A multi-objective dynamic path planning method and system for a logistics vehicle

Abstract

This invention relates to the field of intelligent transportation and logistics technology, and discloses a multi-objective dynamic path planning method and system for logistics vehicles. The method includes collecting data on the vehicle's surrounding environment and real-time location coordinates, performing time-series alignment and anomaly removal processing; generating lane availability mapping based on road feature recognition and traffic status verification, and fusing roadside traffic signal status to obtain a real-time traffic condition feature set; performing global path search and multi-dimensional optimization to obtain a macro-path sequence, dynamically adjusting to generate lane-level path planning; generating a vehicle driving operation sequence through dynamic traffic scenario deduction, and obtaining the final path instruction by real-time monitoring of trajectory deviation. This invention effectively solves the problem that existing technologies lack the ability to adaptively handle complex dynamic environments and cannot effectively distinguish between static road information and real-time traffic condition changes.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and logistics technology, and in particular to a method and system for multi-objective dynamic path planning for logistics vehicles. Background Technology

[0002] With the development of intelligent transportation technology and the increasing demand for logistics transportation, vehicle route planning systems have been widely applied in key scenarios such as urban distribution, trunk transportation, and park scheduling. As the core carrier combining cargo transportation and distribution, logistics vehicles, combined with high-precision positioning and vehicle networking technology, achieve intelligent scheduling. Users have placed higher demands on the accuracy of their planning, dynamic adaptability, and transportation efficiency, which directly affects the operational efficiency and service quality of logistics companies.

[0003] Existing route planning methods for logistics vehicles mostly rely on static, fixed road network data for one-time route planning, using distance or theoretical travel time as the sole optimization objective. Furthermore, most methods only perform global coarse planning, lacking detailed lane-level status analysis and mechanisms for real-time data updates and local replanning. Consequently, the system cannot dynamically distinguish between static road attributes and real-time traffic flow or unexpected conditions. Road condition updates are delayed, and planned routes deviate from actual driving scenarios, ultimately resulting in significant planning deviations and untimely responses to unexpected road conditions, further reducing overall logistics efficiency and delivery on-time rates. Summary of the Invention

[0004] This invention provides a multi-objective dynamic path planning method and system for logistics vehicles to solve the problems of existing technologies lacking the ability to adapt to complex dynamic environments, failing to effectively distinguish between static road information and real-time road condition changes, resulting in large planning deviations and untimely responses, and reducing the utilization efficiency and delivery timeliness of logistics transportation systems.

[0005] Firstly, to address the aforementioned technical problems, this invention provides a multi-objective dynamic path planning method for logistics vehicles, comprising: Collect vehicle surrounding environment data and real-time location coordinates, perform time-series alignment and anomaly removal on the surrounding environment data and real-time location coordinates, and obtain a preliminary dynamic road dataset. Based on the preliminary dynamic road dataset, road feature recognition and traffic status verification are performed to obtain lane availability mapping; Based on the lane availability mapping and the roadside-transmitted traffic signal status, path calculation is paused when the traffic signal status is prohibited, and a real-time traffic feature set is obtained. Based on the real-time traffic feature set, a global path search and multi-dimensional optimization process are performed to obtain an optimized macro-path sequence. Local road segment information is extracted based on the macro-path sequence and dynamically adjusted in combination with real-time traffic information. When a road segment obstacle is detected, the local path is replanned to obtain a refined lane-level path plan. Based on the lane-level path planning and dynamic traffic scenario simulation, speed control commands and lane operation commands for vehicle driving are generated, and the speed control commands and lane operation commands are fused to obtain a lane-level operation sequence. The vehicle is controlled to drive in real time according to the lane-level operation sequence and the driving trajectory deviation is monitored. When the trajectory deviation exceeds the preset deviation threshold, the real-time traffic information is re-integrated and the planning is updated to obtain the final path instruction.

[0006] Secondly, the present invention provides a multi-objective dynamic path planning system for logistics vehicles, comprising: The data acquisition and processing module is used to collect vehicle surrounding environment data and real-time location coordinates, and to perform time-series alignment and anomaly removal processing on the surrounding environment data and real-time location coordinates to obtain a preliminary dynamic road dataset. The lane status verification module performs road feature recognition and traffic status verification processing based on the preliminary dynamic road dataset to obtain a lane availability mapping. The traffic signal fusion module fuses the traffic signal status transmitted from the roadside according to the lane availability mapping, and suspends path calculation when the traffic signal status is prohibited from passing, thereby obtaining a real-time traffic feature set; The macro-path planning module performs global path search and multi-dimensional optimization processing based on the real-time traffic feature set to obtain an optimized macro-path sequence. The local path refinement module extracts local road segment information based on the macro path sequence and makes dynamic adjustments in combination with real-time traffic information. When a road segment obstacle is detected, the local path is replanned to obtain a refined lane-level path plan. The driving instruction generation module generates speed control instructions and lane operation instructions for vehicle driving based on the lane-level path planning and dynamic traffic scenario deduction, and merges the speed control instructions and lane operation instructions to obtain a lane-level operation sequence. The path execution correction module controls the vehicle to drive in real time and monitors the driving trajectory deviation according to the lane-level operation sequence. When the trajectory deviation exceeds a preset deviation threshold, the real-time traffic information is re-integrated and the planning is updated to obtain the final path instruction.

[0007] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention collects multi-source data through vehicle-mounted sensing devices and satellite positioning, and generates high-precision lane availability mapping by combining image recognition technology, thereby realizing the accurate acquisition and dynamic updating of real-time road information.

[0008] (2) By integrating traffic signal status and real-time road condition characteristics, the present invention dynamically adjusts the timing of path calculation, thereby achieving global optimization of macro-path and adaptive adjustment of local road segments.

[0009] (3) This invention generates an executable vehicle operation sequence by combining lane-level path planning with dynamic traffic scenario simulation, thereby achieving precise control of logistics vehicle driving and real-time correction of path deviation. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of a multi-objective dynamic path planning method for logistics vehicles provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a multi-objective dynamic path planning system for logistics vehicles provided in the second embodiment of the present invention. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] In step S101, the vehicle's surrounding environment data and real-time location coordinates are collected. Temporal alignment and anomaly removal are performed on the surrounding environment data and real-time location coordinates to obtain a preliminary dynamic road dataset, including: The system collects obstacle data, road image data, and real-time location coordinates around the vehicle using vehicle-mounted lidar, cameras, and satellite positioning devices. The obstacle data, road image data, and real-time location coordinates are then time-series aligned according to the collection timestamps to obtain aligned multi-source data. Outlier removal is performed on the aligned multi-source data to remove invalid data that exceeds the preset value range, and lane layout information and temporary closed area information are extracted from the road image data. By integrating real-time location coordinates, lane layout information, and temporary closure area information, a preliminary dynamic road dataset is obtained.

[0013] It should be noted that the vehicle-mounted LiDAR, cameras, and satellite positioning devices collect data synchronously, maintaining a unified time reference throughout the collection process. Obstacle data, road image data, and real-time location coordinates are aligned point-by-point according to the collection timestamp, forming aligned multi-source data with consistent time dimensions. The unified time reference is provided by the vehicle control system, and the time accuracy meets the requirements for synchronous multi-source data collection.

[0014] It should be noted that the anomaly removal process includes two stages: sensor effective range verification and data quality anomaly detection.

[0015] The first stage involves validating the sensor's effective range. All sampling points in the aligned multi-source data are traversed, and data items that are invalid due to sensor physical limitations or are explicitly invalid are removed: For LiDAR, the effective obstacle distance range is 0.1–100 meters (sensor range). Sampling points with a distance <0.1 meters (blind zone) or >100 meters (exceeding the maximum detection distance) are marked as invalid and removed. For cameras, the valid pixel brightness range is 0–255 (complete dynamic range of an 8-bit grayscale image). No brightness values ​​are removed in this stage; only this range is recorded for subsequent anomaly detection input validity checks. For satellite positioning, coordinate points with a horizontal positioning error (parsed from NMEA statements) ≥10 meters are marked as invalid and removed. This threshold is determined based on a 1.5 times margin of typical accuracy for civilian single-point positioning (CEP 95% ≤ 5–8 meters).

[0016] This stage only removes data points that are physically impossible or clearly invalid by the sensors, and does not involve any statistically significant anomaly judgments.

[0017] The second stage involves data quality anomaly detection. For valid data that passed the first stage verification, a depth anomaly detection based on statistical characteristics and physical kinematic constraints is further performed: LiDAR outlier filtering employs a statistical outlier filtering method. For each point, the average distance to all neighboring points (with a neighborhood radius of 0.5 meters) is calculated. Assuming the obtained average distance distribution follows a Gaussian distribution, the global mean and standard deviation are calculated. Outliers exceeding the average distance threshold are then excluded from the detection process. Points marked as outliers are removed. The threshold was determined by statistical analysis of LiDAR measurement data from 100 different scenarios (urban roads, highways, and industrial parks), and it can remove about 95% of isolated noise while retaining effective obstacle information.

[0018] In addition, a temporal consistency check is performed on the lidar point cloud: if the depth change of the same spatial location point exceeds 5 meters between two consecutive frames (0.1 seconds apart) (corresponding to a displacement of approximately 180 km / h for a vehicle), it is considered an abnormal jump and is removed. This threshold is determined based on a margin of 3 times the maximum driving speed of logistics vehicles (80 km / h) and the maximum reasonable displacement between two frames (approximately 2.2 meters).

[0019] Secondly, camera image quality is detected. For all-black and all-white detection, the global grayscale mean of the image is calculated. If the mean is <10 or >245, and effective lane line edge responses cannot be extracted (the number of Canny edge points is <0.1% of the total number of image pixels), it is judged as an invalid frame and discarded. The thresholds of 10 and 245 are determined by statistical histograms of 1000 normal, overexposed, and underexposed road image samples, which can cover 99% of normal lighting scenes. Structural similarity detection is performed by calculating the structural similarity index (SSIM) between two adjacent frames. If the SSIM is lower than 0.3, it indicates that the image content has undergone discontinuous abrupt changes (such as the camera being momentarily illuminated or blocked by strong light), and the frame is marked as abnormal and discarded. The SSIM threshold of 0.3 was determined through 500 sets of experiments on image changes caused by bumps and sudden changes in lighting during normal driving. Under this threshold, the pass rate of normal frames is 98%, and the recognition rate of abnormal frames is 92%.

[0020] Finally, there is satellite positioning kinematic constraint detection, specifically for position jump detection, based on the position from the previous timestamp. ,speed and time interval Predict the current position If the actual measured location The distance from the predicted position is greater than the sum of the vehicle's maximum reasonable displacement and the positioning error tolerance, i.e. ,in (90km / h) If the error is more than 3 times the GPS positioning error, it is considered an abnormal jump.

[0021] Speed ​​jump detection, calculation of instantaneous speed ,like ,in If the maximum braking acceleration of the logistics vehicle is 1.5 times the margin, it is considered abnormal. Based on the industry standard for the maximum deceleration of fully loaded vehicles, approximately... And consider the GPS noise amplification effect.

[0022] All data retained after the two-stage detection are marked as valid data and used for subsequent construction of the initial dynamic road dataset.

[0023] It should be noted that the lane layout information is extracted directly from the road image data and includes lane boundaries, number of lanes, and lane direction. The temporary closure area information is obtained by identifying closure signs and roadblock locations from the road image data and includes the closure range and spatial coordinates.

[0024] It is worth noting that real-time location coordinates, lane layout information, and temporary closure area information are sequentially integrated according to the collection time sequence to form a structured preliminary dynamic road dataset. The format of this preliminary dynamic road dataset is consistent with the input format for subsequent road feature recognition and traffic status verification, and can be directly used for subsequent processing steps.

[0025] In step S102, road feature identification and traffic status verification are performed based on the preliminary dynamic road dataset to obtain lane availability mapping, including: Based on the preliminary dynamic road dataset, road image sequences are extracted, and edge feature extraction processing is performed on the road image sequences to obtain lane line edge contour data; The lane boundary position is determined based on the lane line edge contour data, and the initial traffic status of each lane is marked by combining the temporary closed area information in the preliminary dynamic road dataset. The lane boundary location and the initial traffic status are compared and verified with high-precision map data in the cloud. After correcting the deviation, a lane availability mapping containing lane number and corresponding traffic status is generated.

[0026] It should be noted that the road image sequence is extracted from the preliminary dynamic road dataset in chronological order of acquisition timestamps, with each frame corresponding to a timestamp, forming a continuous road image sequence. Edge feature extraction processing performs grayscale conversion on each road image frame, extracting edge contour information to obtain lane line edge contour data. The edge intensity threshold is set to 100 by default and can be automatically adjusted according to light intensity: 120 for sunny days with sufficient light, 80 for cloudy days with insufficient light, and 150 for rainy days with glare. This edge intensity threshold was determined statistically from 500 sets of urban road image samples under different lighting and weather conditions, covering common driving scenarios such as sunny, cloudy, and rainy days.

[0027] The following multi-stage processing is performed on each frame of road image to cope with complex scenes such as changes in lighting, shadows, wear, and stains.

[0028] The RGB image was converted to grayscale using the formula Gray = 0.299·R + 0.587·G + 0.114·B. Adaptive histogram equalization (CLAHE) was used to enhance image contrast. The block size was set to 8×8 pixels, and the contrast limit was 2.0. These parameters were determined through contrast enhancement experiments on 100 sets of low-light and shadowed road images, and can improve the grayscale difference between lane lines and road surface by at least 30% without amplifying noise. Gaussian filtering was applied to remove high-frequency noise, with a kernel size of 5×5 and a standard deviation σ = 1.0. This parameter strikes a balance between maintaining lane line edge sharpness (loss <5%) and noise suppression (signal-to-noise ratio improvement of 12dB).

[0029] The Canny edge detection algorithm is used instead of the static thresholding method. The Canny algorithm includes two adaptive thresholds: a low threshold T_low and a high threshold T_high. A histogram of the image's gray-level gradient is calculated, and the gradient magnitudes are sorted from high to low. The top 70% quantile is taken as T_high, where T_low = 0.5 × T_high. This 70% quantile is determined statistically from 500 road images of different scenes, ensuring continuous edge extraction even in low-contrast scenes (lane line gray-level difference < 15) while suppressing road texture noise. The Sobel operator kernel size in the Canny algorithm is 3×3, and the gradient calculation uses the L2 norm. The non-maximum suppression window is 3×3 pixels.

[0030] In a cloudy, low-light scenario, the 70th percentile of the gradient magnitude is 35, so T_high = 35 and T_low = 17.5; in a sunny, high-light scenario, this percentile is 65, so T_high = 65 and T_low = 32.5. For the edge image output by Canny, Hough Transform is applied to extract candidate line segments. The Hough Transform parameters are set as follows: accumulator threshold of 30 (at least 30 edge points constitute candidate lines), minimum line segment length of 40 pixels, and maximum line segment gap of 20 pixels. These parameters are determined based on the projection geometry of the vehicle camera resolution (1280×720), actual lane width (10-15cm), and shooting distance (10-30 meters).

[0031] Candidate lines are selected based on lane line geometric constraints. Straight line angle constraint: The angle relative to the horizontal direction of the image should be between 30° and 150° (excluding horizontal or approximately horizontal lines); Length constraint: The projected length of the line segment in the image should not be less than 50 pixels (corresponding to an actual road length of approximately 5-8 meters); Width constraint: The horizontal distance between the left and right edges of the same lane (in the image) should be between 40-180 pixels, corresponding to an actual lane width of 2.5-3.75 meters.

[0032] A sliding window method is used to search for lane boundaries row by row, starting from the bottom of the image (closest to vehicles). The window width is set to 50 pixels. Within each window, the weighted average position of edge points is calculated (weights are considered as gradient magnitudes) to obtain the candidate lane center points. Then, the window slides upwards vertically in 10-pixel steps until it reaches the top of the image. For each lane, the sequence of window center points is fitted using a quadratic polynomial: x = A·y² + B·y + C, where y is the image row coordinate (increasing from top to bottom), and x is the column coordinate. The quadratic polynomial can adapt to changes in curves and lane curvature.

[0033] Kalman filtering was applied to the fitted lane line parameters (A, B, C) of consecutive video frames. The prediction model was a uniform motion model with a state vector of [A, B, C, dA / dt, dB / dt, dC / dt]. The diagonal of the measurement noise covariance matrix was set to 0.1, and the diagonal of the process noise covariance matrix was set to 0.05. The parameters were calibrated through comparative experiments on 50 sets of measured urban road sequences (including curves, straight sections, and lane line wear sections). The parameters can maintain smooth tracking when lane lines are briefly lost (e.g., obscured by the vehicle in front), with a frame loss recovery time of less than 0.3 seconds.

[0034] It should be noted that the lane boundary positions are obtained by fitting lane line edge contour data. The fitting process involves matching edge points line by line along the longitudinal direction of the image to determine the left and right boundary coordinates of each lane. The initial traffic status marking is completed by combining the temporary closure area information in the preliminary dynamic road dataset. When the coordinates of a lane area overlap with those of a temporary closure area for more than 30 seconds, the lane is marked as impassable. The 30-second duration threshold was determined by statistically analyzing 200 sets of measured data from temporary construction and accident road closure scenarios, which can effectively distinguish between temporary obstacles and long-term closed areas.

[0035] It should be noted that the high-precision map data in the cloud includes static lane boundary positions, lane numbers, and lane direction information. The comparison and verification process involves comparing the extracted lane boundary positions point-by-point with the corresponding lane boundaries on the cloud map, calculating the positional deviation. When the positional deviation exceeds 0.2 meters, the lane boundary positions are corrected based on the high-precision map data in the cloud. This 0.2-meter deviation threshold was determined through statistical experiments comparing 1000 sets of vehicle-mounted sensing data with the cloud map, balancing the real-time performance of the sensing data with the accuracy of the map data.

[0036] It should be noted that lane availability mapping includes three core types of information: lane number, lane boundary coordinates, and traffic status. Traffic status is divided into two categories: passable and impassable, corresponding to different route planning permissions. The generation process arranges lane information in lane number order to ensure that the lane order is consistent with the actual road travel direction. Lane numbers start from the rightmost lane and increase sequentially, maintaining the same numbering rule as the high-precision cloud map.

[0037] It is worth noting that the lane availability mapping is structured and stored in the order of collection timestamps, with each mapping record corresponding to a timestamp. The storage interval is consistent with the update interval of the preliminary dynamic road dataset. The stored data is synchronously backed up to the onboard local storage unit and the cloud storage platform. The onboard local storage retains the mapping data for the most recent 7 days, and the cloud storage retains the mapping data for the most recent 180 days. The storage duration is determined through statistical analysis of the traceability needs of logistics vehicle operation data.

[0038] It should be noted that the information on temporary closed areas includes the type of closed area (construction, accident, temporary traffic control), the pixel mask of the closed area in the road image, and its 3D spatial coordinates in the world coordinate system.

[0039] A deep learning-based semantic segmentation model is used to perform pixel-level classification on each frame of road images, identifying temporary closure categories such as "construction cones," "barriers," "warning signs," and "accident vehicles." Specifically, the model architecture uses a DeepLabv3+ network with a ResNet-50 backbone, an output stride of 16, and dilated convolution rates of 6, 12, and 18 for the ASPP module. The training dataset uses the "construction" and "obstacle" categories from the BDD100K road scene dataset, and additionally collects and annotates 2000 images of temporary construction and accident scenes on Chinese urban roads (image resolution 1280×720, labeled categories include: cones, crash barriers, warning signs, accident vehicles, and road closure barriers). The training / validation / test sets are split in a 7:2:1 ratio. Weights are initialized using an ImageNet pre-trained model with a batch size of 8, an initial learning rate of 0.001 (using a multinomial decay strategy), a momentum of 0.9, and a weight decay of 0.0001. The training run consisted of 200 epochs, with 20 epochs ending early. The model outputs the classification confidence score for each pixel, with a confidence threshold of 0.7. Pixels below this threshold were excluded from further processing. This threshold was determined using the precision-recall curve of the validation set, selecting the threshold corresponding to the point with the highest F1 score (measured at 0.73, rounded down to 0.7). Post-processing employed connected component analysis to remove isolated noise regions smaller than 500 pixels (corresponding to an actual area of ​​approximately 0.2 square meters, effectively filtering out small-sized clutter).

[0040] The pixel coordinates of the detected closed regions in the image are converted to 3D positions in the vehicle coordinate system. This assumes the vehicle-mounted camera has been pre-calibrated with intrinsic parameters. Point coordinates The parameters are: height H = 1.5 meters above the ground, pitch angle θ = 5°, roll angle = 0°. For any pixel (u, v) in the image, the vehicle coordinates (X, Y, Z) of the corresponding ground point satisfy the following: assuming the ground is flat, the Z-axis is forward, the X-axis is right, and the Y-axis is upward. Based on the pinhole model and the road surface plane constraint (Y = -H), it can be derived that:

[0041] For each closed region, the median of the 3D coordinates of all its pixels is taken as the spatial location of the region, and the convex hull boundary is calculated as the influence range. In a certain frame of the image, a closed region consisting of three cones is detected. After transformation, the set of pixel coordinates is obtained as the world coordinate range X∈[-1.2m,1.8m], Z∈[15m,25m]. The spatial coordinates of this region are described as "centered at (0.3,20), with a front-to-back range of 10 meters and a left-to-right range of 3 meters".

[0042] The convex hull of the closed region in the world coordinate system is projected onto the lane-level grid of the high-precision map. If the minimum distance between the centerline of a lane and the convex hull of the closed region is less than half the lane width (1.85 meters), and the overlap length is greater than 5 meters, then the lane is considered to be affected by closure. Duration counter: If the affected lane is detected for 5 consecutive frames (approximately 0.5 seconds), and the duration exceeds 30 seconds, it is marked as impassable. The continuity requirement of 5 frames is used to avoid false detections in a single frame (such as brief pedestrian obstruction), and the 30-second threshold is used to distinguish between temporary stops (such as parking for unloading) and long-term closures.

[0043] In step S103, based on the lane availability mapping and the roadside-transmitted traffic signal status, path calculation is paused when the traffic signal status is prohibited, resulting in a real-time traffic feature set, including: The vehicle-to-everything (V2X) communication interface is used to obtain information on traffic signal status, remaining signal duration, and intersection number transmitted by roadside equipment. The traffic signal status, the remaining duration of the signal, and the lane availability mapping are weighted and fused to generate an intersection traffic condition feature vector. If the traffic signal status is prohibited from passing, the global path calculation is paused based on the remaining duration of the signal, and the current traffic restriction level of the intersection is determined based on the intersection traffic condition feature vector; The intersection traffic condition feature vector and the traffic restriction level are integrated to obtain a real-time traffic feature set.

[0044] It should be noted that the vehicle-to-everything (V2X) communication interface receives data transmitted from roadside equipment once per second. When the vehicle is less than 50 meters from the intersection, the communication frequency automatically increases to twice per second. The data includes traffic signal status, remaining signal duration, and intersection number. This communication frequency is determined based on statistical data of urban intersection traffic signal update cycles, covering the phase change frequencies of common traffic signals. A timestamp is added to each data packet during the reception process, with a timestamp accuracy in milliseconds.

[0045] It should be noted that the weighted fusion process first quantizes the discrete state values: traffic signal status (allowing passage) is quantized as 1, and prohibiting passage as 0; lane availability mapping (available) is quantized as 1, and unavailable as 0. Then, a weight of 0.6 is assigned to the traffic signal status, and a weight of 0.4 is assigned to the lane availability mapping, calculating the intersection traffic condition value. The sum of these two weights is 1. The weight values ​​are determined through statistical analysis of 500 sets of measured data from different intersection traffic scenarios, covering different traffic directions such as straight, left turn, and right turn, as well as different road conditions such as smooth traffic, slow traffic, and congestion. The final weight allocation is determined using traffic efficiency and route planning rationality as evaluation indicators.

[0046] It should be noted that the global path calculation pause duration is the remaining signal duration plus 0.5 seconds. This 0.5-second increment is determined through statistics on the vehicle control system's response time, covering the startup delay of the path calculation module and data transmission delay. During the pause, data reception from roadside equipment continues, and the remaining signal duration is updated in real time. If the signal status switches to allow passage ahead of schedule, global path calculation immediately resumes.

[0047] It should be noted that traffic restriction levels are divided into three levels: Level 1, Level 2, and Level 3. Traffic restriction values ​​are calculated based on the intersection's traffic condition feature vector. A restriction value greater than or equal to 0.8 is classified as Level 1, greater than or equal to 0.5 but less than 0.8 as Level 2, and less than 0.5 as Level 3. Level 1 restrictions mean the intersection is completely impassable, requiring a replanned global route to bypass it. Level 2 restrictions mean the intersection's capacity is reduced by more than 50%, prioritizing alternative parallel intersections. Level 3 restrictions mean the intersection is passable, allowing travel along the original planned route. The thresholds of 0.8 and 0.5 were determined through statistical analysis of experimental data from 300 different traffic restriction scenarios, corresponding to different route planning adjustment strategies.

[0048] It should be noted that the real-time traffic feature set includes six categories of information: intersection number, traffic signal status, remaining signal duration, lane availability information, intersection traffic condition feature vector, and traffic restriction level. The generation process arranges each feature record in timestamp order, with the recording interval consistent with the data reception interval of the vehicle-to-everything (V2X) communication interface. The real-time traffic feature set is synchronously backed up to both the onboard local storage unit and the cloud storage platform. The onboard local storage retains the feature data for the most recent 7 days, while the cloud storage retains the feature data for the most recent 180 days. The storage duration is determined through statistical analysis of the logistics vehicle operation data traceability requirements.

[0049] For example, in an urban logistics delivery scenario, the vehicle-to-everything (V2X) communication interface receives data transmitted from roadside equipment once per second. When the vehicle is less than 50 meters from the intersection, the communication frequency automatically increases to twice per second. The current intersection number is identified as N012, the traffic signal status is "prohibited," and the remaining signal duration is 25 seconds. The traffic signal status, remaining signal duration, and lane availability mapping are weighted and fused, assigning a weight of 0.6 to the traffic signal status and 0.4 to the lane availability mapping, generating an intersection passage condition feature vector. Based on the remaining signal duration of 25 seconds, global path calculation is paused for 25.5 seconds, during which time data from roadside equipment is continuously received. The passage restriction value calculated based on the intersection passage condition feature vector is 0.85, which is greater than or equal to 0.8, determining the current intersection passage restriction level to be Level 1. The intersection number, traffic signal status, remaining signal duration, lane availability information, intersection passage condition feature vector, and Level 1 passage restriction level are integrated to generate a real-time traffic feature set, which is stored by timestamp in the vehicle's local storage and on the cloud platform.

[0050] In step S104, a global path search and multi-dimensional optimization process are performed based on the real-time traffic feature set to obtain an optimized macro-path sequence, including: Based on the real-time traffic feature set, urban roads are divided into intersection nodes and road segments, and a dynamic traffic weight is assigned to each road segment. The dynamic traffic weight is determined by a combination of the congestion level, length and estimated travel time of the corresponding road segment. Based on the dynamic access weight, the A* algorithm is used to perform a global path search and generate candidate global paths. Calculate a comprehensive score for each candidate global path based on average traffic efficiency and historical traffic stability, and select the candidate path with the highest comprehensive score as the macroscopic path sequence.

[0051] It should be noted that the urban road division process is based on the intersection numbering information in the real-time traffic feature set. Each intersection is marked as a node, and the road between two adjacent nodes is marked as a road segment. The node numbering is consistent with the intersection numbering in the real-time traffic feature set, and the road segment numbering is generated sequentially according to the node order, in the format "Node 1 number - Node 2 number". Each node contains three types of information: intersection number, intersection type, and turning rules. Each road segment contains four types of information: road segment number, road type, number of lanes, and speed limit. The resulting road network topology includes all passing intersections and corresponding road segments, covering all feasible paths from the starting point to the ending point, excluding road segments that are prohibited from passage and one-way reverse road segments.

[0052] For road sections Its dynamic passage weight The sum is obtained by weighting the three parts:

[0053] in, The length of the road segment is (km). The maximum segment length (km) in the current candidate path set; , Given by the modified Green Shields model (unit: h). Maximum estimated travel time (h); Dimensionless, value range [0,3]; weighting coefficient The sum of these coefficients is 1. These coefficients were calibrated through 1000 A* route-finding experiments under different road conditions to minimize the actual travel time of the planned route. It is a dimensionless scalar and can be directly used for cost accumulation in the A* algorithm.

[0054] The congestion level ranges from 0 to 1, with higher values ​​indicating more severe congestion. The congestion level is calculated from traffic flow density data in a real-time road condition feature set. Road segment length is in kilometers, and estimated travel time is in minutes, calculated by dividing the road segment length by the average speed of the road segment.

[0055] It should be noted that the traffic flow density data acquisition in this method does not rely on the private interface of an external urban traffic monitoring system. Instead, it uses vehicle-to-everything (V2X) communication and roadside unit (RSU) broadcast parsing to obtain real-time traffic flow density data. Specifically, vehicles receive BSM (Basic Safety Message) and SPAT (Signal Phase and Time) messages broadcast by roadside equipment via DSRC (Dedicated Short Range Communication, IEEE 802.11p) or C-V2X (Cellular Vehicle-to-Everything, 3GPPRelease 14 and above) protocols. The roadside equipment transmits the average number of vehicles and average speed in each lane within its sensing range every 100 milliseconds.

[0056] If roadside equipment is not covered or communication is interrupted, vehicle-mounted sensor fusion estimation is used: a forward-facing camera and lidar are used to detect vehicles within 100 meters ahead of the vehicle. Combined with the vehicle's speed, a macroscopic traffic flow model (such as the Green Shields model) is used to infer the average density of the road segment. Specifically, the inference formula is that density equals the current number of vehicles divided by the detection distance multiplied by the number of lanes, and an exponentially weighted moving average (smoothing factor 0.3) is used for time-series filtering.

[0057] To avoid the drawbacks of the linearly simplified model, a modified Greenshields model is adopted, which incorporates traffic flow density. Convert to speed And the passage time penalty factor.

[0058] Define the average driving speed of the road segment:

[0059] in, Free-flow speed (0.9 times the legal speed limit of the road, in km / h); Current density (vehicles / km / lane); For congestion density, we take 140 vehicles / km / lane (based on actual measurements on urban expressways). The nonlinear exponent is set to 1.5 (obtained by fitting 500 sets of measured density-speed data to better reflect the speed decline trend during moderate congestion). Then, the degree of congestion was redefined as the travel time amplification factor:

[0060] This coefficient is dimensionless and ranges from [0, 3]. Indicates no congestion. This indicates that travel time has doubled. The "Congestion Level" item in the dynamic traffic weighting actually takes... This is not the original density ratio. For example, if , Then the calculation yields , .

[0061] It should be noted that the global path search process starts at the starting node and ends at the ending node, using the A* algorithm for global path search. This method uses Euclidean distance divided by the maximum speed (i.e., the minimum possible travel time) as the heuristic function, rather than Manhattan distance. (Define the heuristic function.) To start from the current node Minimum possible travel time to the target node:

[0062] in, The straight-line Euclidean distance (in km) between the current node and the target node; The maximum legal speed limit for all road segments in the road network (take the speed limit of 80km / h for urban expressways).

[0063] Since the actual road length is at least as long as the straight-line distance, and the actual driving speed cannot exceed the maximum speed limit. Therefore, in reality from Shortest travel time to the destination Must be satisfied:

[0064] therefore It is a lower bound estimate of the actual cost, satisfying the admissibility condition. The A* algorithm guarantees finding the global optimum (given dynamic weights). Compared with other heuristic functions, the Manhattan distance... In non-grid road network layouts, straight-line distances are significantly overestimated, violating admissibility (e.g., the actual path length of an angled road may be less than the Manhattan distance), causing A* to potentially skip the optimal path. Therefore, this patent does not use the Manhattan distance. Euclidean distance is a safer choice, computationally efficient, and highly adaptable to irregular urban road networks.

[0065] To improve search speed, the Euclidean distance is pre-multiplied by a road curvature coefficient during actual calculation. (Obtained by statistically analyzing the average curvature of urban road network samples), resulting in a tighter heuristic function. The coefficient still satisfies the acceptability requirement because the actual path length is usually ≤1.2 times the straight-line distance.

[0066] It should be noted that the overall score calculation assigns a weight of 0.6 to average traffic efficiency and a weight of 0.4 to historical traffic stability, with the sum of the two weights being 1. Average traffic efficiency is calculated by dividing the total length of the candidate routes by the estimated total travel time, expressed in kilometers per hour.

[0067] This method establishes a historical route travel time record library in the vehicle's local storage unit and on the cloud platform, which is used to calculate the "historical travel stability" score for each candidate route. The specific construction method is as follows: After each trip, record the following fields: Route identifier (hash value of the sequence of origin-destination-intersections passed); Date and time period (divided into: morning peak 7:00-9:00, morning off-peak 9:00-12:00, noon 12:00-14:00, afternoon off-peak 14:00-17:00, evening peak 17:00-19:00, night 19:00-next day 7:00); Total actual travel time (seconds); Weather of the day (sunny / rainy / snowy / foggy); Whether any emergencies occurred (yes / no) Excessively long trips (travel time exceeding three standard deviations of the average travel time for the same period) caused by driver rest or abnormal stops are excluded; trip segments with GPS signal loss exceeding 10 seconds are also excluded; records with sudden accidents are marked as "atypical" and are not included in stability calculations by default, but are stored separately for training accident scenario models. Retain valid records from the most recent 180 days. Data older than 180 days is automatically deleted daily at 2:00 AM. For the same path-time combination, at least 10 valid records are required for stability calculation; if fewer than 10 records are found, the historical pass stability score is set to the default value of 0.5 (corresponding to moderate stability). Coefficient of Variation The historical stability score is defined as:

[0068] The denominator 0.5 is an empirical threshold (CV exceeding 0.5 indicates extreme instability), with a score range of [0.1, 1.0], which is then linearly mapped to the [0, 10] interval for comprehensive scoring. Travel times for different weather conditions and seasons have been controlled for using the above time period divisions and will not be further normalized. If there are insufficient records for a route on rainy days, the average value of sunny days in the same time period is multiplied by a weather coefficient (rainy days = 1.2, snowy days = 1.5) as an estimate.

[0069] The overall score ranges from 0 to 10, with higher values ​​indicating better overall route performance. The overall score is calculated by weighting two parts: the first part is the average traffic efficiency percentage, calculated by dividing the average traffic efficiency by the weighted average legal speed limit of the candidate route, then multiplying by 6; the second part is the historical traffic stability percentage, directly multiplied by 4. The weighted average legal speed limit of the candidate route is the average speed limit obtained by weighting the legal speed limits of each road segment according to their proportion to the total route length, expressed in kilometers per hour.

[0070] It should be noted that the macro-path sequence includes seven types of information: node number, road segment number, dynamic traffic weight, estimated travel time, comprehensive score, intersection turning instructions, and road segment speed limits. The generation process arranges the node and road segment information in the order of travel, ensuring that the path direction is consistent with the vehicle's travel direction. Intersection turning instructions are divided into four categories: straight, left turn, right turn, and U-turn, corresponding to different intersection traffic rules. Road segment speed limits are consistent with the legal speed limits for the corresponding road segments. The macro-path sequence is synchronously backed up to the vehicle's local storage unit and the cloud storage platform. The onboard local storage retains the current trip's path data, while the cloud storage retains all historical path data from the last 180 days. The storage duration is determined through logistics vehicle operation data analysis and traceability requirements, meeting the traceability requirements of logistics companies for vehicle travel trajectories.

[0071] For example, in an urban logistics delivery scenario, the starting node is N001 and the ending node is N010. Based on real-time traffic features, the route is divided into 9 intersection nodes and 9 road segments. Each node includes an intersection number, intersection type, and turning rules; each road segment includes a road segment number, road type, number of lanes, and speed limit. Dynamic traffic weights are assigned to each road segment. Road segment N001-N002 has a length of 1.2 kilometers, a congestion level of 0.3, and an estimated travel time of 2.4 minutes. The longest road segment in the current candidate path set is 2.4 kilometers long, with a longest estimated travel time of 4.8 minutes. After normalizing the length and time, the normalized value for both is 0.5. Substituting these values ​​into the dynamic traffic weight formula, we get: 0.3×0.5 + 0.5×0.5 + 0.2×0.3 = 0.46.

[0072] A global path search is performed based on dynamic traffic weights. The maximum path length is set to 1.5 times the shortest path length. After traversing all feasible paths, they are sorted by total weight, and the three candidate global paths with the lowest total weights are selected: Path 1 has a total weight of 18.3, Path 2 has a total weight of 20.1, and Path 3 has a total weight of 19.5. The comprehensive scores of the three candidate paths are calculated. Path 1 has a total length of 12.5 kilometers, an estimated total travel time of 30 minutes, an average traffic efficiency of 25 kilometers per hour, a standard deviation of actual travel time over the past 30 days of 2.5 minutes, and a historical traffic stability of 0.4. The weighted average legal speed limit for this path is 50 kilometers per hour, and the average actual travel time over the past 30 days is 30 minutes with a standard deviation of 2.5 minutes. The calculated coefficient of variation is 2.5 / 30 ≈ 0.083, and the historical traffic stability is 1 / 0.083 ≈ 12, which is 4.0 after normalization to the 0-10 range.

[0073] The overall score is calculated as 25 / 50×6 + 4.0×0.4 = 3.0 + 1.6 = 4.6; the average traffic efficiency of route 2 is 22 km / h, the historical traffic stability is 0.35, and the overall score is 3.6; the average traffic efficiency of route 3 is 23 km / h, the historical traffic stability is 0.38, and the overall score is 3.8. Route 1, with the highest overall score, is selected as the macro-path sequence. Node numbers N001 to N010, corresponding road segment numbers, dynamic traffic weights, estimated travel time, overall score, intersection turning instructions, and road segment speed limits are arranged in the order of travel to generate a complete macro-path sequence, which is then simultaneously stored in the vehicle's local storage and on the cloud platform.

[0074] In step S105, local road segment information is extracted based on the macro-path sequence and dynamically adjusted in conjunction with real-time traffic information. When a road segment obstacle is detected, the local path is replanned to obtain a refined lane-level path plan, including: Real-time traffic flow density data for each local road segment is extracted based on the macroscopic path sequence; Based on the location information of each local road segment, real-time emergency information of the corresponding road segment is obtained, and the affected lane range and expected duration of impact are determined based on the emergency information. Based on the real-time traffic flow density data of the local road segment, the affected lane range, and the expected impact duration, local path calculation is performed with the goal of minimizing traffic delay to obtain the local optimal driving sub-path. By integrating the local optimal driving sub-paths corresponding to all local road segments according to the driving sequence, a lane-level path planning that covers the entire macro-path sequence is obtained.

[0075] It should be noted that the division of local road segments is based on the road segments in the macro-path sequence, with each road segment corresponding to one local road segment. The division process proceeds sequentially according to the driving order of the macro-path sequence, covering all road segments included in the macro-path sequence. The length of the local road segment is consistent with the length of the corresponding road segment, without any additional segmentation or merging. Each local road segment is assigned a unique local road segment number, in the format "L + road segment number", for example, the local road segment number corresponding to road segment N003-N004 is L-N003-N004.

[0076] It should be noted that the real-time traffic flow density data is obtained using the V2X communication and roadside unit (RSU) broadcast parsing method described in step S104. If the roadside equipment is not covered or communication is interrupted, vehicle-mounted sensor fusion estimation is used, and the data update frequency is once per minute. The update frequency of once per minute is determined by statistical data on the rate of change of urban road traffic flow, covering three time periods: morning peak, evening peak, and off-peak, and three road types: urban arterial roads, secondary arterial roads, and local roads. The statistical results show that the average change cycle of urban road traffic flow is 1.2 minutes, and the update frequency of once per minute can reflect the real-time change characteristics of traffic flow. The real-time traffic flow density data includes the real-time number of vehicles in each lane within each local road segment, in units of vehicles / km, with a data precision of 1 vehicle / km.

[0077] It should be noted that real-time emergency information is obtained through V2X communication by receiving event announcements broadcast by other vehicles or roadside equipment, or by identifying temporary construction signs and visual features of the accident scene through vehicle-mounted cameras. The information includes the event type, location, time of occurrence, and estimated end time. When external emergency information is unavailable, this step only performs local path calculations based on real-time traffic flow density data, setting the affected lane range as an empty set. Event types are categorized into three types: construction, accident, and temporary traffic control. The affected lane range is determined based on the location and impact area of ​​the emergency. The impact range for construction events is 50 meters before and after the construction area; for accident events, it is 100 meters before and after the accident area; and for temporary traffic control events, it is the entire controlled section. The estimated impact duration is determined based on the type of emergency and statistical analysis of historical similar events' processing times. The statistical sample size for historical similar events is no less than 1000 sets, with an average processing time of 120 minutes for construction events, 30 minutes for accident events, and 60 minutes for temporary traffic control events. After obtaining real-time emergency information, the affected lane range and estimated duration of impact are updated every 10 minutes until the event ends.

[0078] It should be noted that the local path calculation aims to minimize travel delay. Input parameters include real-time traffic flow density data, the affected lane range, and the estimated impact duration. The calculation process first establishes a lane-level road network model for the local road segment, including the lane number, location, length, and real-time capacity of each lane. The real-time capacity is calculated based on the real-time traffic flow density data, with the conversion relationship being: real-time capacity equals the road's saturation capacity multiplied by 1 minus the difference between the traffic flow density and the saturation traffic flow density. The road's saturation capacity is 1800 vehicles per lane per hour. Then, a dynamic programming algorithm is used to calculate the travel delay for all feasible lane combinations within the local road segment.

[0079] Define the state variable as a tuple s(i,k), representing the state of a vehicle located in the i-th position segment and in the k-th lane of a local road segment. The position segments are uniformly discretized at 10-meter intervals. Define the decision variable as a∈{keep lane, change lane to the left, change lane to the right}. The state transition cost for each decision is: c(s(i,k),a)=t_travel(i,k)+t_wait(i,k)+δ(a)·c_laneChange, where t_travel(i,k) is the travel time to pass through the i-th position segment in the current lane k, t_wait(i,k) is the queuing time in the i-th position segment in the current lane k, c_laneChange is a single lane change penalty term of 3 seconds, and δ(a) is 1 when a is changing lanes and 0 when keeping lanes.

[0080] The dynamic programming recurrence relation is: dp[s(i+1,k')]=min{dp[s(i,k)]+c(s(i,k),a)}, where k' is the lane number after executing decision a. The boundary condition is dp[s(0,k_start)]=0, where k_start is the initial lane when the vehicle enters the local road segment. Finally, the state transition path corresponding to the minimum value in dp[s(N,k_end)] is the locally optimal driving sub-path, where N is the total number of position segments and k_end is any feasible lane.

[0081] Traffic delay consists of two parts: travel time and waiting time. Travel time is calculated by dividing the segment length (10 meters) by the average lane speed. Waiting time is calculated by multiplying the queue length by the average headway. The average headway is set to 2 seconds, which is the standard safe interval for vehicles on urban roads. Finally, the lane combination with the minimum traffic delay is selected as the locally optimal travel sub-path. It should be noted that the integration of locally optimal driving sub-paths is performed sequentially according to the driving order, with adjacent local road segments connecting at corresponding intersection nodes. During the integration process, the lane numbers of adjacent sub-paths are checked for matching; if they do not match, a lane-changing operation is added at the intersection exit. The starting position of the lane-changing operation is no less than 50 meters from the intersection stop line, and the ending position is no less than 100 meters from the intersection stop line. The 50-meter distance threshold is determined statistically based on the minimum safe distance required for logistics vehicles to change lanes at different speeds, covering a common driving speed range of 20 km / h to 60 km / h. When the vehicle speed is 60 km / h, the minimum safe distance required for lane changing is 45 meters; therefore, 50 meters is used as the general threshold. If there are multiple lane-changing needs between adjacent local road segments, the lane-changing operations are performed sequentially, with an interval of no less than 100 meters between adjacent lane-changing operations.

[0082] It should be noted that after lane-level route planning is generated, real-time traffic information is checked every 5 minutes. If new road obstructions are detected, or if the impact range or estimated duration of existing obstructions changes, the route for the corresponding local road segment is immediately recalculated. After recalculation, the content of the corresponding local road segment in the lane-level route planning is updated, and the data on the vehicle's local storage and cloud storage platform are updated simultaneously. Lane-level route planning includes six core types of information: local road segment number, lane number, travel direction, estimated travel time, lane change location, and lane change direction. The generation process arranges the information of each local road segment in the order of travel to ensure that the route direction is consistent with the vehicle's travel direction. Lane-level route planning is synchronously backed up to the vehicle's local storage unit and the cloud storage platform. The vehicle's local storage retains the route data for the current trip, while the cloud storage retains all historical route data for the past 180 days. The storage duration is determined through statistical analysis of logistics vehicle operation data and traceability needs.

[0083] For example, in an urban logistics delivery scenario, the macro-path sequence includes a local road segment from node N003 to node N004, with a total length of 1.8 kilometers and three parallel lanes. The local road segment is numbered L-N003-N004. Real-time traffic flow density data for this local road segment is obtained through the vehicle-to-everything (V2X) communication interface. The leftmost lane has 50 vehicles per kilometer, the middle lane has 45 vehicles per kilometer, and the rightmost lane has 40 vehicles per kilometer. The data was updated 30 seconds before the current time. Simultaneously, a construction event is detected on this road segment. The construction location is 200 meters ahead of node N003, the construction area is 100 meters long, the leftmost lane is affected, and the estimated impact time is 6.5 minutes. This data was obtained 1 minute before the current time. Local path calculations were performed with the goal of minimizing traffic delay. A lane-level road network model was established for this local road segment. The calculated real-time capacity of the leftmost lane was 1050 vehicles per hour, with a travel time of 4.7 minutes, a waiting time of 3.0 minutes, and a total traffic delay of 7.7 minutes. The real-time capacity of the middle lane was 1125 vehicles per hour, with a travel time of 4.3 minutes and no waiting time, and a total traffic delay of 4.3 minutes. The real-time capacity of the right lane was 1200 vehicles per hour, with a travel time of 4.0 minutes and no waiting time, and a total traffic delay of 4.0 minutes. The right lane with the minimum traffic delay was selected as the locally optimal travel sub-path. The locally optimal driving sub-path is integrated with the sub-paths of other local road segments in the macro-path sequence according to the driving order. The lane number matching of adjacent sub-paths is checked. It is found that the lane number of the sub-path of the previous local road segment is 2, and the lane number of the sub-path of the current local road segment is 3. A lane switching operation is added at the exit of node N003. The switching start position is 60 meters away from the intersection stop line, the ending position is 120 meters away from the intersection stop line, and the switching direction is to the right. Finally, a lane-level path plan containing all local road segment numbers, corresponding lane numbers, driving directions, estimated travel time, lane switching positions, and lane switching directions is generated and synchronously stored in the vehicle's local storage and the cloud platform. After driving for 5 minutes, the estimated impact duration of the construction event is detected to be updated to 10 minutes. The path of this local road segment is immediately recalculated. After recalculation, the total travel delay of the right lane becomes 5.0 minutes, and the total travel delay of the middle lane remains 4.8 minutes. The locally optimal driving sub-path is updated to the middle lane, and the corresponding content in the lane-level path plan is updated synchronously.

[0084] In step S106, based on the lane-level path planning and dynamic traffic scenario deduction, a lane-level operation sequence is generated, including: Based on the lane-level path planning, the traffic signal change patterns and surrounding vehicle driving status at the intersections are extracted to obtain the dynamic traffic scenario within a preset time period. Based on the dynamic traffic scenario, the optimal driving speed of the vehicle on each road segment and the speed of passing through intersections are calculated, and speed control commands are generated. Based on the lane switching requirements in the lane-level path planning and the dynamic traffic scenario, a safe lane-changing interval is determined, and lane operation instructions are generated. The speed control command and the lane operation command are combined in chronological order to obtain a lane-level operation sequence.

[0085] It should be noted that the preset duration is 30 seconds. This 30-second duration is determined by statistical data on the traffic signal phase change cycle at urban intersections and the average time logistics vehicles take to pass through intersections. The statistics cover three time periods: morning peak, evening peak, and off-peak, as well as three common intersection types: crossroads, T-junctions, and roundabouts, thus fully covering the entire process of a vehicle passing through a single intersection. Traffic signal change patterns are extracted from real-time traffic feature sets, including four types of information: signal phase type, remaining phase duration, phase switching time, and duration of each phase. The driving status of surrounding vehicles is collected through vehicle-mounted LiDAR and cameras, with a collection range of 100 meters in front of and behind the vehicle and 10 meters to the left and right. The collected data includes the position coordinates, speed, and direction of surrounding vehicles, with a data update frequency of 10 times per second.

[0086] It should be noted that the optimal driving speed is determined comprehensively based on the real-time traffic flow density of the road segment, the legal speed limit, and the driving status of vehicles ahead. The calculation process first determines the upper speed limit based on the legal speed limit, then adjusts the base driving speed based on the real-time traffic flow density, and finally corrects the final speed by combining the speed of vehicles ahead and the following distance. The following distance is not less than the safe braking distance corresponding to the current speed. The safe braking distance is determined by statistical analysis of braking test data of logistics vehicles at different speeds. When the vehicle speed is 30 km / h, the safe braking distance is 9 meters; when the vehicle speed is 60 km / h, the safe braking distance is 36 meters. The speed for passing through intersections is calculated based on the remaining time of the traffic signal at the intersection and the length of the intersection. The calculation process ensures that the vehicle completely crosses the stop line at the intersection before the signal phase changes.

[0087] It should be noted that the safe lane-changing interval is determined based on lane-changing requirements and the driving conditions of surrounding vehicles. The length of the lane-changing interval is not less than the minimum distance required for the vehicle to complete the lane change. The minimum distance required for a lane change is determined through statistical analysis of lane-changing test data for logistics vehicles at different speeds. When the vehicle speed is 20 km / h, the minimum distance required for a lane change is 20 meters; when the vehicle speed is 40 km / h, the minimum distance required for a lane change is 40 meters; and when the vehicle speed is 60 km / h, the minimum distance required for a lane change is 60 meters. The starting position of the lane-changing interval is not less than the minimum distance required for a lane change, and the ending position is not less than 10 meters from the lane-changing point. There must be no other vehicles within the lane-changing interval. If other vehicles are present, the lane-changing interval should be extended backward until a safe interval that meets the requirements is found.

[0088] It should be noted that speed control commands include three types of information: command type, execution time, and target speed. Lane operation commands include four types of information: command type, execution time, lane change direction, and lane change range. The precision of command execution time is in seconds, the precision of target speed is in 1 km / h, and the precision of lane change range is in 1 meter. The combination process arranges all commands in the order of their execution times. If two commands have the same execution time, the speed control command precedes the lane operation command. After the lane-level operation sequence is generated, it is updated every 5 seconds, adjusting the command content and execution time based on the latest dynamic traffic scenario.

[0089] It is worth noting that lane-level operation sequences are simultaneously backed up to both the onboard local storage unit and the cloud storage platform. The onboard local storage retains the operation sequence data for the current trip, while the cloud storage retains all historical operation sequence data for the past 180 days. The storage duration is determined through statistical analysis of logistics vehicle operation data and traceability requirements, meeting the traceability requirements of logistics companies for vehicle operation. Operation sequence data is stored in association with corresponding timestamped vehicle driving status data and road condition data, facilitating subsequent data analysis and optimization.

[0090] For example, in an urban logistics delivery scenario, lane-level path planning includes a road segment from node N005 to node N006, with a total length of 2.5 kilometers, containing 4 parallel lanes. The legal speed limit is 50 km / h, and the intersection is numbered N005 with a length of 30 meters. The traffic signal change patterns at the intersection are extracted: the current phase is a green light for straight traffic with 12 seconds remaining, and the next phase is a red light with a duration of 30 seconds. The driving status of surrounding vehicles is collected using onboard LiDAR and cameras, with a collection range of 100 meters in front of and behind the vehicle, and 10 meters to the left and right. The collection results show that there is a vehicle 50 meters ahead at a speed of 40 km / h, no vehicle in the left lane, and one vehicle in the right lane at a speed of 35 km / h, thus obtaining the dynamic traffic scene for the next 30 seconds. Based on the dynamic traffic scenario, the optimal driving speed is calculated to be 45 km / h, and the intersection crossing speed is 30 km / h. A speed control command is generated, executed at the current time, with a target speed of 45 km / h, adjusted to 30 km / h when crossing the intersection. According to the lane-switching requirements in the lane-level path planning, a switch from lane 2 to lane 3 is needed at the N005 exit. Combining the dynamic traffic scenario, the safe lane-changing interval is determined to be 60 to 120 meters from the intersection stop line. A lane operation command is generated, executed 5 seconds after the current time, with the lane-changing direction to the right and the lane-changing interval between 60 and 120 meters. The speed control command and lane operation command are combined in the order of execution time to obtain a complete lane-level operation sequence, which is simultaneously stored in the vehicle's local storage and on the cloud platform. After 5 seconds of driving, if the speed of the vehicle ahead is detected to have dropped to 35 km / h, the optimal driving speed is immediately updated to 38 km / h, the target speed of the corresponding speed control command is adjusted, and the lane-level operation sequence is updated simultaneously.

[0091] In step S107, the vehicle is controlled to drive in real time according to the lane-level operation sequence, and the driving trajectory deviation is monitored. When the trajectory deviation exceeds a preset deviation threshold, the real-time traffic information is re-integrated and the planning is updated to obtain the final path instruction, including: The lane-level operation sequence is encrypted and transmitted to the vehicle control system, which controls the vehicle to drive in real time according to the lane-level operation sequence. By integrating high-precision map data and real-time road information collected by vehicle-mounted sensing devices, vehicle trajectory deviation data is calculated. If the trajectory deviation data exceeds the preset deviation threshold, the latest real-time traffic information is re-acquired and the route planning is updated, and an adjustment instruction is generated. The adjustment command is integrated with the lane-level operation sequence to obtain the final path command.

[0092] It should be noted that the transmission process uses the AES-128 encryption algorithm, with a transmission frequency of 10 times per second and a latency controlled within 50 milliseconds. The transmitted data includes speed control commands and lane operation commands, each with a unique timestamp accurate to the millisecond level. Upon receiving the commands, the vehicle control system executes them in the order of their timestamps, recording the execution time and result.

[0093] It should be noted that the high-precision map data has an accuracy of 0.1 meters and includes lane centerline coordinates, lane boundary coordinates, and road curvature information. The vehicle-mounted sensing equipment includes LiDAR and cameras, collecting data within a 100-meter radius in front of and behind the vehicle, and 10 meters to the left and right, at a frequency of 10 times per second. Trajectory deviation data includes the lateral distance between the vehicle's actual position and the centerline of the planned path, and the longitudinal deviation between the vehicle's actual travel time and the planned travel time. The calculation process is performed 10 times per second, and the results are rounded to two decimal places. The preset lateral deviation threshold is 0.2 meters, and the longitudinal deviation threshold is 5 seconds, determined through statistical experiments on vehicle trajectory control accuracy under 1000 different driving scenarios. When the lateral deviation exceeds 0.2 meters or the longitudinal deviation exceeds 5 seconds, path replanning is triggered. The experiments cover a driving speed range of 20 km / h to 60 km / h, as well as common weather conditions such as sunny, cloudy, and rainy days.

[0094] It should be noted that the reacquired real-time traffic information includes real-time traffic flow density data, real-time emergency information, and real-time traffic signal status. The input parameters for updating the route planning process include the latest real-time traffic information, the vehicle's current location, and its speed. Adjustment commands include speed adjustment commands and lane adjustment commands, with execution starting from the current moment and covering the next 30 seconds of the driving process. Once generated, the adjustment commands are immediately transmitted to the vehicle control system, simultaneously updating the lane-level operation sequence.

[0095] It should be noted that the integration process is performed in the order of instruction execution time. If the execution time of the adjustment instruction overlaps with that of the instructions in the original lane-level operation sequence, the adjustment instruction overrides the original instruction. The final route instructions include all unexecuted original instructions and newly added adjustment instructions, arranged in chronological order of execution. The final route instructions are synchronously backed up to the onboard local storage unit and the cloud storage platform. The onboard local storage retains the instruction data for the current trip, while the cloud storage retains all historical instruction data for the most recent 180 days. Trajectory deviation data and correction records are synchronously uploaded to the cloud database for subsequent training of the route optimization model.

[0096] For example, in an urban logistics delivery scenario, the lane-level operation sequence includes speed control instructions and lane operation instructions 10 seconds after the current moment, with a target speed of 40 km / h and lane 3 as the driving lane. The lane-level operation sequence is transmitted to the vehicle control system using AES-128 encryption at a frequency of 10 times per second with a delay of 30 milliseconds. The vehicle control system executes the instructions sequentially according to timestamps, controlling the vehicle to travel at 40 km / h in lane 3. By integrating high-precision map data and real-time road information collected by onboard LiDAR and cameras, the lateral distance between the vehicle's actual position and the centerline of the planned path is calculated to be 0.3 meters, exceeding the preset deviation threshold of 0.2 meters. The latest real-time traffic information is then retrieved; the current traffic flow density is 40 vehicles per kilometer, there are no emergencies, and the traffic light is green. The path planning is updated, generating adjustment instructions to adjust the steering angle by 2 degrees and the target speed to 35 km / h, with the execution time being the current moment. These adjustment instructions are then integrated with the original lane-level operation sequence, covering the current speed and lane control content from the original instructions, resulting in the final path instructions. The final route instruction is simultaneously backed up to the vehicle's local storage and the cloud platform, while the trajectory deviation data of 0.3 meters and the correction record are simultaneously uploaded to the cloud database. After driving for 5 seconds, the trajectory deviation is recalculated to 0.1 meters, which does not exceed the preset deviation threshold, and the final route instruction continues to be executed.

[0097] In summary, this invention provides a multi-objective dynamic path planning method for logistics vehicles, which effectively solves the problems of existing technologies lacking the ability to adapt to complex dynamic environments, failing to effectively distinguish between static road information and real-time road condition changes, resulting in large planning deviations and untimely responses, and reducing the utilization efficiency and delivery punctuality of logistics transportation systems.

[0098] refer to Figure 2 The second embodiment of the present invention provides a multi-objective dynamic path planning system for logistics vehicles, comprising: The data acquisition and processing module is used to collect vehicle surrounding environment data and real-time location coordinates, and to perform time-series alignment and anomaly removal processing on the surrounding environment data and real-time location coordinates to obtain a preliminary dynamic road dataset. The lane status verification module performs road feature recognition and traffic status verification processing based on the preliminary dynamic road dataset to obtain a lane availability mapping. The traffic signal fusion module fuses the traffic signal status transmitted from the roadside according to the lane availability mapping, and suspends path calculation when the traffic signal status is prohibited from passing, thereby obtaining a real-time traffic feature set; The macro-path planning module performs global path search and multi-dimensional optimization processing based on the real-time traffic feature set to obtain an optimized macro-path sequence. The local path refinement module extracts local road segment information based on the macro path sequence and makes dynamic adjustments in combination with real-time traffic information. When a road segment obstacle is detected, the local path is replanned to obtain a refined lane-level path plan. The driving instruction generation module generates speed control instructions and lane operation instructions for vehicle driving based on the lane-level path planning and dynamic traffic scenario deduction, and merges the speed control instructions and lane operation instructions to obtain a lane-level operation sequence. The path execution correction module controls the vehicle to drive in real time and monitors the driving trajectory deviation according to the lane-level operation sequence. When the trajectory deviation exceeds a preset deviation threshold, the real-time traffic information is re-integrated and the planning is updated to obtain the final path instruction.

[0099] It should be noted that the multi-objective dynamic path planning system for logistics vehicles provided in this embodiment of the invention is used to execute all the process steps of the multi-objective dynamic path planning method for logistics vehicles in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0100] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0101] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A multi-objective dynamic path planning method for a logistic vehicle, characterized in that, include: Collect vehicle surrounding environment data and real-time location coordinates, and perform time-series alignment and anomaly removal processing on the surrounding environment data and real-time location coordinates to obtain a preliminary dynamic road dataset. Based on the preliminary dynamic road dataset, road feature recognition and traffic status verification are performed to obtain lane availability mapping; Based on the lane availability mapping and the roadside-transmitted traffic signal status, path calculation is paused when the traffic signal status is prohibited, thus obtaining a real-time traffic feature set; Based on the real-time traffic feature set, a global path search and multi-dimensional optimization process are performed to obtain an optimized macro-path sequence. Local road segment information is extracted based on the macro-path sequence and dynamically adjusted in combination with real-time traffic information. When a road segment obstacle is detected, the local path is replanned to obtain a refined lane-level path plan. Based on the lane-level path planning and dynamic traffic scenario simulation, speed control commands and lane operation commands for vehicle driving are generated, and the speed control commands and lane operation commands are fused to obtain a lane-level operation sequence. The vehicle is controlled to drive in real time according to the lane-level operation sequence and the driving trajectory deviation is monitored. When the trajectory deviation exceeds the preset deviation threshold, the real-time traffic information is re-integrated and the planning is updated to obtain the final path instruction.

2. The method of claim 1, wherein, The process involves collecting vehicle surrounding environment data and real-time location coordinates, performing time-series alignment and anomaly removal on the surrounding environment data and real-time location coordinates, and obtaining a preliminary dynamic road dataset, including: Vehicle-mounted LiDAR, cameras, and satellite positioning devices are used to collect obstacle data, road image data, and real-time location coordinates around the vehicle. The obstacle data, road image data, and real-time location coordinates are then time-sequentially aligned according to the collection timestamps to obtain aligned multi-source data. The surrounding environment data includes the obstacle data and the road image data. Outlier removal is performed on the aligned multi-source data to remove invalid data that exceeds a preset value range, and lane layout information and temporary closed area information are extracted from the road image data. By integrating the real-time location coordinates, the lane layout information, and the temporary closed area information, a preliminary dynamic road dataset is obtained.

3. The method of claim 1, wherein, The process of identifying road features and verifying traffic status based on the preliminary dynamic road dataset to obtain lane availability mapping includes: Based on the preliminary dynamic road dataset, road image sequences are extracted, and edge feature extraction processing is performed on the road image sequences to obtain lane line edge contour data; The lane boundary position is determined based on the lane line edge contour data, and the initial traffic status of each lane is marked by combining the temporary closed area information in the preliminary dynamic road dataset. The lane boundary location and the initial traffic status are compared and verified with high-precision map data in the cloud. After correcting the deviation, a lane availability mapping containing lane number and corresponding traffic status is generated.

4. The method of claim 1, wherein, The method involves fusing the traffic signal status transmitted from the roadside based on the lane availability mapping, pausing path calculation when the traffic signal status is prohibited, and obtaining a real-time traffic feature set, including: The vehicle-to-everything (V2X) communication interface is used to obtain information on traffic signal status, remaining signal duration, and intersection number transmitted by roadside equipment. The traffic signal status, the remaining duration of the signal, and the lane availability mapping are weighted and fused to generate an intersection traffic condition feature vector. If the traffic signal status is prohibited from passing, the global path calculation is paused based on the remaining duration of the signal, and the current traffic restriction level of the intersection is determined based on the intersection traffic condition feature vector; The intersection traffic condition feature vector and the traffic restriction level are integrated to obtain a real-time traffic feature set.

5. The multi-objective dynamic path planning method for logistics vehicles according to claim 1, characterized in that, The step of performing global path search and multi-dimensional optimization based on the real-time traffic feature set to obtain an optimized macro-path sequence includes: Based on the real-time traffic feature set, urban roads are divided into intersection nodes and road segments, and a dynamic traffic weight is assigned to each road segment. The dynamic traffic weight is determined by a combination of the congestion level, length and estimated travel time of the corresponding road segment. Based on the dynamic access weight, the A* algorithm is used to perform a global path search and generate candidate global paths. Calculate a comprehensive score for each candidate global path based on average traffic efficiency and historical traffic stability, and select the candidate path with the highest comprehensive score as the macroscopic path sequence.

6. The multi-objective dynamic path planning method for logistics vehicles according to claim 1, characterized in that, The step of extracting local road segment information based on the macro-path sequence and dynamically adjusting it in conjunction with real-time traffic information, and replanning the local path when a road segment obstacle is detected, to obtain a refined lane-level path plan, includes: Real-time traffic flow density data for each local road segment is extracted based on the macroscopic path sequence; Based on the location information of each local road segment, real-time emergency information of the corresponding road segment is obtained, and the affected lane range and expected duration of impact are determined based on the emergency information. Based on the real-time traffic flow density data of the local road segment, the affected lane range, and the expected impact duration, local path calculation is performed with the goal of minimizing traffic delay to obtain the local optimal driving sub-path. By integrating the local optimal driving sub-paths corresponding to all local road segments according to the driving sequence, a lane-level path planning that covers the entire macro-path sequence is obtained.

7. The multi-objective dynamic path planning method for logistics vehicles according to claim 1, characterized in that, The process of generating a lane-level operation sequence based on the lane-level path planning and dynamic traffic scenario deduction includes: Based on the lane-level path planning, the traffic signal change patterns and surrounding vehicle driving status at the intersections are extracted to obtain the dynamic traffic scenario within a preset time period. Based on the dynamic traffic scenario, the optimal driving speed of the vehicle on each road segment and the speed of passing through intersections are calculated, and speed control commands are generated. Based on the lane switching requirements in the lane-level path planning and the dynamic traffic scenario, a safe lane-changing interval is determined, and lane operation instructions are generated. The speed control command and the lane operation command are combined in chronological order to obtain a lane-level operation sequence.

8. The multi-objective dynamic path planning method for logistics vehicles according to claim 1, characterized in that, The process of controlling the vehicle to drive in real time according to the lane-level operation sequence and monitoring the driving trajectory deviation, and when the trajectory deviation exceeds a preset deviation threshold, re-integrating the real-time traffic information and updating the plan to obtain the final path instruction includes: The lane-level operation sequence is encrypted and transmitted to the vehicle control system, which controls the vehicle to drive in real time according to the lane-level operation sequence. By integrating high-precision map data and real-time road information collected by vehicle-mounted sensing devices, vehicle trajectory deviation data is calculated. If the trajectory deviation data exceeds the preset deviation threshold, the latest real-time traffic information is reacquired and the route planning is updated to generate an adjustment instruction that minimizes safety hazards. The adjustment command is integrated with the lane-level operation sequence to obtain the final path command.

9. A multi-objective dynamic path planning system for logistics vehicles, characterized in that, include: The data acquisition and processing module is used to collect vehicle surrounding environment data and real-time location coordinates, and to perform time-series alignment and anomaly removal processing on the surrounding environment data and real-time location coordinates to obtain a preliminary dynamic road dataset. The lane status verification module performs road feature recognition and traffic status verification processing based on the preliminary dynamic road dataset to obtain a lane availability mapping. The traffic signal fusion module fuses the traffic signal status transmitted from the roadside according to the lane availability mapping, and suspends path calculation when the traffic signal status is prohibited from passing, thereby obtaining a real-time traffic feature set; The macro-path planning module performs global path search and multi-dimensional optimization processing based on the real-time traffic feature set to obtain an optimized macro-path sequence. The local path refinement module extracts local road segment information based on the macro path sequence and makes dynamic adjustments in combination with real-time traffic information. When a road segment obstacle is detected, the local path is replanned to obtain a refined lane-level path plan. The driving instruction generation module generates speed control instructions and lane operation instructions for vehicle driving based on the lane-level path planning and dynamic traffic scenario deduction, and merges the speed control instructions and lane operation instructions to obtain a lane-level operation sequence. The path execution correction module controls the vehicle to drive in real time and monitors the driving trajectory deviation according to the lane-level operation sequence. When the trajectory deviation exceeds a preset deviation threshold, the real-time traffic information is re-integrated and the planning is updated to obtain the final path instruction.

Images