A method for loading and unloading goods on unmanned logistics vehicles based on a swap-and-hook approach.

Abstract

This application relates to a method for loading and unloading goods from unmanned logistics vehicles based on a swap-and-hook mechanism, comprising: collecting unmanned logistics vehicle operation data, swap-and-hook device status data, and cargo box storage data; constructing a cargo box-vehicle matching degree calculation model, outputting a quantitative matching score between the cargo box and the unmanned logistics vehicle, and generating a loading and unloading scheduling task based on the quantitative matching score; generating a collision-free driving trajectory based on the unmanned logistics vehicle operation data, swap-and-hook device status data, and cargo box storage data; the unmanned logistics vehicle completing dynamic parking along the collision-free driving trajectory; and the swap-and-hook device performing the docking, undocking, and loading / unloading operations of the cargo box and the unmanned logistics vehicle. This invention achieves automatic docking, automatic docking, and rapid loading and unloading of the cargo box and the unmanned logistics vehicle without human intervention, eliminating efficiency bottlenecks and safety hazards caused by manual intervention; and significantly improving loading and unloading efficiency and docking accuracy through collaborative mechanisms such as cargo box-vehicle matching, path planning, obstacle avoidance, and dynamic adaptation.

Description

Technical Field

[0001] This application relates to the field of logistics and transportation technology, specifically to a method for loading and unloading goods on unmanned logistics vehicles based on a swapping method. Background Technology

[0002] With the rapid development of autonomous driving technology, unmanned logistics vehicles have been gradually applied to various transportation scenarios such as park logistics, urban delivery, and trunk line transshipment. In these scenarios, the efficiency of loading and unloading goods directly affects the timeliness and capacity utilization of the entire transportation chain. If the loading and unloading process is inefficient, even if the transportation process is unmanned and automated, the overall operation time will still be significantly constrained.

[0003] Currently, most unmanned logistics vehicles still rely on manual operation or semi-automatic equipment for loading and unloading, such as having staff load or unload packages or goods onto or from the vehicle. This method suffers from slow loading and unloading speeds, high labor intensity, and low automation, especially at night or in unattended scenarios, where manual loading and unloading methods cannot sustainably support high-frequency transportation demands.

[0004] To improve overall logistics efficiency, some solutions attempt to assist operations with loading and unloading equipment such as automatic lifting platforms and robotic arms. However, these solutions often require real-time docking between vehicles and station equipment, and loading and unloading operations can only be carried out after the vehicles arrive. This results in the entire system being in a synchronous loading and unloading mode, which cannot make full use of the idle time of stations and vehicles.

[0005] In contrast, the trailer swapping mode has been widely adopted in traditional manned freight transport. Its core advantage lies in decoupling the transport vehicle from the loading and unloading operations, thereby improving vehicle utilization and operational parallelism. In the trailer swapping mode, the vehicle and cargo box are separated, and the cargo box can be loaded and unloaded independently.

[0006] However, existing technologies have significant shortcomings: in unmanned logistics scenarios, the swap-and-hook model lacks mature automated loading and unloading control methods and collaborative mechanisms, resulting in the underutilization of this highly efficient model. How to introduce a swap-and-hook mechanism into an unmanned logistics system to achieve automatic docking, automatic attachment, and rapid loading and unloading of cargo containers and unmanned logistics vehicles, thereby eliminating manual intervention, improving loading and unloading efficiency, and realizing asynchronous collaboration between transport vehicles and station operations, has become a pressing technological challenge.

[0007] Therefore, this invention proposes a method for loading and unloading goods on unmanned logistics vehicles based on a swapping method. Summary of the Invention

[0008] To overcome the shortcomings of the prior art, this application provides a method for loading and unloading goods on unmanned logistics vehicles based on a swapping method, specifically adopting the following technical solution.

[0009] A method for loading and unloading goods on unmanned logistics vehicles based on a swapping mechanism includes the following steps.

[0010] S1. Cloud-based Task Initialization and Pre-scheduling: Constructing a multi-dimensional quantitative matching model and a self-learning scheduling mechanism. The cloud-based scheduling system collects multi-source data from unmanned logistics vehicles, automatic trailer swapping devices, and cargo container storage in real time, establishing a global state database to achieve full system monitoring. A matching degree calculation model is constructed based on the analytic hierarchy process (AHP), clarifying the weights of each dimension and the scoring formula to achieve accurate quantitative matching between cargo containers and vehicles, avoiding wasted transportation capacity. Task generation adopts an automatic-first, manual emergency coverage mode, synchronously linking with the warehousing and sorting system to achieve asynchronous parallel loading, unloading, and transportation.

[0011] S2. Autonomous Vehicle Localization and Parking Path Planning: A collaborative design of differentiated positioning strategies and optimized path planning addresses the issues of insufficient positioning accuracy and path planning not fitting real-world scenarios. After the autonomous vehicle arrives at the preset area, differentiated positioning is performed for different types of trailer hitches: fixed hitches directly utilize the precise location assigned by the task, while integrated hitches combine real-time positioning, historical data review, and vehicle-side perception to achieve accurate positioning. HybridA is constructed based on multi-source information fusion. The search map employs a planning strategy with 6 discrete steering directions and RS curve termination, clearly defining the specific values ​​of the front wheel steering angle and fine-tuning rules. It uses a cost function with safety margin to generate smooth, collision-free parking paths that conform to vehicle kinematic constraints, balancing search efficiency and driving safety.

[0012] S3. Dynamic Parking Execution and Safe Obstacle Avoidance: Employing a scenario-based adaptation and dynamic path correction mechanism based on dynamic collision detection overcomes the shortcomings of existing parking processes, such as inaccurate collision detection and operational interruptions caused by dynamic obstacles. The unmanned vehicle completes speed planning and trajectory prediction based on the planned path, and conducts dynamic collision detection using a differentiated time window of 3-5 seconds to achieve accurate identification of collision risks. Through high-frequency detection and the continued use of speed adjustment strategies, redundant speed adjustments are avoided, improving obstacle avoidance efficiency. When a dynamic obstacle lingers for more than 30 seconds, path correction is automatically triggered, increasing the number of node expansion directions to eight, dynamically adjusting the cost function weights, prioritizing obstacle avoidance while quickly returning to the original trajectory, achieving a balance between dynamic obstacle avoidance and parking efficiency, and ensuring continuous and controllable operation.

[0013] S4. Automatic Hook-and-Truck Docking and Loading / Unloading Execution: Adopting an adaptive attitude hierarchical correction and device type adaptation mechanism solves the problems of large docking deviations, poor adaptability, and excessive manual intervention in existing systems. After the unmanned vehicle completes parking, the cloud-based scheduling system triggers adaptation adjustments based on the hook-and-truck device type: fixed devices adjust the loading / unloading mechanism's attitude, while integrated devices adapt to different network scenarios based on their flexibility. A hierarchical PID adaptive correction algorithm is used to clearly define the PID coefficients corresponding to each deviation, achieving three-dimensional closed-loop correction in a vertical-horizontal-angle sequence. Combined with scenario adaptation optimization and a preset parameter library, this enables docking between the cargo box and the vehicle.

[0014] S5. Fault Self-Healing and Task Closure: Employing multi-dimensional fault monitoring, tiered self-healing, and emergency response mechanisms, this system addresses the issues of untimely fault handling and easy disruption of the logistics chain. The status monitoring module uses multi-sensor fusion, clearly defining sampling frequency and connection logic to collect core operational data in real time and implement tiered early warning through fault warning formulas. Upon completion of loading and unloading, a task closure is achieved, synchronously feeding back to the corresponding client based on task type, ensuring continuous logistics and improving operational reliability and fault tolerance.

[0015] The technical solution of this application has achieved the following beneficial effects.

[0016] This technology effectively addresses the technical challenge of lacking mature automated loading and unloading control methods and collaborative mechanisms in the swap-and-hook mode of unmanned logistics scenarios. It successfully introduces the swap-and-hook mechanism into the unmanned logistics system, enabling automatic docking, automatic attachment, and rapid loading and unloading of cargo boxes and unmanned logistics vehicles without human intervention. This eliminates efficiency bottlenecks and safety hazards caused by manual intervention. Through collaborative mechanisms such as cargo box-vehicle matching, path planning, safety obstacle avoidance, and dynamic adaptation, it significantly improves loading and unloading efficiency and docking accuracy. At the same time, it achieves asynchronous collaboration between transport vehicles and station operations, fully leveraging the high efficiency advantages of the swap-and-hook mode and promoting the large-scale development of the unmanned logistics industry. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method, as described in this application.

[0018] Figure 2 A schematic diagram of loading and unloading boxes for a fixed drop-and-hook device.

[0019] Figure 3 The integrated trailer is in the retracted state, and the unmanned logistics vehicle is waiting for the device to support it.

[0020] Figure 4 This diagram illustrates how unmanned logistics vehicles can be parked for loading or unloading containers, supported by a fixed trailer hitch.

[0021] Figure 5A schematic diagram of the parking trajectory search for unmanned logistics vehicles. Detailed Implementation

[0022] The present application will now be further described with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application and should not be construed as limiting the scope of protection of the present application. It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present application.

[0023] like Figure 1 As shown, the present invention discloses a method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method, comprising the following steps.

[0024] S1. Cloud-based Task Initialization and Pre-scheduling: Constructing a multi-dimensional quantitative matching model and a self-learning scheduling mechanism. The cloud-based scheduling system collects multi-source data from unmanned logistics vehicles, automatic trailer swapping devices, and cargo container storage in real time, establishing a global state database to achieve full system monitoring. A matching degree calculation model is constructed based on the analytic hierarchy process (AHP), clarifying the weights and scoring formulas for each dimension to achieve accurate quantitative matching between cargo containers and vehicles, avoiding wasted transportation capacity. Task generation adopts an automatic priority, manual emergency coverage mode, synchronously linking with the warehousing and sorting system to achieve asynchronous parallel loading, unloading, and transportation. Simultaneously, the model supports self-learning optimization, dynamically adjusting weights based on historical data to solve the problems of low matching accuracy and insufficient capacity utilization in existing scheduling systems.

[0025] S2. Precise Localization and Parking Path Planning for Autonomous Vehicles: This system employs a collaborative design of differentiated positioning strategies and optimized path planning to address the issues of insufficient positioning accuracy and path planning not fitting real-world scenarios. After the autonomous vehicle arrives at the preset area, differentiated positioning is performed based on different types of trailer hitchs: fixed hitchs directly utilize the precise location assigned by the task, while integrated hitchs combine real-time positioning, historical data review, and vehicle-side perception to achieve accurate positioning. A Hybrid A system is constructed based on multi-source information fusion. The search map employs a planning strategy with 6 discrete steering directions and RS curve termination, clearly defines the specific values ​​of the front wheel steering angle and fine-tuning rules, designs a cost function with safety margin, and generates a smooth, collision-free parking path that conforms to vehicle kinematic constraints, balancing search efficiency and driving safety.

[0026] S3. Dynamic Parking Execution and Safe Obstacle Avoidance: Employing a scenario-based adaptation and dynamic path correction mechanism based on dynamic collision detection overcomes the shortcomings of existing parking processes, such as inaccurate collision detection and operational interruptions caused by dynamic obstacles. The unmanned vehicle completes speed planning and trajectory prediction based on the planned path, and conducts dynamic collision detection using a differentiated time window of 3-5 seconds to achieve accurate identification of collision risks. Through high-frequency detection and the continued use of speed adjustment strategies, redundant speed adjustments are avoided, improving obstacle avoidance efficiency. When a dynamic obstacle lingers for more than 30 seconds, path correction is automatically triggered, increasing the number of node expansion directions to eight, dynamically adjusting the cost function weights, prioritizing obstacle avoidance while quickly returning to the original trajectory, achieving a balance between dynamic obstacle avoidance and parking efficiency, and ensuring continuous and controllable operation.

[0027] S4. Automated Drop-and-Hook Docking and Loading / Unloading: Adopting an adaptive attitude hierarchical correction and device type adaptation mechanism, this system addresses the existing problems of large docking deviations, poor adaptability, and excessive manual intervention. After the unmanned vehicle completes precise parking, the cloud-based scheduling system triggers adaptation adjustments based on the type of drop-and-hook device: fixed devices adjust the attitude of the loading / unloading mechanism, while integrated devices adapt to different network scenarios based on their flexibility. A hierarchical PID adaptive correction algorithm is used to clearly define the PID coefficients corresponding to each deviation, achieving three-dimensional closed-loop correction in a vertical-horizontal-angle sequence. Combined with scenario adaptation optimization and a preset parameter library, the docking deviation is controlled within ±3.0mm, with a total correction time ≤5s, achieving precise docking of the cargo box and vehicle without any manual intervention, thus improving docking efficiency and accuracy.

[0028] S5. Fault Self-Healing and Task Closure: Employing multi-dimensional fault monitoring, tiered self-healing, and emergency response mechanisms, this system addresses the issues of untimely fault handling and easy disruption of the logistics chain. The status monitoring module uses multi-sensor fusion, clearly defining sampling frequency and connection logic to collect four types of core operational data in real time. Tiered warnings are implemented through fault early warning formulas. For minor faults, self-healing operations are automatically triggered, with the effectiveness evaluated using a self-healing success rate formula. For severe faults, the cargo container is immediately locked, the issue is reported to the cloud, idle vehicles are coordinated to take over the task, and a maintenance work order is generated. After loading and unloading, a task closure is achieved, synchronously feeding back to the corresponding client based on the task type, ensuring continuous logistics and improving operational reliability and fault tolerance.

[0029] Example 1.

[0030] This embodiment 1 discloses a method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method, wherein step S1 includes the following steps.

[0031] S11: Multi-source information acquisition and real-time status monitoring: The cloud-based dispatch system collects and synchronizes real-time data from unmanned logistics vehicles, automatic trailer swapping devices, and cargo storage data, establishing a global status database to achieve information acquisition and status monitoring across the entire system.

[0032] Real-time data for unmanned logistics vehicles includes vehicle parameters, high-precision positioning information, vehicle status, unique identification ID, maximum load capacity, battery range, and current mission status.

[0033] Automatic trailer swapping device data includes device type, equipment ID, installation location, operating status, fault information, attitude adjustment capability, and loading point location. Automatic trailer swapping devices are available in fixed and integrated types.

[0034] Cargo container storage data includes cargo container weight, volume, dimensions, destination, transit time, cargo priority, urgency, pre-loading and unloading status. It also connects with the warehousing and sorting system and the automated sorting system to obtain the cargo container pre-loading progress, enabling multi-system linkage and improving the overall automation level of the logistics system.

[0035] S12: Cargo Box-Vehicle Intelligent Matching and Task Generation: The cloud-based scheduling system implements a cloud-based intelligent scheduling mechanism. Based on the transportation plan and multi-source data collection, the cloud scheduling service constructs a multi-dimensional matching degree calculation model. Input parameters are the unmanned logistics vehicle parameters and the cargo box parameters. Unmanned logistics vehicle parameters include maximum load capacity, cargo box size adaptation range, battery range, current location, and historical operational efficiency. Cargo box parameters include weight, volume, dimensions, transportation destination, transportation time limit, and cargo priority. A matching degree score is calculated using the analytic hierarchy process (AHP) to achieve quantitative matching and precise binding, avoiding wasted transportation capacity. The matching degree calculation model is as follows.

[0036] ; .

[0037] ;

[0038] ; ; ; ; in: It is the total score of the matching degree between the cargo box and the vehicle, with a value range of 0 to 100. A score of ≥80 is the best match, 60 to 79 is a qualified match, and <60 is a mismatch. For load adaptability weight, Weighting for size fit As a weight for battery life matching, As the path fit weight, The priority matching degree weight is initially set as: load capacity adaptability. Size fit Battery life matching Path fit Priority matching degree ; For the load-bearing adaptability score, Score for size fit. The score is for the battery life matching score. The score is for the path fit item. The score is for the priority matching degree. This refers to the weight of the cargo box. As the upper limit of vehicle load, when hour, ; These are the length, width, and height of the cargo box, respectively. These are the length, width, and height that the vehicle's cargo box is adapted to; For the vehicle's remaining range, For transportation distance; The distance from the vehicle's current location to the loading / unloading point. The optimal distance from the vehicle to the loading / unloading point; This is a task priority coefficient, used to quantify the score weight of tasks with different priorities, including urgent tasks. routine tasks Ordinary task .

[0039] Task generation and assignment employ two methods: 1) Manual tasks issued by personnel through the scheduling platform; 2) Automated tasks automatically generated by the system based on transportation plans or station operation queues. Automated tasks generated by the system are executed first, while manual tasks can override automated tasks for emergency adjustments.

[0040] Each task includes high-precision loading / unloading locations, device type, unique cargo container identifier, and task priority. Task priorities include urgent, routine, and ordinary, and the corresponding attitude adjustment parameters of the trailer hitch are marked, such as adjustment stroke, fine-tuning accuracy, and response speed. Simultaneously, by decoupling cargo container loading / unloading from vehicle transportation, the tasks are synchronously linked to the warehouse sorting system, ensuring that the site can complete the pre-loading of cargo containers before the arrival of unmanned vehicles. This lays the foundation for subsequent asynchronous collaborative loading / unloading, achieving asynchronous parallel loading / unloading and transportation, and significantly improving overall operational efficiency.

[0041] S13: Task Issuance and Preliminary Vehicle Route Planning: After the cloud-based dispatch system generates and binds the task, it issues the loading and unloading task to the target unmanned logistics vehicle. Simultaneously, it initiates task issuance and execution monitoring, real-time monitoring of the vehicle's driving trajectory and remaining range, and updating road conditions around the loading and unloading points. If congestion occurs at the site, it automatically triggers load balancing scheduling to adjust the task execution order and avoid multiple vehicles waiting in a cluster. After receiving the task, the vehicle preliminarily plans its driving route based on a high-precision map and proceeds to the designated loading and unloading point, requiring no manual intervention throughout the entire process.

[0042] This technical solution addresses the problems of inefficient matching of cargo containers and vehicles, wasted capacity, low scheduling efficiency, and insufficient multi-system collaboration in existing scheduling processes. It establishes a global state database through multi-source information collection, enabling real-time monitoring of the entire system and ensuring accurate and reliable scheduling data. A quantitative matching model is constructed based on the analytic hierarchy process (AHP), clarifying the weights and scoring formulas for each dimension to achieve precise binding of cargo containers and vehicles, avoiding wasted capacity. Compared to conventional experience-based matching methods, matching accuracy is improved by more than 30%. Simultaneously, an automatic priority, manual emergency coverage task generation mode is adopted, linking the warehousing and sorting systems to achieve asynchronous parallel loading, unloading, and transportation, shortening the operation cycle. The self-learning optimization function of the matching model can dynamically adjust weights based on historical data, continuously improving scheduling rationality and adapting to multi-vehicle dynamic scheduling scenarios, significantly enhancing the intelligence level and capacity utilization of logistics scheduling.

[0043] Example 2.

[0044] Reference Figure 2 , Figure 3 and Figure 4 This embodiment 2 discloses a method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method, wherein step S2 includes the following steps.

[0045] S21. Differentiated Positioning and Virtual Parking Space Modeling of the Trailer-Swapping Device: After the unmanned logistics vehicle arrives at the preset area of ​​the loading and unloading point, it enters the first stage of the automatic trailer-swapping docking mechanism. The unmanned vehicle is precisely positioned, and differentiated positioning is performed according to the type of trailer-swapping device to realize virtual parking space and target area modeling. That is, based on the actual layout and location information of the trailer-swapping device, a precise virtual parking space model is constructed in a high-precision map as a target reference for the unmanned vehicle's path planning and parking control. The specific positioning method is as follows.

[0046] S21.1 Fixed-position trailer hitch: Its position is usually fixed after installation and deployment, so the unmanned vehicle can directly trust the high-precision position information issued during the mission.

[0047] S21.2 Integrated trailer swapping device: Its position may change due to the last detachment, being transported, or its own movement. Therefore, multiple mechanisms are used to determine its precise position: (1) Real-time positioning of the intelligent cargo box: If the integrated trailer swapping device has a built-in positioning module, it can report its own high-precision position information in real time; (2) Historical position retrieval: The scheduling system will record the high-precision position of the unmanned vehicle when the integrated trailer swapping device was last detached from the unmanned vehicle. When issuing the subsequent docking task, the scheduling system will use this historical position as the initial reference point to issue to the vehicle end; (3) Vehicle end perception precise positioning: When the unmanned vehicle drives to the vicinity of the historical position, it can accurately determine the relative position relationship between the vehicle and the integrated trailer swapping device through perception recognition and vehicle body QR code positioning, thereby obtaining the current absolute position of the device. A QR code is preset on the front of the vehicle body and is identified in conjunction with the cargo box logo to improve positioning accuracy.

[0048] The specific perception and recognition method is as follows: A vehicle-mounted high-definition camera captures pre-set, unique markings on the cargo box surface, such as cargo box serial number inkjet printing, QR codes / barcodes, etc. The marking information is then parsed using the YOLOv5 algorithm. Simultaneously, the vehicle-mounted LiDAR scans the cargo box outline and surrounding obstacles, and the PCL point cloud processing algorithm library obtains the spatial outline coordinates of the cargo box. These two methods are then fused for calculation, accurately identifying the cargo box position and the distribution of surrounding obstacles. This determines the relative positional deviation between the unmanned vehicle and the integrated trailer hitch. Combined with the unmanned vehicle's own high-precision positioning information, the current absolute position of the integrated trailer hitch is finally calculated. The YOLOv5 algorithm is a mature target detection algorithm with a recognition accuracy of ≥99%, fast response speed, and adaptability to dynamic vehicle scenarios. This invention adds a cargo box marking pose calculation head to the standard YOLOv5 network structure, directly outputting the relative position and angle offset of the target detection results, instead of just outputting the detection box, achieving integrated detection and positioning and improving real-time positioning performance. The PCL point cloud processing algorithm library is an open-source, mature algorithm that can quickly extract the cargo box outline and obtain spatial coordinates. When the deviation of the three positioning results exceeds 5 mm, the vehicle-side perception positioning result is taken as the most accurate. When the deviation is between 3 and 5 mm, the average of the three positioning results is taken as the final position. When all three positioning methods fail, the unmanned vehicle suspends operation, reports to the cloud, and coordinates with maintenance personnel for on-site calibration. At the same time, the cloud records the fault location and time for subsequent positioning mechanism optimization. The relative position deviation between the vehicle and the cargo box is used as the arbitration basis. All positioning data is unified to the vehicle's local coordinate system. When any sensor fails, it automatically degrades to single-sensor positioning and reports it. The traditional PCL point cloud registration algorithm is enhanced with cargo box contour features to remove irrelevant point cloud interference from the ground and fences, retaining only the rigid contour of the cargo box for iterative fitting, improving the robustness of position recognition in weak texture and multi-obstacle scenarios. The visual positioning result and the point cloud positioning result are weighted and fused with a weight of 0.6:0.4 to finally output the absolute pose of the integrated trailer hitch.

[0049] S21.3 Unmanned vehicle uncoupling task: The uncoupling position is specified by the cloud scheduling system, and the unmanned vehicle can directly trust the high-precision location information issued in the task.

[0050] After obtaining the high-precision position of the swapping device, the autonomous vehicle virtually generates a corresponding parking space model on a high-precision map, marks the deviation compensation range, and adapts to subsequent attitude adjustment requirements, laying the foundation for the next step of building a Hybrid A Searching maps and planning berthing routes provide accurate target references.

[0051] S22, Multi-source fusion Hybrid A Search map construction: After completing the virtual parking space modeling, the unmanned logistics vehicle uses Hybrid A A drivable path is planned using RS curves, constructing an accurate search map to ensure the safety and feasibility of the parking path. This map is represented in a two-dimensional grid format, with a grid size of 0.1m × 0.1m to balance positioning accuracy and algorithm efficiency. Each grid cell is marked as drivable or impassable, constraining Hybrid A. The search space during node expansion.

[0052] The search map is generated by integrating information from multiple sources to ensure its integrity and accuracy.

[0053] (1) High-precision map lane topology information: Provides the structured road network skeleton of the site, including lane line positions, lane connection relationships and intersection topology, as the underlying prior constraints of the search map, defining the basic drivable range of vehicles.

[0054] (2) Pre-defined drivable areas in high-precision maps: The boundaries of drivable areas, such as site fences and restricted areas, are pre-marked and verified in high-precision maps and used to directly mark them as impassable areas in raster maps, thus constraining the search space from a global perspective.

[0055] (3) Perceive static obstacle information: Static obstacles such as cones and temporarily parked vehicles are detected in real time by vehicle-mounted sensors and their occupied grid areas are marked as impassable, so as to realize dynamic updates of the search map and enable the planned path to avoid real-time environmental risks.

[0056] (4) Occupied grid map: The gridded environmental perception result constructed based on sensor data directly provides the occupancy status of each grid unit. It is fused and superimposed with the above information to supplement the obstacle information not covered by the structured map and improve the completeness of the search map.

[0057] After unified coordinate alignment and rasterization fusion, the aforementioned multi-source data generates a comprehensive search map that combines prior structured constraints with real-time dynamic perception. Hybrid A The algorithm performs node expansion and collision detection on the map, thereby efficiently searching for a collision-free and kinematically feasible parking path from the current vehicle pose to the target virtual parking space in complex parking scenarios, laying the foundation for the generation and execution of subsequent optimized parking paths.

[0058] S23. Optimized Parking Path Search and Generation: The unmanned logistics vehicle adopts a Hybrid A Algorithm + RS curve joint planning of berthing path, balancing kinematic constraints, search efficiency and path smoothness, Hybrid A The algorithm is a hybrid of the Algorithm A* and uses the Ritz-Shepp curve, which is the analytical solution for the shortest path that satisfies the vehicle's minimum turning radius constraint. It consists of straight line segments and circular arc segments, supporting both forward and backward movement of the vehicle and enabling rapid and precise alignment of the target pose. This invention performs collision prediction and safety margin correction on the standard RS curve, actively expanding the safety boundary by 0.5m after curve generation, rather than directly using the original shortest path, thus improving parking reliability. The specific execution flow is as follows.

[0059] (1) Node Expansion Strategy: During each node expansion, the algorithm simulates the forward and backward movements of the vehicle along six discrete steering directions based on the current node pose, including turning left, moving forward, turning right, turning left in reverse, turning right in reverse, and reversing, generating candidate successor nodes. The six expansion directions correspond to different discrete values ​​of the front wheel steering angle, covering the control range from the maximum left turn to the maximum right turn, achieving a balance between search efficiency and path quality. Each candidate node must pass collision detection: based on the vehicle's geometric envelope, the algorithm verifies whether the vehicle overlaps with an impassable area under the node pose in the search map, and only retains legal nodes without collisions to enter the open list. Turning left in forward corresponds to -35°, moving forward corresponds to 0°, turning right in forward corresponds to +35°, turning left in reverse corresponds to -35°, turning right in reverse corresponds to +35°, and reversing corresponds to 0°. The angle can be finely adjusted by ±2° according to the minimum turning radius of different vehicle models to ensure the standardization and feasibility of node expansion.

[0060] (2) Cost function design: The total cost f(n) of a node consists of the actual cost g(n) and the heuristic cost h(n).

[0061] ; .

[0062] In addition to the usual path length cost, direction switching penalty, and steering angle change penalty, the actual cost g(n) also introduces the distance cost to the impassable area: when a node is close to an obstacle or impassable boundary, an additional penalty weight is applied to guide the search path to actively move away from the obstacle boundary, so as to reserve sufficient safety margin for actual driving. The cost is the path length, used to constrain the overall path length and avoid redundant travel. For the number of direction switching, This is the penalty coefficient for direction switching, used to penalize frequent turns and improve the smoothness of the driving path; This represents the change in steering angle. This is the penalty coefficient for changes in steering angle, used to adapt to vehicle kinematic constraints and avoid exceeding the vehicle's steering limits; The cost of distance from impassable areas, This is the distance cost penalty coefficient, applied when a node is close to an obstacle or impassable boundary. As the value increases, through Applying additional penalty weights guides the search path to actively move away from obstacle boundaries, reserving sufficient safety margin for actual driving (≥0.5m). This addresses the technical pain point of existing path planning technologies that only consider no collisions and do not actively reserve safety margins, thereby improving the safety and practicality of parking paths. , and The specific value is: =0.5、 =0.8、 =1.2, a standard value, which can be dynamically adjusted according to the vehicle model. These are the coordinates of the current node. These are the coordinates of the target parking space.

[0063] (3) Reeds-Shepp Curve Fast Termination: During node expansion, when the distance between the current expanded node and the target parking space is less than the 8m threshold, the algorithm attempts to directly connect the current node and the target pose using an RS curve. The RS curve is the shortest path analytical solution that satisfies the minimum turning radius constraint of the vehicle. It is composed of straight line segments and circular arc segments and supports forward and backward movement. If the generated RS curve is verified to be collision-free by collision detection and its distance from the impassable area meets the safety margin requirement, then the curve is used as the final path segment from the current node to the target, and the search process is terminated. This strategy significantly improves the search efficiency near the target pose and avoids redundant node expansion in the target area.

[0064] This invention addresses the traditional Hybrid A The algorithm is improved in three aspects: 1) the discrete steering angle is optimized to a dedicated 6-direction control set; 2) a dynamic safety margin penalty term is introduced into the cost function; 3) in dynamic obstacle scenarios, it is adaptively extended to an 8-direction search and the weights are dynamically adjusted. This makes it more suitable for unmanned logistics vehicle docking scenarios, rather than a general parking algorithm.

[0065] Through the above process, a smooth, collision-free, and vehicle-steering-compliant complete parking trajectory is generated, providing stable path support for subsequent dynamic parking execution, safe obstacle avoidance, and precise docking.

[0066] This technical solution employs a differentiated positioning strategy, using appropriate positioning methods for different types of trailer hitchs. In particular, the multi-positioning mechanism for integrated devices controls positioning errors within ±3.0mm, significantly improving positioning accuracy. The multi-source fusion-based search map balances prior constraints and real-time perception, ensuring map integrity and accuracy. A planning strategy with six discrete steering directions and RS curve termination clarifies angle values ​​and fine-tuning rules. Combined with a cost function containing safety margins, the generated parking path is smooth and collision-free, satisfying vehicle kinematic constraints while improving search efficiency. Compared to existing path planning methods, parking time is reduced by 20%, effectively mitigating safety risks during parking and laying a solid foundation for subsequent precise docking.

[0067] Example 3.

[0068] Reference Figure 5 This embodiment 3 discloses a method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method, wherein step S3 includes the following steps.

[0069] S31. Parking Speed ​​Planning and Motion Trajectory Prediction: After obtaining Hybrid A After determining the parking path, the unmanned logistics vehicle further performs speed planning to ensure that it can safely handle dynamic obstacles during parking, thus improving parking safety. The process is as follows.

[0070] (1) Motion trajectory prediction: Based on the vehicle orientation θ and the corresponding front wheel turning angle δ of each path point on the planned path, combined with the vehicle kinematic model, the expected area occupied by the vehicle at different times is calculated, and a motion trajectory with timestamp is generated.

[0071] (2) Dynamic collision detection: Spatiotemporal collision detection is performed between the vehicle's expected trajectory and the predicted trajectory of dynamic obstacles output by the onboard sensors to determine whether there is a collision risk within a time window of 3 to 5 seconds. 3 seconds is used for normal parking scenarios and 5 seconds is used for complex obstacle scenarios.

[0072] (3) Adaptive speed adjustment: If a collision risk is detected within a certain time period, the autonomous vehicle will actively reduce its speed until it stops, wait for the dynamic obstacle to pass through the conflict area, and then resume normal parking speed to continue the parking action. This strategy ensures the safety of the parking process, avoids unnecessary path replanning, and works in conjunction with the next step of spatiotemporal dynamic collision detection to further enhance parking safety.

[0073] S32. Spatiotemporal Dynamic Collision Detection and Adaptive Speed ​​Adjustment: During parking operations, the unmanned logistics vehicle continuously synchronizes with the expected vehicle trajectory generated in stage S31. Combined with real-time updates of dynamic obstacle trajectories from onboard sensors, high-frequency spatiotemporal collision detection is performed to further improve the accuracy and timeliness of collision risk identification, ensuring a safe and controllable parking process. Dynamic obstacles include pedestrians and temporarily moving vehicles. During detection, the vehicle focuses on monitoring changes in obstacle movement. If a collision risk is detected, the adaptive speed adjustment strategy from S31 is used to actively reduce speed until the vehicle stops, waiting for the obstacle to pass the conflict area before resuming normal parking speed, avoiding repeated speed adjustment procedures. If a dynamic obstacle remains on the parking path for an extended period, exceeding 30 seconds, and cannot be avoided by stopping and waiting, the next step of the dynamic path correction mechanism is automatically triggered, connecting to step S33 to achieve dynamic obstacle avoidance and path optimization, ensuring that overall operational efficiency is not affected.

[0074] S33. Path Correction in Continuous Dynamic Obstacle Scenarios: If a dynamic obstacle remains on the parking path for an extended period, exceeding 30 seconds, and cannot be avoided by stopping and waiting, the unmanned logistics vehicle automatically triggers a dynamic path correction mechanism, combined with the improved Hybrid A... The algorithm cost function formula enables dynamic obstacle avoidance and path optimization. The specific process is as follows.

[0075] (1) The vehicle-mounted sensors update the dynamic obstacle trajectory, speed and dwell time in real time, and synchronously update the impassable grid area of ​​the search map and mark the obstacle movement trend prediction area; the vehicle-mounted sensors include high-definition cameras and lidar, etc.

[0076] (2) Readjust Hybrid A The algorithm's node expansion strategy increases the number of expansion directions from six to eight: forward (-35°, -15°, 0°, +15°, +35°) and backward (-35°, 0°, +35°), improving the flexibility of path search and expanding the node expansion range to avoid missing optimal paths. A small forward left turn corresponds to -15°, and a small forward right turn corresponds to +15°, which, combined with the original six directions, further enhances path search flexibility. The turn angle values ​​can be dynamically fine-tuned based on the vehicle model.

[0077] (3) Dynamically adjust the weights of the cost function. The dynamic adjustment model of the cost function is as follows.

[0078] .

[0079] . .

[0080] . .

[0081] in: It is the comprehensive cost of node n, a core decision-making indicator; This is the actual cost from the starting point to node n, including path length cost, direction switching penalty, steering angle change penalty, and distance cost from impassable areas. For path length, For the number of direction switching, This represents the change in steering angle. This corresponds to the penalty coefficient; The cost of distance from impassable areas, This represents the distance cost penalty coefficient. It is the heuristic cost from node n to the target parking space, calculated using Manhattan distance. Let n be the coordinates of node n. The coordinates of the target parking space; It is the distance cost from node n to the nearest impassable region. The minimum distance from node n to the obstacle. The minimum safe distance is set at 0.5m. It's distance cost weighting, in typical scenarios. Dynamic obstacle scenarios are improved by 30%, that is ; It is the risk and cost of dynamic obstacles. For dynamic obstacle speed, Let n be the predicted collision time between node n and the obstacle. For dynamic risk cost weighting, in conventional scenarios Dynamic obstacle scenarios are improved by 30%, that is . It's a heuristic cost weighting, in typical scenarios. =1.0, Dynamic Obstacle Scene =1.2, used to dynamically adjust the proportion of heuristic cost in the overall cost, to adapt to path search needs in different scenarios.

[0082] By dynamically adjusting the model using the aforementioned cost function, the model prioritizes avoiding the predicted movement area of ​​obstacles while retaining the ability to quickly return to the original berthing trajectory after obstacles have passed. This ensures that dynamic obstacle avoidance does not affect berthing efficiency, addressing the pain point of berthing interruption caused by dynamic obstacles in existing technologies. After completing the dynamic berthing execution, the model enters the fourth stage of automatic trailer docking and loading / unloading execution.

[0083] This technical solution addresses the problems of inaccurate dynamic collision detection, operation interruptions caused by dynamic obstacles, and low obstacle avoidance efficiency in existing berthing processes. It employs dynamic collision detection with a differentiated time window of 3-5 seconds to adapt to different scenario requirements, improving the accuracy and timeliness of collision risk identification. High-frequency detection combined with a speed adjustment strategy avoids repeated speed adjustments, improving obstacle avoidance efficiency and reducing unnecessary path replanning. When a dynamic obstacle lingers for more than 30 seconds, real-time path correction is achieved through node expansion direction optimization and dynamic adjustment of cost function weights. This prioritizes obstacle avoidance while quickly returning to the original trajectory, ensuring continuous and controllable berthing operations. It solves the pain point of operation interruptions caused by dynamic obstacles in existing technologies, reducing the berthing interruption rate to below 5% and significantly improving the safety and efficiency of the berthing process.

[0084] Example 4.

[0085] This embodiment 4 discloses a method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method, wherein step S4 includes the following steps.

[0086] S41. Confirmation of precise parking of unmanned vehicle: The unmanned vehicle completes precise parking according to the corrected parking trajectory. After parking, it uses onboard sensors to detect the relative position deviation between itself and the trailer hitch, including vertical, horizontal and angular deviations, to confirm that the parking position meets the docking requirements. Then, it reports the parking completion status to the cloud scheduling system through the communication interface, triggering the cloud scheduling system to start the subsequent trailer hitch type adaptation and attitude pre-adjustment operation.

[0087] S42. Adaptation and attitude pre-adjustment of the trailer swapping device: After receiving the arrival signal of the unmanned vehicle, the cloud dispatching system automatically triggers the corresponding automatic trailer swapping device to perform loading and unloading operations. First, it adapts and adjusts the device according to the device type to ensure that the device attitude matches the vehicle docking requirements.

[0088] (1) Fixed trailer swapping device: This type of device is fixed on the ground or in a specific location and has the ability to accommodate and handle multiple cargo boxes. It is usually linked with automatic sorting, warehouse scheduling and other systems to form an efficient unmanned loading and unloading line in scenarios such as transit yards or distribution centers. At this time, the device does not need to move, but only needs to adjust its loading and unloading mechanism posture according to the task instructions to adapt to the vehicle docking height and angle. The fixed trailer swapping device adopts existing mature products. The model can be CIMC Lingyu ZJV9400TJZ, which is suitable for heavy unmanned logistics vehicles, supports automatic locking docking, and is suitable for warehouse line linkage; or Fuhua Machinery FHW-1200, a light fixed trailer swapping device, which is suitable for small and medium-sized logistics transit scenarios and has high operation precision.

[0089] (2) Integrated trailer swapping device: This type of device is designed as an integrated unit with the cargo box, meaning the trailer swapping device itself constitutes part of the cargo box. It can be used as a standard cargo box for batch loading and unloading by a fixed trailer swapping device, adapting to large-scale operation scenarios in transit hubs. It also has strong flexibility, not only able to independently complete vehicle loading and unloading operations, but also adapting to network points where fixed equipment cannot be installed. After completing the device type adaptation and attitude pre-adjustment, the core adaptive attitude correction and precise docking steps are entered. The integrated trailer swapping device adopts existing mature products. The model can be FAW Jiefang J6P integrated trailer swapping box, which is adapted to heavy-duty unmanned logistics vehicles, supports automatic support and docking, and integrates a positioning module; or Dongfeng Commercial Vehicle DFH5180TSZ, a lightweight integrated trailer swapping device, which is highly flexible and adaptable to last-mile delivery network points without fixed equipment.

[0090] S43. Adaptive Attitude Correction and Precise Docking: Achieves precise alignment, attachment, and unattachment. Through adaptive attitude correction, it ensures precise docking between the cargo box and the vehicle without manual intervention. The specific process is as follows.

[0091] (1) Deviation Detection: The horizontal, vertical, and angular deviations between the cargo box and the vehicle saddle are collected in real time using a laser rangefinder and a vision sensor. The sampling frequency is 10.0Hz during the attitude correction phase, 2.0Hz during normal monitoring, and 3.0Hz during docking verification. The frequency is dynamically switched according to the phase to ensure that the deviation data is accurate in real time. When the deviation data collected by the two sensors exceeds 1.0mm, the average value is taken as the actual deviation. When the deviation exceeds 2.0mm, attitude correction is paused, data is collected again, and the valid data is taken after comparison. When one sensor fails, the backup sensor is activated. For example, if the laser sensor fails, the vision sensor is activated, and the sampling frequency is reduced to 5Hz. At the same time, the sensor fault information is reported to the cloud to ensure continuous operation.

[0092] (2) Graded PID adaptive correction: The PID adaptive correction algorithm is adopted, combined with the exclusive deviation compensation formula, to realize the three-dimensional deviation graded closed-loop correction. The core correction formula is as follows.

[0093] .

[0094] in: It is the correction control quantity at time t, used to adjust the attitude of the detachment device / vehicle; It is the deviation value at time t, including horizontal / vertical / angular deviation. ; It is the proportionality coefficient, the horizontal deviation. Vertical deviation , angle deviation ; These are integral coefficients, and should be uniformly taken as follows: , used to eliminate steady-state deviations; These are the differential coefficients, and should be uniformly taken as... Used to suppress correction overshoot. Control quantity Limits: displacement ±5mm / cycle, angle ±0.5° / cycle. The integral term is limited within the range of [-2.0, 2.0]. When the value exceeds the limit range, the extreme value remains unchanged to suppress integral saturation.

[0095] The tiered correction process is as follows: First, correct the vertical deviation by adjusting the cargo box height using the hydraulic outriggers. Calculate the extension / retraction control amount using the formula above, adjusting with an accuracy of ≤0.1mm / time, until the vertical deviation is ≤2mm. Next, correct the horizontal deviation by finely adjusting the saddle position left and right using the saddle lateral fine-tuning mechanism, adjusting the stroke from 0-10mm, until the horizontal deviation is ≤2mm. Finally, correct the angular deviation by adjusting the saddle tilt angle using the saddle angle fine-tuning mechanism, until the angular deviation is ≤0.3°. Simultaneously, a deviation dead zone threshold is set: when the detected horizontal and vertical deviations are less than 0.1mm and the angular deviation is less than 0.05°, the control module does not output a correction action, avoiding repeated adjustments for small deviations that could cause mechanism vibration and improving docking stability.

[0096] (3) Real-time feedback and verification: During the correction process, the laser ranging sensor collects deviation data 3 times per second and feeds it back to the control module. Update the deviation value to form a closed-loop adjustment of detection-correction-verification, ensuring that the final docking deviation is compensated to within ±3.0mm.

[0097] (4) Scene adaptation optimization: For scenes with uneven ground, ground flatness is detected, and the force on each outrigger is detected by pressure sensors. Dynamically adjust the outrigger extension / retraction range. For the i-th outrigger, The elastic coefficient, The system measures the outrigger extension range to ensure the cargo box is level when docking with the vehicle. For minor vehicle parking errors or cargo box misalignment, it automatically matches a pre-set correction parameter library to shorten correction time, with a total correction time ≤ 5 seconds. Once the attitude correction meets the docking standard, the system immediately executes the next automatic attachment or detachment operation. The parameter library contains PID coefficient fine-tuning values ​​corresponding to different horizontal, vertical, and angular deviation ranges, adapting to deviation correction in different scenarios.

[0098] S44. Automatic Hook-and-Unhook Execution: After attitude correction, the automatic hook-and-unhook device performs the corresponding hook-and-unhook operation according to the loading / unloading task type: During loading, the device precisely mates the pre-prepared cargo box with the vehicle's saddle and locks it in place; during unloading, the device unlocks the saddle locking mechanism, separating the cargo box from the vehicle and placing it in the designated location. The entire process requires no manual intervention, realizing the core operation of automatic hook-and-unhook docking. During loading and unloading, the device's status monitoring and fault self-healing mechanism are simultaneously activated, entering the S5 stage of the operation process.

[0099] This technical solution addresses the problems of large docking deviations, poor device adaptability, excessive manual intervention, and low docking efficiency in existing methods. Through a device type adaptation mechanism, it adjusts the attitude of both fixed and integrated devices to adapt to different operating scenarios, expanding the application scope of automatic docking technology. The application of a graded PID adaptive correction algorithm clarifies the PID coefficients corresponding to each deviation, achieving three-dimensional closed-loop correction in a vertical-horizontal-angle sequence. Combined with scenario adaptation optimization and a preset parameter library, the docking deviation is controlled within ±3.0mm, and the total correction time is ≤5s. Compared to existing docking methods, docking accuracy and efficiency are improved by more than 40%. The entire process requires no manual intervention, achieving precise docking between the cargo box and the vehicle, avoiding human error, reducing labor costs, and adapting to different vehicle models and cargo box types, thus improving the versatility and practicality of the technical solution.

[0100] Example 5.

[0101] This embodiment discloses a method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method, wherein step S5 includes the following steps.

[0102] S51. Equipment Status Monitoring and Fault Self-Healing: During loading and unloading operations, the equipment's built-in status monitoring module employs a multi-sensor fusion monitoring method to collect four types of core operating data in real time, including hydraulic system pressure, locking mechanism torque, outrigger extension accuracy, and communication link signal strength. The overall sampling frequency of the status monitoring module is 2 times / second, which is increased to 3 times / second when the laser rangefinder is used for deviation detection. After the deviation detection is completed, it automatically returns to the overall sampling frequency.

[0103] The core function of the status monitoring module is to provide accurate data support for the fault self-healing mechanism, realizing real-time monitoring, data acquisition, and anomaly identification of the device's operating status. Its core components and data processing flow are as follows: The status monitoring module includes a laser rangefinder, a pressure sensor, a hydraulic sensor, a communication status sensor, an edge computing module, a data storage unit, and an early warning triggering unit, realizing the integration and adaptation of monitoring functions with the device body; Data processing flow: Real-time operating data collected by sensors is first transmitted to the edge computing module for noise reduction to remove abnormal interference data. Then, the processed data is compared with preset standards, and abnormal data is quickly identified through fault warning formula. At the same time, the monitoring data is synchronized to the cloud scheduling system in real time, forming a complete data link of acquisition-processing-identification-synchronization. This provides core data support for subsequent fault warning and self-healing assessment, enabling 24-hour uninterrupted monitoring with a data sampling error of ≤0.1%. Based on the safety threshold of each monitoring parameter, the anomaly identification response time is ≤0.5s.

[0104] Simultaneously, real-time data analysis is performed through the edge computing module, combined with fault diagnosis models and early warning formulas to achieve early warning of faults, graded diagnosis, and self-healing. The specific process is as follows.

[0105] (1) Fault warning: Preset safety thresholds for each monitoring parameter, such as hydraulic pressure 0.8-1.2MPa, locking torque ≥50N·m, outrigger extension accuracy ≤0.2mm, and communication signal strength ≥-80dBm. Calculate the warning index through the fault warning formula. When the index reaches the warning threshold, the warning will be automatically triggered.

[0106] .

[0107] in: It is a fault warning index, with a value range of 0 to 1. The Level 1 warning has been triggered, and the threshold is approaching. A minor anomaly was triggered, triggering a Level 2 warning. It is the weight of the j-th type of monitoring parameter, hydraulic pressure. Lock-up torque outrigger extension accuracy Communication signal strength ; It is the real-time value of the j-th type of monitoring parameter; It is the lower safety limit value of the j-th type of monitoring parameter; This is the safety upper limit value for the j-th type of monitoring parameter. When the data approaches the threshold, an early warning is automatically triggered, and the warning information is reported to the cloud. At the same time, abnormal data is recorded to provide a basis for subsequent troubleshooting.

[0108] (2) Self-healing for minor faults: If a minor fault is detected, such as incomplete locking, low hydraulic pressure, or brief interruption of communication signal, the corresponding self-healing command will be automatically triggered. The self-healing effect evaluation formula is as follows.

[0109] .

[0110] in: To increase the success rate of self-healing, Considered a successful self-healing. These are the parameter values ​​when a fault occurs. These are parameter values ​​after self-healing. These are normal parameter values; specific self-healing operations: If locking is not fully engaged, the locking mechanism is controlled to re-execute the locking operation until the torque reaches the safe threshold; if the hydraulic pressure is low, the hydraulic pump is started to replenish hydraulic oil until the pressure returns to the safe range; if the communication signal is briefly interrupted, the backup communication link is automatically switched, with redundancy between the CAN bus and Ethernet to ensure communication continuity. When the communication signal strength remains below -80dBm for 3 seconds, the system automatically switches to the backup link, with a switching delay of ≤0.3s.

[0111] (3) Handling serious faults: If a serious fault is detected, such as hydraulic system leakage, locking mechanism damage, outrigger jamming, and communication link interruption for more than 10 seconds, the fault level is marked by the fault level determination model.

[0112] (when )or (when ).

[0113] in: This is a fault level index. This is a Level 1 fault and requires immediate attention. For level 2 faults, processing can be delayed; upon detection of a serious fault, loading and unloading operations should be stopped immediately, the cargo container should be locked to prevent it from falling, and the fault level should be reported to the cloud.

[0114] (4) Emergency Response: After receiving information about a serious fault in the cloud, the system automatically connects nearby available vehicles to take over the current task based on the fault level and the station's capacity. At the same time, a maintenance work order is generated, specifying the fault location, fault type, and handling suggestions, and pushed to the maintenance personnel's terminal. The maintenance personnel can remotely view the device's operating data through the cloud, prepare repair parts in advance, shorten maintenance time, and avoid interruption of the logistics chain. If no fault occurs or a minor fault heals successfully on its own, the loading and unloading operation is completed, and the task closure and feedback step S52 is entered.

[0115] S52. Task Closed Loop and Feedback: After the automatic swapping device completes the loading and unloading operation, it will feed back the operation status and result information to the cloud scheduling system to realize task feedback and closed-loop control. After receiving the feedback, the cloud scheduling system records the task execution status and updates the status, and generates new task instructions according to the configured business logic or rules. The specific feedback processing mode is as follows.

[0116] (1) If it is a delivery task, the cloud can automatically notify the client system.

[0117] (2) If it is a transportation transit task, the cloud can directly trigger the next stage of transportation or loading and unloading tasks.

[0118] (3) If an anomaly is detected, such as task failure or abnormal equipment status, the cloud will report to the operation and maintenance module and can automatically generate manual intervention instructions. At this point, a single loading and unloading task forms a complete closed loop. The coordinated execution of the above steps jointly realizes the unmanned, efficient and safe loading and unloading process of unmanned logistics vehicles, corresponding to the realization of the following technical effects.

[0119] This technical solution addresses the problems of untimely fault handling, lack of self-healing capabilities, easy interruption of logistics links, and low operation and maintenance efficiency in existing systems. A multi-sensor fusion status monitoring module enables 24 / 7 uninterrupted monitoring with a data sampling error ≤0.1% and an anomaly detection response time ≤0.5s, quickly detecting equipment malfunctions. A tiered early warning and self-healing mechanism automatically triggers self-healing operations for minor faults, achieving a success rate ≥98% and preventing minor faults from escalating. For serious faults, emergency linkage measures such as emergency locking, idle vehicle replacement, and maintenance work order pushes shorten fault handling time and prevent logistics link interruptions. A closed-loop task feedback mechanism ensures full traceability and controllability of operations, timely synchronization of task progress with relevant parties, improving operational reliability and fault tolerance, reducing the failure rate and operation and maintenance costs of logistics operations, and ensuring the stable operation of the entire unmanned logistics loading and unloading process.

[0120] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for loading and unloading goods on unmanned logistics vehicles based on a swapping mechanism, characterized in that, Includes the following steps: Collect operational data of unmanned logistics vehicles, status data of trailer swapping devices, and cargo storage data; A cargo box vehicle matching degree calculation model is constructed. The input of the cargo box vehicle matching degree calculation model is the unmanned logistics vehicle operation data, the trailer swapping device status data and the cargo box storage data. The output is the quantitative matching score between the cargo box and the unmanned logistics vehicle. Loading and unloading scheduling tasks are generated based on the quantitative matching score. After receiving the loading and unloading scheduling task, the unmanned logistics vehicle generates a collision-free driving trajectory based on the unmanned logistics vehicle operation data, the status data of the trailer swapping device, and the cargo box storage data. The unmanned logistics vehicle completes dynamic parking along the collision-free driving trajectory; the trailer swapping device performs the docking and undoing operations of the cargo box and the unmanned logistics vehicle, as well as cargo loading and unloading operations.

2. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 1, characterized in that... The generation of the collision-free driving trajectory includes the following steps: When the unmanned logistics vehicle drives to the preset area of ​​the loading and unloading point, it is positioned according to the type of the trailer swapping device. Combining the unmanned logistics vehicle operation data and the trailer swapping device status data, a virtual parking space model is constructed on the map as the target benchmark for path planning. Based on the target benchmark, a two-dimensional grid search map is constructed, marking passable and impassable areas; Combining the search map, a hybrid A* algorithm and Ritz-Shepp curve are used to plan the trajectory of the unmanned logistics vehicle.

3. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 2, characterized in that... The trailer-swapping device includes a fixed trailer-swapping device and an integrated trailer-swapping device; the trailer-swapping device is communicatively connected to the cloud-based dispatch system to receive location commands and feedback status information; Building a virtual parking space model on a map involves the following steps: For the fixed trailer swapping device, the unmanned logistics vehicle uses the location information issued by the task to construct a virtual parking space model; For the integrated trailer swapping device, the location is determined by cargo box positioning, historical location retrieval and vehicle-side perception positioning mechanism, and a virtual parking space model is constructed; For the task of unloading, the unmanned logistics vehicle uses the unloading location information specified in the cloud to build a virtual parking space model.

4. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 3, characterized in that... The positioning method of the integrated trailer-slinging device includes: Positioning is achieved using onboard sensors, including an onboard high-definition camera and a lidar. The markings on the surface of the cargo box are captured by the vehicle-mounted high-definition camera, and then the marking information is analyzed by a corresponding algorithm; The outline of the cargo box is scanned by the LiDAR, and the spatial coordinates of the cargo box are obtained by the point cloud processing algorithm library based on the scan data. The location of the cargo box is determined by fusing the identification and analysis information obtained from the vehicle's high-definition camera with the spatial coordinates of the cargo box acquired by the lidar.

5. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 2, characterized in that... The search map is generated by fusing lane topology information, preset drivable areas, perceived static obstacle information, and occupied grid map, and then merging them through coordinate alignment and rasterization. The lane topology information and preset drivable areas are obtained through the map, the perceived static obstacle information is obtained through vehicle-mounted sensor detection, and the occupied grid map is generated through vehicle-mounted sensor scanning and point cloud processing.

6. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 2, characterized in that... The method of using a hybrid A* algorithm and Ritz-Shep curve to plan the trajectory of the unmanned logistics vehicle includes the following steps: The parking path is planned using a hybrid AlphaGo algorithm and Ritz-Shepp curves. The Ritz-Shepp curves consist of straight segments and circular segments, which meet the minimum turning radius requirements of vehicles and support both forward and reverse movement of vehicles. Collision prediction and safety boundary expansion correction are performed on the Ritz-Shepp curve; The node expansion adopts multiple discrete steering directions, and the corresponding front wheel steering angle can be finely adjusted according to the vehicle model; A weighted penalty mechanism is used to calibrate the path.

7. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 1, characterized in that... After generating the collision-free driving trajectory, the process also includes a safety obstacle avoidance step, which specifically includes the following steps: Parking speed planning and trajectory prediction: Based on the collision-free driving trajectory, parking speed planning is performed to generate the vehicle's expected trajectory with timestamps; Spatiotemporal dynamic collision detection and adaptive speed adjustment: Combine the dynamic obstacle trajectory detected by the vehicle's onboard sensors to carry out collision detection. When a collision risk is detected, the unmanned logistics vehicle actively reduces its speed until it stops. After the obstacle passes, it resumes its normal parking speed. A cost function dynamic adjustment model is constructed; the input of the cost function dynamic adjustment model is the dynamic obstacle trajectory and the expected vehicle motion trajectory, and the output is the cost function weight adjustment parameters. Dynamic obstacle path correction: Based on the cost function weight adjustment parameters, according to the dynamic obstacle trajectory collected in real time by the vehicle sensor, update the impassable grid area in the search map, perform path correction adaptation, adjust the expansion direction of the hybrid A* algorithm nodes, and perform path correction and obstacle avoidance in dynamic obstacle scenarios.

8. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 1, characterized in that... The cargo box vehicle matching degree calculation model is as follows: ； ； ； ； ； ； ； in: It is the total score of the matching degree between the cargo box and the unmanned logistics vehicle. For load adaptability weight, Weighting for size fit. As a weight for battery life matching, As the path fit weight, For priority matching degree weight, The load-bearing adaptability score is given. Score for size fit. The score is based on the battery life matching. Score the path fit. Score the priority matching degree; This refers to the weight of the cargo box. This is the maximum load capacity for unmanned logistics vehicles; These are the length, width, and height of the cargo box, respectively. The length, width, and height of the cargo box for unmanned logistics vehicles are adapted accordingly. For the vehicle's remaining range, For transportation distance; The distance from the vehicle's current location to the loading / unloading point. The optimal distance from the vehicle to the loading / unloading point; This represents the task priority coefficient.

9. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 7, characterized in that... The cost function is dynamically adjusted as follows: ； ； ； ； ； in: It is the total cost of node n. It is the actual cost from the starting point to node n. For path length, For the number of direction switching, This represents the change in steering angle. This corresponds to the penalty coefficient; The cost of distance from impassable areas, This represents the distance cost penalty coefficient. It is the heuristic cost from node n to the target parking space. The coordinates of the target parking space; It is the distance cost from node n to the nearest impassable region. The minimum distance from node n to the obstacle. For minimum safe distance, It is the distance cost weight. It is the risk and cost of dynamic obstacles. For dynamic obstacle speed, Let n be the predicted collision time between node n and the obstacle. For dynamic risk cost weighting, It is a heuristic cost weight; These are the coordinates of the current node. These are the coordinates of the target parking space.

10. The method for loading and unloading goods on an unmanned logistics vehicle based on a swapping method according to claim 1, characterized in that... The aforementioned trailer-swapping device performs the docking of the cargo box with the unmanned logistics vehicle and the loading and unloading of goods, including the following steps: After the unmanned logistics vehicle is parked, the onboard sensors detect the relative positional deviation between it and the trailer hitch. The type of trailer swapping device is adapted and adjusted accordingly. Fixed trailer swapping devices adjust their own posture to adapt to the docking, while integrated trailer swapping devices are linked and adapted to the cargo box. Deviation data is collected by onboard sensors, and proportional-integral-derivative adaptive correction of the deviation is used. When the quantitative matching score reaches the preset threshold, the automatic docking step is triggered. After docking is completed, the automatic attachment or detachment operation is performed. If the threshold is not met, the task is reassigned.

Images