieve rapid convergence of the laser positioning system and thus achieve rapid global positioning of the vehicle.

[0050] This invention adopts a modular global initialization architecture, mainly comprising a candidate location information acquisition and filtering module, a point cloud map matching module, a particle filter pose inference module, a high-weight particle processing module, and a laser positioning initialization module. The candidate location information acquisition and filtering module uses 6G positioning information as the primary constraint source, combining historical location and GNSS positioning information for evaluation and filtering. The point cloud map matching module completes the matching score between the vehicle's real-time point cloud and a high-precision map. The particle filter pose inference module completes particle weight iterative updates, resampling, and convergence judgment under 6G constraints. The high-weight particle processing module fuses converged particles and outputs the global initialization pose. The laser positioning initialization module achieves rapid convergence of the laser positioning system based on this pose. Its core protection points are the modular global initialization system structure and data flow design guided by 6G positioning information, and the system scalability design that supports the access and processing of new external positioning information sources. Simultaneously, the system has three core highlights: a precision-adaptive particle distribution strategy that dynamically adjusts the number and distribution density of particles according to 6G positioning accuracy; and the ability to complete global initialization using 6G even when GNSS fails and historical location information is unavailable. It has the advantages of leading a multi-source fusion initialization mechanism and achieving collaborative optimization of particle filtering and point cloud matching through 6G constraints, effectively reducing the search space and improving the initialization speed and accuracy.

[0051] It should be noted that the point cloud matching algorithm, particle filtering algorithm, weight calculation method, and statistical analysis method involved in this invention can all be implemented using existing mature technologies in the field. This invention does not improve the above-mentioned basic algorithms themselves, but rather constrains and reconstructs the particle distribution and search space in the global initialization process by introducing 6G positioning information, thereby achieving technical effects different from existing technologies. The core of this invention lies in constructing a global initialization framework based on 6G positioning information as the dominant constraint. Through priority fusion of multi-source pose information and adaptive adjustment of particle distribution, it achieves fast and robust global initialization of autonomous vehicles in complex environments. Therefore, without departing from the core idea of ​​this invention, those skilled in the art can replace or adjust the specific implementation method. The method described in this invention can complete the entire process using the above-mentioned existing technologies, and has good engineering feasibility.

[0052] Example 1 refer to Figure 4This embodiment proposes a global initialization method and system for laser positioning, using 6G positioning information as the core prior constraint, for complex environments such as factories, ports, and logistics warehouses. The system includes a candidate location information acquisition and filtering module, a point cloud and map matching module, a particle filter pose inference module, a high-weight particle processing module, and a laser positioning initialization module. (Reference) Figure 2 and Figure 4 The left-hand section, the location information acquisition and filtering module, prioritizes acquiring 6G positioning information when the vehicle powers on, restarts, or recovers from positioning failure. It then combines this information with GNSS and locally stored historical location information to construct a candidate pose set for global initialization, providing spatial prior constraints for subsequent particle generation. Candidate location information includes, but is not limited to, 6G positioning information, GNSS positioning information, and historical location information. 6G positioning information is provided by the 6G positioning network deployed within the factory or port area, serving as the primary external positioning source in complex environments such as closed or semi-closed spaces. GNSS positioning information serves as auxiliary coarse positioning information when available. Historical location information is the historical location information saved during the vehicle's last power-off or operating cycle. This module performs availability assessment, accuracy evaluation, and stability analysis on location information from different sources, and prioritizes and manages the data based on the evaluation results. When 6G positioning information is available and stable, it is used as the highest priority candidate pose information. When 6G positioning information is obstructed or its accuracy decreases, GNSS positioning information and historical location information serve as supplementary candidates. When historical location information is inconsistent with the current vehicle state, its priority can be reduced or it can be eliminated. Based on this, the candidate position information acquisition and filtering module adaptively generates the corresponding particle initialization range according to the accuracy level of each candidate position information: the candidate pose with high accuracy corresponds to a smaller particle spreading range, and the candidate pose with low accuracy corresponds to a larger particle spreading range, thereby significantly reducing the global initialization search space while ensuring search coverage.

[0053] Through the above design, the candidate location information acquisition and filtering module introduces 6G positioning information as the core spatial constraint source in the global initialization stage, effectively avoiding the problem of full-map search without prior conditions in the global initialization of traditional laser positioning, and providing a highly reliable and efficient initialization prior for the subsequent particle filtering process.

[0054] (1) 6G positioning-assisted information processing This embodiment introduces 6G positioning as an important external positioning constraint source in complex environments. It utilizes the positioning capabilities provided by the 6G network to provide highly reliable initial position constraints for the vehicle's global initialization. (Reference) Figure 5 Its processing flow includes: ① 6G location information reception The vehicle reports its identity, location tag, or communication session identifier to the 6G network via its onboard communication unit. Based on measurement information from the base station, the 6G network generates an initial vehicle location result and sends it to the onboard positioning system. The 6G positioning information includes at least two-dimensional or three-dimensional spatial coordinates, a positioning timestamp, and positioning quality-related parameters.

[0055] ② Judgment of positioning effectiveness and stability The received 6G location information is validated, including but not limited to whether the location data is complete, whether the location timestamp is within the valid time window, and whether there are obvious jumps in the location results of multiple consecutive frames.

[0056] When the positioning result meets the stability threshold requirement, the 6G positioning information is determined to be a valid positioning result; otherwise, the positioning information is marked as low confidence or temporarily excluded from the generation of candidate location information.

[0057] ③ Coordinate system transformation and calibration compensation The positioning coordinates output from the 6G network side will be transformed from their original coordinate system to a system compatible with high-precision geodetic coordinates. Figure 1 The global coordinate system is aligned, including: applying pre-calibrated base station coordinate mapping relationships; using coordinate transformation matrices or nonlinear mapping models to complete coordinate alignment; and compensating for and correcting any possible system deviations.

[0058] ④ Positioning accuracy and reliability assessment Based on the measurement parameters in the 6G positioning information, the accuracy and reliability of the current positioning result are evaluated. The evaluation methods include, but are not limited to: error estimation based on the time difference of arrival (TDoA), angle stability analysis based on the angle of arrival (AoA), signal quality evaluation based on channel state information (CSI), or a weighted average of the above evaluation results.

[0059] Based on the evaluation results, a corresponding uncertainty range or weight level is assigned to the 6G positioning information.

[0060] ⑤ Candidate location information generation and priority setting Based on the effective 6G positioning results, corresponding candidate initial pose information is generated and added to the candidate position information set. In the candidate position information set, the 6G positioning information is set as a high-priority candidate, and its corresponding spatial uncertainty range is used to constrain the initial particle distribution area of ​​the particle filter. When GNSS or historical position information is available at the same time, the 6G positioning information is used to guide the distribution of particles within a smaller spatial range.

[0061] ⑥ Output to the particle filter initialization module The processed 6G candidate location information, including its location coordinates, confidence index, and uncertainty constraint parameters, is output to the particle filter initialization module as one of the input conditions for particle generation and propagation.

[0062] In some implementations, when 6G positioning information fails briefly or its accuracy decreases, the system automatically lowers its candidate priority and degenerates into a particle initialization strategy based on GNSS or historical location information.

[0063] (2) GNSS positioning information processing In industrial environments such as factories, GNSS positioning errors can reach 10 meters due to steel structure obstruction and multipath effects, making it difficult to use directly as a positioning result. However, it can be used as a coarse range constraint. The processing flow includes: ① Receive GNSS positioning data (usually in WGS84 coordinate system); ② By using the coordinate system transformation unit, the GNSS coordinates are transformed to a coordinate system compatible with high-precision geodetic coordinates. Figure 1 A unified global coordinate system; ③ Determine the GNSS solution status. If it is a fixed solution or a floating-point solution, add it to the candidate location information set; if the solution status is unstable, do not use it as candidate location information. ④ Set the particle distribution range according to the GNSS solution type. Fixed solutions correspond to a smaller distribution range, while floating-point solutions correspond to a larger distribution range.

[0064] By employing the above methods, large-scale particle searches on a global scale can be avoided while ensuring coverage of uncertainties.

[0065] (3) Historical location information processing When the vehicle is powered off, it stores its current pose information as historical location information and automatically retrieves it upon the next power-on. The processing flow includes: ① After power-on, historical location information is read from the preset path; ② Combine the vehicle's current mileage information with the mileage corresponding to the historical position to make a consistency judgment; ③ If it is determined that the vehicle has not moved significantly or has moved only a small area, then the historical location information is added to the candidate set; ④ Adjust the particle distribution range based on the movement results. If no movement occurs, use a small-range distribution; if a small-range movement occurs, use an expanded-range distribution.

[0066] By incorporating historical location information, the initialization speed can be significantly improved when the vehicle has not moved between the current power-on and the previous power-off.

[0067] refer to Figure 3 and Figure 4The point cloud and map matching module primarily evaluates the matching degree between real-time laser point clouds and high-precision maps under different particle poses, providing a basis for particle weight calculation. In complex environments such as factories, maps typically contain fixed structures like pillars and walls, as well as dynamic interference objects such as cranes and hoists. Its processing flow includes: (1) Downsampling, distortion correction, and stitching are performed on the lidar data to generate real-time online point clouds; (2) Based on the candidate poses represented by particles (the candidate poses are generated by multi-source location information constraints dominated by 6G), the online point cloud is transformed from the vehicle system to the global map coordinate system; (3) Match the converted point cloud with the high-precision map. The matching methods include, but are not limited to, NDT, ICP or GICP. (4) Calculate the matching score for each particle based on the matching results. The score is used to characterize the confidence level of the particle's pose.

[0068] Through the point cloud matching process described above, the point cloud matching module calculates a matching score for each candidate particle pose and outputs the matching score as the particle weight or the basis for weight calculation to the particle filtering inference module. This matching score comprehensively reflects the consistency between the real-time point cloud and the high-precision map in spatial structure under the current particle pose. It can effectively distinguish between high-confidence particles located in the correct position area and low-confidence particles deviating from the map structure, thus providing a reliable basis for subsequent particle resampling, convergence judgment, and global initialization pose estimation modules. Especially in complex environments such as factories, by utilizing stable structural features such as columns and walls for matching evaluation, it can maintain the ability to distinguish the correct initial pose even in the presence of dynamic interference objects such as cranes and hoists, improving the stability and accuracy of the global initialization process.

[0069] refer to Figure 3 and Figure 4 The particle filter pose inference module is used to iteratively evaluate, filter, and determine convergence of particles under the constraint of candidate position information. Its processing flow includes: (1) Particle generation and propagation During the initialization phase, the relationship between candidate position information and corresponding particle distribution is as follows: Figure 6 As shown, a smaller variance indicates a stronger spatial prior constraint on the candidate pose. This embodiment uses 6G positioning information as a high-confidence global coarse positioning prior. Based on different candidate position information sources, it generates particle sets with different variance characteristics. For example, low-variance particles are generated based on 6G positioning information, medium-variance particles are generated based on fixed GNSS solutions or historical position information that has not moved, and high-variance particles are generated based on floating-point GNSS solutions or historical position information with small-scale movement. The number and distribution range of particles are controlled through an adaptive accuracy mechanism.

[0070] (2) Particle scoring and weight update As shown in Table 1, point cloud matching calculations are performed for each particle, and the matching score is normalized to the particle weight, so that the sum of the particle weights is 1.

[0071] Table 1. Matching Score Examples

[0072] (3) Resampling and convergence judgment Resampling is performed based on the particle weight distribution, retaining high-weight particles and discarding low-weight particles. Initialization is considered complete when a particle meets at least one of the following convergence conditions: The particles are highly concentrated in space; The particle variance is less than a preset threshold; The matching score remained stable across multiple consecutive frames.

[0073] Through the aforementioned particle generation, scoring, resampling, and convergence judgment processes, the particle filtering inference module, constrained by multi-source candidate location information, gradually eliminates low-confidence particles that do not conform to the map structure and retains a set of particles with a high degree of matching with the high-precision map. When the particle set meets the preset convergence condition, the module outputs a set of high-weight particles with stable spatial distribution and concentrated weights, which serve as the input for subsequent high-weight particle processing and initialization pose estimation units. By introducing an iterative screening and adaptive convergence mechanism based on matching scores, the particle filtering pose inference module, constrained by multi-source candidate location information (especially the spatial prior provided by 6G positioning), can effectively reduce the global initialization search space in complex environments such as factories, avoid the computational overhead caused by global blind search, and improve the robustness and success rate of the initialization process.

[0074] refer to Figure 4 The high-weighted particle processing module generates a convergent particle set based on candidate position information primarily driven by 6G positioning. This module further filters and statistically analyzes the particle set to obtain globally initialized pose information with consistent spatial distribution and stable statistical characteristics. Its processing flow includes: (1) Sort the converged particle set according to the particle weight, and select a number of particles with the highest weight ranking to form a high-weight particle candidate set. (2) Statistical analysis of the pose distribution of high-weight particles is performed to calculate their spatial mean, variance and weight distribution characteristics, which are used to characterize the concentration and stability of the pose. (3) In order to avoid the adverse effects of a single abnormal particle on the results, the pose is smoothed and the consistency is verified by combining the neighborhood distribution characteristics and covariance information of high-weight particles in the pose space. (4) When the statistical confidence of the high-weight particle set is insufficient or the pose distribution dispersion exceeds the preset threshold, the second-highest weight particle is introduced to participate in the statistical calculation in order to improve the stability and robustness of the pose result.

[0075] Through the above processing, the high-weighted particle processing module outputs a highly reliable and uniformly distributed global initial pose result, which serves as the input for laser positioning initialization, enabling it to converge quickly.

[0076] refer to Figure 4 The laser positioning initialization module is used to input the globally initialized pose information generated from the 6G dominant candidate information and processed by high-weighted particles into the laser positioning system as its initial pose, in order to initiate the subsequent fine positioning process based on the laser point cloud. Its processing flow includes: (1) Receive the global initial pose information output by the high-weight particle processing module, wherein the pose includes the vehicle's position and attitude information in the map coordinate system; (2) Perform basic validity and confidence checks on the global initial pose. When the pose meets the preset consistency conditions, use it as the initial pose for laser positioning. (3) Inject the initial pose into the laser positioning module to initialize the matching relationship between the point cloud and the high-precision map; (4) Under the initial pose constraint, the laser positioning module begins to perform continuous point cloud matching and pose update to achieve stable positioning of the vehicle in the factory environment.

[0077] By utilizing the high-quality initial pose provided during the global initialization phase, laser positioning avoids global search from a state without prior knowledge, thereby significantly improving the positioning convergence speed and stability.

[0078] In summary, this invention uses 6G positioning information as the core global prior constraint, combined with GNSS positioning information and historical location information, to construct a global initialization method for complex environments such as factories, ports, and logistics warehouses. Based on the accuracy characteristics of 6G positioning information, the method adaptively adjusts the particle generation and distribution range, and combines the matching results of laser point clouds and high-precision maps to iteratively evaluate, resample, and determine convergence of particles, thereby obtaining a stable and reliable global initial pose. By integrating the global initial pose into the laser positioning system, rapid initialization and smooth startup of laser positioning are achieved, avoiding the problem of large-scale global search without prior knowledge. This invention can significantly improve the success rate and convergence speed of positioning initialization in complex environments, especially in scenarios with unstable or severely obstructed GNSS signals. Relying on 6G-dominated global prior information enhances the adaptability and robustness of the system, demonstrating significant engineering application value.

[0079] In another embodiment of the invention, a device is also included, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments.

[0080] In addition, in conjunction with the global initialization method for autonomous vehicles in the above embodiments, embodiments of the present invention can provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments.

[0081] This invention constrains and guides the particle filter initialization process by introducing 6G positioning information, achieving the following technical effects: (1) Positioning accuracy is significantly improved By introducing 6G positioning information, which possesses high spatiotemporal continuity and availability, as an important global constraint source in the particle filter initialization stage, the global initialization reference is expanded from the traditional combination of historical pose and GNSS to a multi-source candidate pose system combining 6G positioning information, historical pose, and GNSS. This significantly improves the spatial rationality and accuracy of the initial particle seeding. At the same time, the particle set generated under the constraint of 6G positioning information can more quickly form a consistent solution with the lidar point cloud matching results. Through the joint iteration of particle filtering and point cloud matching, the initial pose can be rapidly converged, effectively reducing the global initialization error in complex environments such as closed or semi-closed environments.

[0082] (2) The initialization speed is significantly faster. Using the accuracy and confidence level of 6G positioning information as the main basis for adaptive adjustment of particle dispersal range, the search space is constrained during the particle generation stage, so that particles are concentrated in high-confidence areas, reducing the number of invalid particles and accelerating the convergence speed of particle filtering. Compared with schemes that rely solely on GNSS or historical poses for initialization, this invention avoids the problem of large-scale search when GNSS is unavailable or historical poses fail by using 6G-assisted particle filtering, significantly shortening the global initialization time and improving the start-up efficiency of autonomous vehicles in complex environments.

[0083] (3) Enhanced system robustness and availability Even when GNSS signals are blocked or interfered with, and the historical location information stored locally on the vehicle is unavailable, 6G positioning information can still participate in the particle filtering process as an independent global initialization constraint source, ensuring the continuous availability of the global initialization function in complex environments. Through the multi-source information fusion mechanism of particle filtering, guided by 6G positioning information, and combined with particle weight update, resampling, and high-weight particle fusion strategies, the impact of single information source errors on the initialization results is effectively suppressed, improving the stability and robustness of the global initialization system.

[0084] It should be noted that the execution order of the above steps is determined by their inherent logic and function. The execution order is sufficient to achieve the desired result of the technical solution disclosed in this invention, and should not impose any limitations or constraints on the implementation of this invention and its embodiments. In addition to the above embodiments, this invention may have other implementation methods. All technical solutions formed by equivalent substitution or equivalent transformation fall within the protection scope claimed by this invention.

Claims

1. A global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments, characterized in that, Includes the following steps: Step 1: Obtain multi-source location information of the autonomous vehicle. The multi-source location information includes at least 6G positioning information, GNSS positioning information, and historical location information stored locally by the vehicle. Step 2: Using the 6G positioning information as the main global initialization constraint source, construct a candidate pose set by combining the GNSS positioning information and historical location information; based on the accuracy and reliability of each candidate pose, adaptively set the corresponding particle initialization distribution range to generate an initial particle set spatially constrained by the 6G positioning information. Step 3: Match the real-time point cloud of the LiDAR with the high-precision map to obtain the matching score of each particle in the initial particle set and update the weight; through resampling and iterative convergence judgment, select the high-weight particle set. Step 4: Perform statistical analysis and weighted fusion on the high-weight particle set to generate a global initial pose containing the vehicle's position and attitude, and use it as the initial input of the laser positioning system to achieve global initialization of autonomous vehicles in complex environments.

2. The global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments according to claim 1, characterized in that, In step 2, the 6G positioning information is used as the primary global initialization constraint source, and a candidate pose set is constructed by combining the GNSS positioning information and historical location information, including: Receive raw location information sent from the 6G network side, the information including at least spatial coordinates, timestamp and location quality parameters; The validity and stability of the raw 6G positioning information are judged, and positioning data with timed-out timestamps or abnormal consecutive frame jumps are removed. The effective 6G positioning information is transformed from its original coordinate system to a global coordinate system consistent with the high-precision map, and its positioning accuracy and reliability are evaluated based on the channel status or signal angle of arrival parameters. When 6G positioning information is available and stable, it is set as the highest priority candidate pose information to dominate the spatial distribution center of subsequent particles. The GNSS positioning information is used to determine the solution state. If it is a fixed solution, it is used as a medium variance constraint source to generate a medium-range particle cloud; if it is a floating-point solution, it is used as a large variance constraint source to generate a large-range particle cloud. The historical location information is used to determine consistency. If the vehicle does not move significantly, it is used as a medium variance constraint source to generate a medium-range particle cloud; if the vehicle moves slightly, it is used as a large variance constraint source to generate a large-range particle cloud. The processed 6G positioning information, GNSS positioning information, and historical location information are sorted according to priority, with 6G positioning information having the highest priority and used to guide the distribution of particles within a small spatial range. GNSS positioning information and historical location information are used as supplementary candidates to enhance global search capabilities.

3. The global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments according to claim 1, characterized in that, In step 2, the corresponding particle initialization distribution range is adaptively set, including: Establish a mapping relationship between candidate pose accuracy and particle dispersion variance, where a smaller particle dispersion range is set for candidate poses with higher accuracy, and a larger particle dispersion range is set for candidate poses with lower accuracy. When the candidate pose is high-precision 6G positioning information, a low-variance particle cloud is generated, with the smallest particle distribution range, which is used to accurately constrain the global initialization search space. When the candidate pose is a fixed GNSS solution or historical location information that has not moved, a particle cloud with moderate variance is generated, with a moderate particle distribution range, to help verify the accuracy of 6G positioning information. When the candidate pose is a GNSS floating-point solution or historical position information with a small range of movement, a particle cloud with large variance is generated, which has the largest particle distribution range. This is used to enhance the global search capability and avoid getting trapped in local optima.

4. The global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments according to claim 1, characterized in that, In step 3, the matching score calculation for each particle in the initial particle set includes: Based on the candidate poses represented by particles, the real-time LiDAR point cloud is transformed from the vehicle system to the global map coordinate system; The normal distribution transformation, iterative nearest point, or generalized iterative nearest point algorithm are used to calculate the degree of matching between the transformed point cloud and the fixed structural features in the high-precision map. The matching degree is normalized into particle weights, so that particles with high consistency with the map structure receive high weights, while particles that match dynamic interference objects or deviate from the map structure receive low weights.

5. The global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments according to claim 1, characterized in that, In step 3, the iterative convergence determination includes: Monitor the spatial distribution variance of the particle set after resampling. When the variance is less than a preset threshold, it is determined that convergence has been achieved. Alternatively, monitor the stability of particle matching scores across multiple consecutive frames, and determine that the particles have converged when the fluctuation in matching scores is less than a preset range. If any of the above convergence conditions are met, the iteration stops and the current set of high-weight particles is output; otherwise, the next round of particle scoring, weight updating, and resampling iterations continues.

6. The global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments according to claim 1, characterized in that, Step 4 involves statistical analysis and weighted fusion of the high-weight particle set, including: The converged particles are sorted according to their weights, and the top-ranked particles are selected to form a high-weight candidate set. The spatial mean and covariance of the high-weight candidate set are calculated, and smoothing is performed in combination with the neighborhood distribution characteristics. The spatial mean and covariance matrix are used to characterize the concentration and stability of the pose. When the statistical confidence of the high-weight particle set is insufficient or the dispersion of the pose distribution exceeds a preset threshold, the second-highest weight particle is introduced to participate in the statistical calculation or the initialization is judged to have failed, so as to improve the stability and robustness of the output pose.

7. The global initialization method for autonomous vehicles based on 6G dominant constraints in complex environments according to claim 1, characterized in that, In step 4, its global initial pose is used as the initial input to the laser positioning system, including: Receive global initial pose information output by the high-weighted particle processing module, wherein the pose includes the vehicle's position and attitude information in the map coordinate system; The global initial pose is subjected to basic validity and confidence verification. When the pose meets the preset consistency condition, it is injected into the laser positioning module as the initial guess value of the point cloud matching algorithm. Under the initial pose constraint, the laser positioning module performs continuous point cloud matching and pose update within a local area to achieve stable positioning of the vehicle in the factory environment.

8. A global initialization system for autonomous vehicles based on 6G dominant constraints in complex environments, characterized in that, include: The system includes a candidate location information acquisition and filtering module, a point cloud and map matching module, a particle filter pose inference module, a high-weight particle processing module, and a laser positioning initialization module. The candidate location information acquisition and filtering module is used to acquire multi-source location information, which includes at least 6G positioning information, GNSS positioning information, and historical location information. It also performs availability assessment and priority ranking on the multi-source location information to generate a candidate pose set dominated by 6G. The point cloud and map matching module is used to receive real-time LiDAR point cloud data of the vehicle and match and score it with a high-precision map. The particle filtering pose inference module is used to generate initial particles based on the candidate pose set, iteratively update the particle weights based on the matching results of the lidar point cloud and the high-precision map, and filter out high-weight particles. The high-weighted particle processing module is used to perform statistical analysis and weighted fusion on the converged high-weighted particle set to generate a global initial pose. The laser positioning initialization module is used to receive the global initial pose and input it as an initial value into the laser positioning system to achieve rapid global positioning of the vehicle.

9. A device, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of a global initialization method for an autonomous vehicle based on 6G dominant constraints in a complex environment as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of a global initialization method for an autonomous vehicle based on 6G dominant constraints in a complex environment as described in any one of claims 1 to 7.

Images