AGV narrow channel accurate picking method based on multi-level reference point series control

Abstract

The present application relates to AGV narrow channel accurate picking technical field, specifically disclose a kind of AGV narrow channel accurate picking method based on multistage fiducial point series connection control, including open space navigation module, narrow channel navigation module and warehouse entry navigation module;Wherein open space navigation module includes: for AGV dolly receives the command receiving module of picking instruction from upper system;For the point cloud map information call module of target information and first fiducial point position from point cloud map;According to the target information and first fiducial point position of point cloud map information call module call, plan a global path planning module from AGV dolly current position to the global edge path of first fiducial point.The application is not based on the navigation of fixed beacon, and thus can avoid the problem of lack of flexibility, and can ensure terminal accuracy, in addition, AGV can also ensure adaptability in structured narrow lane, so that AGV automatically adapts to the situation of narrow channel, to ensure that AGV accurate picking in narrow channel.

Description

Technical Field

[0001] This invention relates to the field of detection and ranging system technology, specifically to a method for precise cargo retrieval in narrow channels of AGVs based on multi-level reference point serial control. Background Technology

[0002] In modern industries such as intelligent manufacturing, warehousing and logistics, and e-commerce fulfillment, automated logistics systems have become core infrastructure for improving operational efficiency, reducing labor costs, and ensuring operational accuracy. Automated guided vehicles (AGVs), as key execution units for automated material transfer, are widely used in scenarios such as shelf picking, workstation feeding, and finished product delivery due to their advantages such as autonomous navigation, flexible scheduling, and all-weather operation. Especially in scenarios with stringent requirements for efficiency and accuracy, such as high-density warehousing and high-speed production, the performance of AGV picking systems directly determines the turnover efficiency and operating costs of the entire logistics chain.

[0003] As industrial upgrading demands increasingly higher levels of logistics automation, traditional AGV picking methods are gradually revealing numerous technical bottlenecks, making it difficult to meet the needs of efficient operations in complex scenarios. Firstly, in terms of picking and positioning accuracy, traditional AGVs mostly rely on fixed-path navigation methods such as magnetic strips and QR codes, and their picking point positioning errors are usually at the centimeter level. For picking and placing precision parts and small materials, problems such as picking deviation and material damage are prone to occur. Although trackless navigation technologies such as laser SLAM and visual navigation have been gradually applied, their navigation and positioning stability is insufficient in dynamic environments (such as personnel movement and temporary obstacles), resulting in reduced repeatability and reliability of picking actions.

[0004] Secondly, regarding the flexibility of AGV picking strategies, existing AGV picking methods are mostly designed for materials or shelves with fixed specifications. The picking mechanisms (such as robotic arms and forks) have low degrees of freedom, making it difficult to adapt to the picking and placing needs of materials of different sizes, weights, and shapes (such as boxes, bags, and irregular parts). When material specifications change, picking parameters (such as clamping force and picking height) need to be manually readjusted, which not only reduces operational efficiency but also increases the cost of manual intervention, failing to meet the needs of flexible manufacturing and personalized warehousing.

[0005] Furthermore, in multi-AGV collaborative picking scenarios, traditional scheduling systems often adopt a simple first-come-first-served scheduling logic, lacking dynamic optimization of picking task priority, AGV load status, and path congestion. This leads to problems such as path conflicts, excessive waiting time, and uneven resource utilization during the picking process. Especially in large-scale warehouses or complex production workshops, the low efficiency of multi-AGV collaboration directly affects the throughput of the entire logistics system.

[0006] Furthermore, traditional AGV picking methods lack sufficient intelligence and the ability to adaptively handle abnormal situations during the picking process. For example, when materials are misplaced, shelves are deformed, or the picking mechanism malfunctions, AGVs often cannot autonomously identify and adjust their picking strategies, and can only stop and wait for manual intervention, leading to operational interruptions and affecting the continuity of the logistics chain. At the same time, existing picking methods have limited environmental perception dimensions, relying mostly on single sensors (such as infrared and ultrasound) for obstacle detection, which makes them less adaptable to complex environments (such as strong light, dust, and obstructions), easily causing collision risks and reducing operational safety.

[0007] To address the aforementioned technical challenges, the industry is gradually exploring AGV retrieval solutions that integrate technologies such as high-precision navigation, multi-sensor fusion perception, intelligent scheduling algorithms, and flexible retrieval mechanisms. For example, by using fusion positioning of visual sensors and LiDAR, the positioning accuracy of retrieval points can be improved to the millimeter level. A scalable, multi-degree-of-freedom robotic arm structure can be adopted to adapt to the retrieval and placement requirements of materials of different specifications. A scheduling model based on reinforcement learning or genetic algorithms can optimize the collaborative retrieval path of multiple AGVs. Machine vision recognition technology can be introduced to achieve autonomous correction of material position deviation and real-time early warning of abnormal situations.

[0008] Existing technologies include fixed beacon-based navigation (such as QR codes, magnetic strips, RFID), where magnetic strips are pre-laid on the warehouse floor or QR codes are pasted on the floor. AGVs follow and locate themselves by reading these fixed beacons. Global single path planning uses sensors (such as LiDAR, vision) to detect obstacles in real time and employs algorithms to plan a global path from the starting point to the end point, and tracks the entire path. It also features real-time dynamic obstacle avoidance, local replanning, and terminal visual servo guidance.

[0009] However, the aforementioned existing technologies lack flexibility and have high deployment and maintenance costs: fixed beacon-based navigation technologies rely heavily on pre-laid physical beacons, making changes to routes, stations, or layouts extremely inconvenient and requiring reconstruction. The system has poor flexibility and high deployment and maintenance costs.

[0010] In addition, the existing terminal technology has insufficient accuracy and poor reliability: navigation technology based on full-domain single path planning will inevitably generate positioning accumulation error after running over long distances and in complex environments. This error has little impact in wide areas, but at the entrance of narrow channels, even micron-level pose deviations may cause the AGV to fail to face the storage location, resulting in failure to pick up goods, scratches or even collisions, making it difficult to guarantee the success rate and reliability of the operation.

[0011] Furthermore, existing technologies are not adaptable to structured narrow passages: while real-time dynamic obstacle avoidance technology can effectively deal with dynamic obstacles, its passive and reactive replanning mechanism requires a large amount of computation, and the planned trajectory is unpredictable. In extremely narrow passages where the width of the passage is only slightly larger than the vehicle body, frequent planning can easily lead to uneven movement, shaking, or even jamming, failing to meet the stringent requirements of path repeatability and high predictability for narrow passage operations. To solve the accuracy problem, terminal vision servo technology needs to rely on additional vision sensors and specific markers, increasing system complexity and cost, and its stability is easily affected by environmental factors such as light and occlusion.

[0012] To address the aforementioned issues, there is an urgent need for innovative design based on the existing detection and ranging system. Summary of the Invention

[0013] The purpose of this invention is to provide an AGV narrow-channel precise picking method based on multi-level reference point serial control, in order to solve the problems mentioned in the background art, which are that existing AGV picking relies on fixed beacon navigation, resulting in a lack of flexibility, insufficient terminal accuracy, and insufficient adaptability in structured narrow channels.

[0014] To achieve the above objectives, the present invention provides the following technical solution: an AGV narrow channel precise picking method based on multi-level reference point serial control, including an open space navigation module, a narrow channel navigation module and an inbound navigation module;

[0015] The open space navigation module includes:

[0016] A command receiving module for AGVs to receive picking instructions from the host system;

[0017] A point cloud map information retrieval module for retrieving target information and the location of the first reference point from a point cloud map;

[0018] Based on the target information retrieved by the point cloud map information retrieval module and the location of the first reference point, a global path planning module plans a global edge-keeping path from the current position of the AGV to the first reference point. The global edge-keeping path enables the AGV to travel along open areas first and avoid all narrow spaces.

[0019] The narrow channel navigation module includes:

[0020] This module is used to retrieve digital map information from the location attribute database of a digital map, including the target storage location, its orientation, the precise width of the passageway it belongs to, and the storage location depth.

[0021] This intelligent edge-keeping calculation module automatically calculates the optimal lateral distance based on the width of the passage and the vehicle's own safe operating envelope, ensuring that the AGV maintains a safe distance from obstacles on both sides within the operating passage and avoiding the risk of collision.

[0022] Starting from the position where the AGV has traveled to the first reference point, the optimal lateral distance calculated by the intelligent edge-keeping calculation module is used as a constraint to generate a lateral movement path, which guides the AGV to the second reference point.

[0023] The inbound navigation module is used for the AGV to rotate in place and reverse into the warehouse. The in-place rotation of the AGV includes a status confirmation module, a rotation module, and an inbound path planning module.

[0024] Preferably, the command receiving module is the core hub and data transmission bridge for the AGV to interact with the upper-level scheduling center. It has high real-time and high-reliability command capture and parsing capabilities, and can capture command information including picking coordinates, cargo number, operation priority and related safety specifications in real time.

[0025] Preferably, the point cloud map information retrieval module is a key supporting unit for AGVs to achieve accurate positioning and operation navigation. It provides accurate data support for subsequent path planning and operation execution, adapts to dynamic operation requirements in complex scenarios, and improves the reliability and efficiency of AGV automated operation.

[0026] Preferably, the global path planning module and the local path planning module plan the path through the global planner and the local planner, respectively. The two form a cooperative mode of global direction setting and local safety assurance, which not only ensures the overall efficiency of the operation, but also improves the path adaptability and driving safety in complex environments, thereby jointly ensuring the efficient and accurate driving of the AGV.

[0027] Preferably, the intelligent edge-keeping calculation module calculates the horizontal edge distance as follows: (channel width - AGV vehicle width) / 2 - preset safety margin. This calculation ensures that when the AGV vehicle is driving in the center of the channel or performing subsequent operations, there is sufficient safety margin on both sides to avoid scratching the shelf.

[0028] Preferably, the AGV vehicle's position and posture standard at the second reference point is as follows:

[0029] With the direction parallel, the center line of the AGV vehicle body is completely parallel to the direction of the target channel or storage location, ensuring that the vehicle will not deviate diagonally when traveling in the channel, avoiding collisions with obstacles. This parallel posture allows the AGV vehicle to strictly follow the planned path, reducing the time loss caused by directional correction and improving the continuity of operations.

[0030] With intelligent distance control, the AGV maintains a fixed distance between itself and one side of the passage. This fixed distance is dynamically calculated based on the passage width, rather than being a fixed value. This allows it to respond in real time to changes in the passage width. Combined with the AGV's own safety envelope and preset safety margin, it accurately outputs a fixed distance that is suitable for the current passage, so as to adapt to passages of different widths.

[0031] Facing the parking space, the AGV vehicle is positioned directly in front of the entrance, along the extension line of the parking space entrance. The specific distance can be optimized based on the depth of the parking space and the vehicle's minimum turning radius to ensure optimal operating space for subsequent rotation and reversing. This avoids collisions with the edge of the parking space during turns due to parking too close, or increased invalid driving distance due to parking too far, reducing the frequency of path corrections and improving the efficiency and smoothness of reversing into the parking space.

[0032] Preferably, the confirmation module includes position confirmation, attitude confirmation, and vehicle stationary confirmation. These three together constitute the core prerequisites for the AGV to enter the fixed-point rotation and attitude calibration logic, and none of them can be omitted.

[0033] Position confirmation is used to verify whether the AGV trolley has accurately stopped at the second reference point;

[0034] The posture confirmation is used to verify whether the parallelism between the centerline of the AGV vehicle body and the target storage location and passage meets the operational requirements;

[0035] Vehicle stationary confirmation ensures the vehicle body is completely stationary, without any displacement or shaking.

[0036] Preferably, the rotation module is executed only after the position, posture, and vehicle stationary are confirmed. It is used for the AGV to perform a fixed-angle rotation in place according to the absolute orientation of the target storage location, so that the AGV body is adjusted from a driving posture parallel to the channel to an operating posture perpendicular to the channel and facing the storage door.

[0037] Preferably, after the rotation module completes the rotation of the AGV vehicle body, it confirms through the sensor module including the encoder that the yaw angle has accurately reached the target value, and completes the final attitude self-calibration. The sensor module will simultaneously fuse the attitude data of the gyroscope and the displacement feedback of the encoder. Through multi-source data cross-verification, a triple guarantee mechanism of data consistency verification, outlier elimination and error compensation is constructed to further improve the accuracy and reliability of yaw angle detection and avoid calibration errors caused by single sensor failure or environmental interference.

[0038] Preferably, the warehousing path planning module is used to plan the AGV cart to reverse into the target storage location. It uses a local planner to generate a smooth reversing path with curvature, starting from the precise pose of the AGV cart after rotation and ending at the picking point inside the storage location, so as to avoid the impact of sharp turns and jerking movements on the balance of the vehicle and the stability of the goods.

[0039] Compared with the prior art, the beneficial effects of the present invention are: the AGV picking method is not based on navigation of fixed beacons, thus avoiding the problem of lack of flexibility and ensuring terminal accuracy. In addition, it can also ensure the adaptability of AGV in structured narrow channels, so that AGV can automatically adapt to the narrow channel conditions to ensure accurate picking of goods in narrow channels.

[0040] After receiving the picking command from the host system, the AGV retrieves the target information and the coordinates of the first reference point from the cloud map. Then, it plans a global path from the current position to the first reference point through the global planner. The AGV then travels along the global path and stops at the first reference point. This process enables safe navigation from any starting position to the first reference point. Because it automatically plans the path after receiving the command from the host system, it does not rely on fixed beacons for navigation. This allows it to flexibly adjust the navigation path according to the position of the AGV. In addition, since it does not need to deploy fixed beacons, it can save deployment and maintenance costs. This avoids the problems of the previous AGV picking process, which relied heavily on pre-laid physical beacons, making changes to the path, workstation, or layout extremely inconvenient and requiring reconstruction. The system was inflexible and the deployment and maintenance costs were high.

[0041] After the AGV (Automated Guided Vehicle) navigates to the first reference point from any location through global path planning, it not only retrieves the coordinates and orientation of the target storage location from the storage location attribute database of the digital map, but also retrieves the precise width of the passageway it belongs to. Then, through the local planner, it plans a movement path with the first reference point as the starting point and the second reference point as the ending point. When the AGV moves on the movement path with the first reference point as the starting point and the second reference point as the ending point, the system calculates the optimal lateral distance to the edge based on the width of the passageway and the width of the AGV, so that the AGV automatically adapts to the width of the passageway. This helps the AGV to drive in the center of the passageway, so as to ensure that there is sufficient safety margin on both sides of the AGV during subsequent operations and avoid scratching the shelves.

[0042] After the AGV reaches the second reference point, it rotates a certain angle in place according to the absolute orientation of the storage location. Then, the local planner plans a smooth reversing path with curvature. This reversing path starts from the second reference point and ends at the target storage location. The second reference point in this path is an absolute starting point that has cleared the accumulated error. In addition, on the path from the first reference point to the second reference point, the first reference point is also an absolute starting point. Thus, the accumulated error can be cleared by using the first and second reference points. This avoids the problems caused by the accumulated error generated after the AGV runs in long-distance complex environments, which prevents the AGV from being able to face the storage location in narrow channels, leading to failure in picking up goods, scratches, or even collisions. As a result, the success rate and reliability of the operation cannot be guaranteed. Attached Figure Description

[0043] Figure 1 This is a diagram showing the system module composition of the present invention.

[0044] Figure 2 This is a flowchart of the workflow of the present invention. Detailed Implementation

[0045] 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.

[0046] Example: Please refer to Figures 1-2 The present invention provides the following technical solution: an AGV narrow channel precise picking method based on multi-level reference point serial control, including an open space navigation module, a narrow channel navigation module and an inbound navigation module;

[0047] The open space navigation module includes:

[0048] A command receiving module for AGVs to receive picking instructions from the host system;

[0049] The command receiving module is the core hub and data transmission bridge for the AGV to interact with the upper-level scheduling center. It has high real-time and high reliability command capture and parsing capabilities, and can capture command information including picking coordinates, cargo number, operation priority and related safety specifications in real time.

[0050] The received instructions originate from the upper-level system, which consists of a scheduling management system, a warehouse execution system, a production execution system, and a human-machine interaction terminal. The scheduling management system is responsible for AGV task allocation, path planning, and conflict coordination. The warehouse execution system provides basic data support such as the storage location of goods and inventory status. The production execution system issues material transfer requirements according to the production plan. The human-machine interaction terminal supports manual emergency instruction input and task status monitoring.

[0051] The command receiving module achieves reliable command reception through a collaborative design of the hardware and software layers:

[0052] At the hardware level, industrial Ethernet interfaces, wireless communication modules, or serial ports are used as data transmission carriers to adapt to the communication distance and real-time requirements of different industrial scenarios.

[0053] At the software level, a communication protocol parsing unit, a data verification module, and an instruction cache unit are integrated. During the receiving process, the instruction data packet issued by the upper system is first obtained through the hardware interface. After being parsed and verified by the software layer, key information such as the picking location, the item number, and the priority is extracted, and a receiving confirmation signal is sent back to the upper system. Finally, the valid instruction is transmitted to the AGV's control unit to drive the subsequent picking action.

[0054] A point cloud map information retrieval module for retrieving target information and the location of the first reference point from a point cloud map;

[0055] Based on the target information retrieved by the point cloud map information retrieval module and the location of the first reference point, a global path planning module plans a global edge-keeping path from the current position of the AGV to the first reference point. The global edge-keeping path enables the AGV to travel along open areas first and avoid all narrow spaces.

[0056] The point cloud map information retrieval module is a key supporting unit for AGVs to achieve accurate positioning and operation navigation. It provides accurate data support for subsequent path planning and operation execution, adapts to dynamic operation requirements in complex scenarios, and improves the reliability and efficiency of AGV automated operation.

[0057] The global path planning module and the local path planning module plan paths through the global planner and the local planner, respectively. The two form a cooperative mode of global direction setting and local safety assurance, which not only ensures the overall efficiency of the operation, but also improves the path adaptability and driving safety in complex environments, thus jointly ensuring the efficient and accurate driving of the AGV.

[0058] The point cloud map information retrieval module consists of four core components: map indexing unit, target retrieval unit, reference point positioning unit, and data caching and interface unit.

[0059] The map index unit uses the Octree or KD-Tree algorithm to perform hierarchical and block indexing of massive point cloud data, establishes a fast mapping relationship between spatial location and point cloud data, and greatly improves retrieval efficiency.

[0060] The target retrieval unit has a built-in target feature library, which contains three-dimensional feature parameters such as cargo dimensions, surface texture, and geometric contour. The target retrieval unit supports accurate retrieval based on feature matching.

[0061] The reference point positioning unit pre-stores the three-dimensional coordinate features of the first reference point that has been calibrated in the scene, and is specifically responsible for the rapid identification and location extraction of the reference point;

[0062] The data cache and interface unit is used to temporarily store retrieved point cloud fragments and target or reference point data, and provides a standardized data interface for data interaction with the AGV control unit and navigation module;

[0063] The point cloud map information retrieval module follows a standardized process of "index positioning - feature matching - data extraction - verification feedback" to retrieve information.

[0064] First, the module receives a retrieval command from the AGV control unit. Then, the map indexing unit quickly locates the corresponding area's point cloud sub-map using a pre-established hierarchical index based on the spatial range parameters in the command, avoiding redundant calculations of the full point cloud data. Next, the target retrieval unit extracts 3D features from the point cloud sub-map, compares and matches them with the built-in feature library, and filters out target information that meets the requirements. Simultaneously, the reference point positioning unit identifies and accurately calculates the 3D spatial position of the first reference point in the point cloud sub-map based on pre-stored calibration features. Afterward, the data verification module verifies the consistency between the retrieved target information and the reference point position to ensure data accuracy. Finally, the verified data is temporarily stored in the data cache unit and synchronized to the AGV's navigation module and execution control unit through a standardized interface, providing spatial positioning basis for subsequent path planning and accurate retrieval. At the same time, it sends a "retrieval successful" confirmation signal to the control unit. If no target or reference point is found, an abnormal alarm is triggered and a retrieval failure message is returned.

[0065] The narrow channel navigation module includes:

[0066] This module is used to retrieve digital map information from the location attribute database of a digital map, including the target storage location, its orientation, the precise width of the passageway it belongs to, and the storage location depth.

[0067] The core of the digital map information retrieval module consists of a storage location index parsing unit, an attribute association retrieval unit, a data verification and formatting unit, a local interface adaptation unit, and a cache management unit. The storage location index parsing unit adopts a dual indexing mechanism of spatial index and attribute index to quickly associate the spatial location and attribute data of the storage location.

[0068] The attribute association retrieval unit has a built-in storage location-channel mapping table, which can reversely associate the geometric parameters of its corresponding channel through the target storage location information, and supports multi-condition retrieval; the data verification and formatting unit is responsible for verifying the logical consistency of the retrieved data and converting the raw database data into a standardized format that can be recognized by the AGV control unit;

[0069] The local interface adapter unit adapts to the local database interface and local file interface to ensure efficient connection with the digital map location attribute database.

[0070] The cache management unit is dedicated to storing frequently accessed database location and channel attribute data, reducing the time spent repeatedly reading the local database and improving retrieval response speed;

[0071] The retrieval process of the digital map information retrieval module follows a closed-loop flow of "command parsing - dual-index positioning - attribute association - verification and formatting - data output":

[0072] The first step is for the module to receive the retrieval command issued by the AGV control unit. The command parsing unit quickly extracts the core search conditions and excludes invalid parameters.

[0073] The second step involves the storage location index parsing unit first matching the database primary key of the target storage location using the attribute index, then locating the spatial coordinate range of the storage location in the digital map using the spatial index, and simultaneously locking the channel ID using the channel index, thus achieving rapid positioning in both attribute and spatial dimensions and avoiding full database traversal.

[0074] The third step involves the attribute association retrieval unit retrieving the basic attributes of the target storage location from the storage location attribute database based on the storage location ID. Then, through the storage location-channel mapping table, it retrieves the precise width of the corresponding channel, the start and end coordinates of the channel, and other related parameters based on the channel ID, thus completing the linkage extraction of target information and channel attributes.

[0075] The fourth step is for the data verification and formatting unit to perform dual verification on the retrieved parameters: first, numerical validity verification, and second, logical consistency verification. After the verification is passed, the scattered attribute data is integrated into a standardized data structure.

[0076] Fifth, if the target database location is a frequently accessed object, the cache management unit will store the formatted data in the local cache, and subsequent identical requests can directly read from the cache;

[0077] The sixth step involves the module synchronizing standardized data to the AGV navigation module and execution control unit via the local interface adapter unit. This provides dimensional and orientational information for the AGV to accurately dock and avoid obstacles along its path. Simultaneously, it sends a "data retrieval successful" message and a verification code to the control unit. If the target storage location cannot be found or the data verification fails, a local alarm is triggered and an error code is returned.

[0078] This intelligent edge-keeping calculation module automatically calculates the optimal lateral distance based on the width of the passage and the vehicle's own safe operating envelope, ensuring that the AGV maintains a safe distance from obstacles on both sides within the operating passage and avoiding the risk of collision.

[0079] This intelligent edge-keeping calculation module automatically calculates the optimal lateral distance based on the width of the passage and the vehicle's own safe operating envelope, ensuring that the AGV maintains a safe distance from obstacles on both sides within the operating passage and avoiding the risk of collision.

[0080] The intelligent edge-keeping calculation module calculates the horizontal edge distance as follows: (channel width - AGV vehicle width) / 2 - preset safety margin. This calculation ensures that when the AGV vehicle is driving in the center of the channel or performing subsequent operations, there is sufficient safety margin on both sides to avoid scratching the shelf.

[0081] The intelligent edge-keeping calculation module also includes a level self-adjustment module to ensure that the AGV's cargo platform is level. The level self-adjustment module includes an attitude sensing unit, a signal processing and control unit, and a drive execution unit. The attitude sensing unit senses whether the AGV's cargo platform is level. If it is not level, the signal processing and control unit causes the drive execution unit to perform relevant operations to level the AGV's cargo platform. In addition, the width of the AGV itself is the width of its cargo platform. The level self-adjustment module can avoid errors in the lateral edge-keeping distance calculated by the intelligent edge-keeping calculation module due to uneven road surfaces in narrow passages, which would affect the AGV's driving.

[0082] The core of the intelligent edge calculation module consists of a parameter input and preprocessing unit, a safety boundary modeling unit, an optimal distance solution unit, a dynamic adjustment unit, and a result output and feedback unit. The parameter input and preprocessing unit is responsible for receiving multi-source data, including the precise channel width transmitted by the digital map information retrieval module and the AGV's own preset safety operation envelope parameters. At the same time, it performs noise reduction, unit unification, and validity verification on the input data and eliminates abnormal values.

[0083] Based on the preprocessed parameters, the safety boundary modeling unit constructs a two-dimensional geometric model of the channel and the vehicle, clarifies the boundaries on both sides of the channel, the outline boundary of the AGV body, and the safety redundancy boundary, forming a three-layer constraint model of "channel boundary - safety boundary - body boundary".

[0084] The optimal distance calculation unit has a built-in multi-objective optimization algorithm, which establishes constraint equations with the optimization objectives of "balancing the safety distance on both sides, prioritizing the side closer to the target storage location, and maximizing the working space".

[0085] The dynamic adjustment unit receives dynamic obstacle information in the channel from the AGV sensors in real time and pre-stores adjustment strategies for different operating scenarios.

[0086] The result output and feedback unit is responsible for converting the calculation results into an instruction format that the AGV control unit can recognize, while providing real-time feedback on the calculation status.

[0087] The intelligent edge computing module follows a closed-loop logic of "parameter input - model construction - algorithm solution - dynamic correction - result output":

[0088] The first step involves the parameter input and preprocessing unit simultaneously collecting two types of core data: the precise width of the channel on the digital map and the safe operation envelope parameters of the AGV itself. Data exceeding the reasonable range is removed through a verification mechanism, and unit unification is completed.

[0089] The second step is to establish a two-dimensional coordinate system with the center line of the channel as the origin in the safety boundary modeling unit, and then determine the left and right limits of the vehicle body contour based on the AGV safety operation envelope. Combined with the basic safety redundancy, the safety boundaries on the left and right sides of the AGV are delineated.

[0090] The third step is to substitute the constraint conditions into the optimal distance solution unit. If the target storage location is located on the right side of the channel, the optimization target will first move to the right side and satisfy the right safety boundary and the left safety boundary. The optimal lateral offset of the AGV relative to the center line of the channel is calculated by the algorithm.

[0091] The fourth step is to dynamically adjust the unit to monitor sensor data in real time. If a temporary obstacle is found on the right side of the channel that reduces the actual usable width, the correction mechanism is immediately triggered to recalculate the optimal offset.

[0092] In the fifth step, the result output and feedback unit converts the final lateral offset into a control command and sends it to the AGV navigation unit to drive the AGV to the side along the optimal path. At the same time, it sends a "calculation completed" signal back to the control unit. If there is an abnormal parameter or the safety constraint cannot be met, an alarm is triggered and an error message "cannot calculate optimal distance" is returned.

[0093] Starting from the position where the AGV has traveled to the first reference point, the optimal lateral distance calculated by the intelligent edge-keeping calculation module is used as a constraint to generate a lateral movement path, which guides the AGV to the second reference point.

[0094] The core of the local path planning module consists of a path constraint parsing unit, a path generation unit, a path smoothing optimization unit, a collision detection unit, and a path instruction conversion unit. The path constraint parsing unit is responsible for extracting the core constraints, including the three-dimensional coordinates of the first and second reference points, the optimal lateral edge distance output by the intelligent edge calculation module, the AGV's own kinematic constraints, and the physical constraints of the channel boundary.

[0095] The path generation unit has a built-in multi-segment path algorithm library, which supports selecting the appropriate path model based on the distance between two points and constraints.

[0096] The path smoothing optimization unit uses gradient descent or moving average algorithm to eliminate inflection points and abrupt changes in the path, ensuring smooth and shock-free AGV movement.

[0097] The collision detection unit integrates a real-time collision prediction mechanism based on a grid map, and performs point-by-point collision verification on the planned path by combining the AGV body contour and the work envelope; the path command conversion unit is responsible for converting the planned path data into control commands that the AGV drive system can recognize, while adapting to different motion control protocols.

[0098] The local path planning module follows a closed-loop process of "constraint modeling - path generation - smooth optimization - collision verification - instruction output":

[0099] The first step is to synchronously collect multi-source constraint data by the path constraint parsing unit, obtain the real-time coordinates of the first reference point and the calibration coordinates of the second reference point through the positioning module, extract the lateral constraints of the intelligent edge calculation module, and load the AGV kinematic parameters.

[0100] The second step is for the path generation unit to select a path model based on the constraint characteristics between two points. Since the core requirement is "lateral movement" and the attitude angle must remain unchanged, a straight path model is preferred. A basic lateral movement path is generated by fitting the coordinates of two points, and the starting point, ending point, length and key node coordinates of the path are defined.

[0101] The third step is to refine the basic straight path by using the path smoothing optimization unit. If there is a slight attitude deviation between the starting point of the path and the first reference point, it is corrected by the arc transition section to eliminate sudden steering changes. At the same time, the path speed curve is planned according to the AGV acceleration constraint to ensure smooth movement.

[0102] The fourth step is that the collision detection unit performs a full-path collision scan on the planned path based on the channel boundary data of the digital map and the AGV safety operation envelope. The path is divided into several small segments in a gridded manner, and the AGV body outline is checked segment by segment to see if it exceeds the channel safety range or overlaps with obstacles on both sides. If a potential collision risk is detected, it is immediately fed back to the path generation unit to readjust the path offset until the collision-free constraint is met.

[0103] Fifth, the path command conversion unit converts the optimized path data into pulse commands or bus control commands and sends them to the AGV's drive unit and steering unit to guide the AGV to move laterally from the first reference point to the second reference point along the planned path. At the same time, it receives AGV position feedback data in real time. If a path deviation occurs, it triggers a real-time correction mechanism to adjust the control commands to ensure accurate stopping at the second reference point. After completion, it sends a "path execution completed" signal to the AGV control unit. If the constraints cannot be met during the planning process, an alarm is triggered and a "path planning failed" error message is returned.

[0104] The inbound navigation module is used for the AGV to rotate in place and reverse into the warehouse. The inbound rotation of the AGV includes a status confirmation module, a rotation module, and an inbound path planning module.

[0105] The status confirmation module has a built-in positioning accuracy verification unit, a surrounding environment monitoring unit, and an equipment status self-test unit. The positioning accuracy verification unit confirms the AGV's current position and attitude angle error by comparing the coordinates of the first and second reference points with those of the lidar and vision sensor.

[0106] The surrounding environment monitoring unit scans dynamic obstacles within a 1.5m range of the rotation and entry area in real time and outputs an unobstructed judgment signal;

[0107] The equipment status self-test unit verifies the operating status of the AGV drive motor, steering system, and braking device to ensure there are no fault alarms.

[0108] The specific AGV trolley position and posture standards at the second reference point are as follows:

[0109] With the direction parallel, the center line of the AGV vehicle body is completely parallel to the direction of the target channel or storage location, ensuring that the vehicle will not deviate diagonally when traveling in the channel, avoiding collisions with obstacles. This parallel posture allows the AGV vehicle to strictly follow the planned path, reducing the time loss caused by directional correction and improving the continuity of operations.

[0110] With intelligent distance control, the AGV maintains a fixed distance between itself and one side of the passage. This fixed distance is dynamically calculated based on the passage width, rather than being a fixed value. This allows it to respond in real time to changes in the passage width. Combined with the AGV's own safety envelope and preset safety margin, it accurately outputs a fixed distance that is suitable for the current passage, so as to adapt to passages of different widths.

[0111] Facing the storage space, the AGV vehicle is positioned directly in front of the storage space entrance along the extension line of the storage space entrance. The specific distance can be optimized based on the storage space depth and the vehicle's minimum turning radius to ensure optimal operating space for subsequent rotation and reversing. This avoids collisions with the storage space edge during turns due to parking too close, or increased invalid driving distance due to parking too far, reducing the frequency of path corrections and improving the efficiency and smoothness of reversing into the storage space.

[0112] The confirmation module includes position confirmation, attitude confirmation, and vehicle stationary confirmation. These three components together constitute the core prerequisites for the AGV to enter the fixed-point rotation and attitude calibration logic, and none of them can be omitted.

[0113] Position confirmation is used to verify whether the AGV trolley has accurately stopped at the second reference point;

[0114] The posture confirmation is used to verify whether the parallelism between the centerline of the AGV vehicle body and the target storage location and passage meets the operational requirements;

[0115] Vehicle stationary confirmation ensures the vehicle body is completely stationary, without any displacement or shaking;

[0116] The rotation module consists of an attitude closed-loop control unit, a speed adjustment unit, and a rotation angle calibration unit. The attitude closed-loop control unit is based on a PID algorithm, which compares the deviation between the target rotation angle and the actual angle in real time and dynamically adjusts the steering force.

[0117] The rotation module is executed only after the position, posture, and vehicle stationary are confirmed. It is used for the AGV to perform a fixed-angle rotation in place according to the absolute orientation of the target storage location, so that the AGV body is adjusted from a driving posture parallel to the channel to an operating posture perpendicular to the channel and facing the storage door.

[0118] After the rotation module completes the rotation of the AGV, it confirms that the yaw angle has accurately reached the target value through the sensor module including the encoder, and completes the final attitude self-calibration. The sensor module will simultaneously fuse the attitude data of the gyroscope and the displacement feedback of the encoder. Through multi-source data cross-verification, a triple guarantee mechanism of data consistency verification, outlier elimination and error compensation is constructed to further improve the accuracy and reliability of yaw angle detection and avoid calibration errors caused by single sensor failure or environmental interference.

[0119] The triple guarantee mechanism of data consistency verification, outlier removal and error compensation first filters out invalid data, then locates and handles anomalies, and finally corrects inherent errors. Through standardized preprocessing, precise algorithm modeling and dynamic parameter adjustment, it achieves high-precision detection. In the AGV yaw angle calibration scenario, this mechanism can improve the detection accuracy from ±0.1° of a single sensor to within ±0.05°, and improve the reliability by more than 40%.

[0120] The speed adjustment unit matches the rotation speed according to the size of the rotation angle;

[0121] The rotation angle calibration unit uses both encoder and reference point visual positioning for dual verification to ensure that the rotation angle error is ≤ ±0.3°.

[0122] The inbound path planning module includes a reversing path modeling unit, a motion constraint adaptation unit, and a path correction unit. The reversing path modeling unit adopts a combination model of "circular arc + straight line" to generate a reversing trajectory from the rotation endpoint to the target position of the storage space.

[0123] The inbound path planning module is used to plan the AGV cart to reverse into the target storage location. It uses a local planner to generate a smooth reversing path with curvature, starting from the precise pose of the AGV cart after rotation and ending at the picking point inside the storage location, so as to avoid the impact of sharp turns and jerking movements on the balance of the vehicle and the stability of the goods.

[0124] The motion constraint adaptation unit loads parameters such as the maximum reverse speed and minimum turning radius of the AGV to ensure that the path conforms to the kinematic limits.

[0125] The path correction unit has a reserved real-time adjustment interface, which can correct path deviations based on sensor feedback.

[0126] The complete working process of the inbound navigation module follows an orderly flow of "status confirmation - in-situ rotation - path planning - reversing into the parking space":

[0127] The first step involves the status confirmation module initiating a comprehensive self-test. The positioning accuracy verification unit acquires the real-time coordinates and attitude angles of the AGV at the second reference point and compares them with the calibration values. If the error exceeds the threshold, fine-tuning calibration is triggered. The surrounding environment monitoring unit scans the work area through sensors. If an obstacle is detected, the process is paused and an alarm is issued until the obstacle is cleared. The equipment status self-test unit completes the functional verification of the drive, steering, and braking systems. After all states are "normal," a start signal is sent to the rotation module.

[0128] The second step is the stationary rotation execution stage. After receiving the status confirmation signal, the rotation module determines the target rotation angle according to the storage requirements. The attitude closed-loop control unit initializes the PID parameters, and the speed adjustment unit sets the speed curve of "start acceleration - uniform speed rotation - deceleration and stop". During the rotation, the rotation angle calibration unit collects the steering angle data in real time through the encoder and performs visual positioning calibration every 10° to dynamically correct the rotation deviation and ensure that the AGV accurately turns to the storage orientation.

[0129] The third step is the inbound path planning stage. The path modeling unit takes the center point of the rotated AGV as the starting point and the target stopping point in the storage space as the ending point. Combined with parameters such as storage space depth and channel width, it generates a combined path of "first small radius arc turning - then straight reversing". The motion constraint adaptation unit embeds parameters such as reversing speed and acceleration into the path to ensure that the trajectory does not exceed the safe operation envelope. The path correction unit simultaneously obtains the storage space edge distance data fed back by the LiDAR and presets a dynamic adjustment threshold of ±10mm.

[0130] The fourth step, the reverse parking execution phase, involves converting the planned path into drive commands to control the AGV to reverse along the trajectory. During this process, the path correction unit compares the actual position with the planned path in real time. If deviations occur due to ground friction or load offset, a correction signal is immediately output to adjust the steering angle. When the AGV reaches the target parking point within the parking space, the positioning module confirms the parking accuracy, the braking system triggers parking, and simultaneously sends a "parking complete" signal to the AGV control unit. If obstacles, equipment malfunctions, or insufficient accuracy are detected during the process, parking is immediately stopped and an alarm is issued, awaiting manual intervention or automatic retry.

[0131] The global planner and local planner are the core components of the AGV navigation system. They work together, complement each other, and jointly build a complete navigation loop of "macro decision-making + micro execution" to ensure that the AGV can complete tasks efficiently, safely, and accurately in complex scenarios.

[0132] The global planner plays the role of a "strategic decision-maker," focusing on the macro-level layout of the entire process path. Based on a pre-built global static map, it integrates static environmental information such as warehouse layout, workshop aisles, fixed shelves, and permanent obstacles. At the same time, it receives the starting and ending coordinates of the task from the upper system. Through path search algorithms, it plans a macro-level path framework that meets optimization objectives such as "shortest path, lowest energy consumption, and fewest turns" within the globally feasible area. This path is usually output in the form of a sequence of key nodes or a general path outline. It can avoid static obstacles in advance, provide AGVs with a clear global driving direction, and only performs low-frequency replanning when the global environment changes statically. It does not rely on real-time sensor feedback, has strong stability, and outstanding global optimality.

[0133] The local planner acts as the "tactical executor," focusing on real-time adjustments and refined control during path execution. It uses the macroscopic path output by the global planner as a reference benchmark, combined with real-time local dynamic environmental data collected by multiple sources such as LiDAR, vision sensors, and ultrasonic sensors, as well as the AGV's own real-time position, attitude, speed, and other status information. Through efficient local algorithms such as dynamic windowing, model predictive control, and time-series optimization, it generates continuous, smooth, and collision-free short-distance local trajectories while satisfying the AGV's kinematic constraints. At the same time, it corrects path deviations caused by factors such as ground friction, load offset, and sensor errors during AGV travel in real time, ensuring that the AGV always travels along the core direction of the global path and avoiding the problem of "local optimum, global suboptimal".

[0134] In actual navigation, the collaboration between the two is close and efficient: when the task starts, the global planner first completes the macro path planning from the starting point to the end point, and defines the overall driving route for the AGV.

[0135] After the AGV starts, the local planner continues to work. If no dynamic obstacles are detected and the path deviation is within the allowable range, it executes control commands such as speed and steering according to the global path. If a temporary obstacle is detected or the deviation exceeds the threshold, the local planner will quickly generate obstacle avoidance or correction trajectory without deviating from the core framework of the global path. After the obstacle is cleared or the deviation is corrected, it will return to the global planned path. When the AGV approaches the target area, the local planner further improves the planning frequency and control accuracy, and completes fine-tuning operations such as lateral fine-tuning, in-situ rotation, and reversing into the parking space. Ultimately, it achieves the complete goal of "global optimal navigation + local safe execution", neither of which can be dispensed with.

[0136] Without a global planner, AGVs may get stuck in local obstacle avoidance or path redundancy due to a lack of macro guidance;

[0137] Without a local planner, AGVs cannot cope with dynamic environmental changes and are prone to collision risks. It is this division of labor and cooperation of "macroscopic direction setting and microscopic safety assurance" that allows AGVs to flexibly adapt to the navigation needs of diverse scenarios such as warehousing and factories.

[0138] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0139] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AGV narrow passage precise picking method based on multi-level reference point series control, characterized in that, The method comprises open space navigation, narrow channel navigation and warehouse entry navigation: The open space navigation comprises: Receiving a picking instruction for the AGV from the upper system through a command receiving module; The target information and the first reference point position are called from the point cloud map through a point cloud map information calling module; Based on the global path planning module, a global edge path is planned from the current position of the AGV to the first reference point, and the AGV is preferentially driven along the open area to avoid all narrow spaces, according to the target information and the first reference point position called by the point cloud map information calling module. The narrow channel navigation comprises: The target storage location and the orientation, the accurate width of the channel and the storage location depth are called from the storage location attribute database of the digital map through a digital map information calling module; An optimal lateral edge distance is automatically calculated by the intelligent edge calculation module according to the channel width and the safety operation envelope of the vehicle itself, to ensure that the AGV maintains a safe distance from the obstacles on both sides in the operation channel; A lateral movement path leading the AGV to the second reference point is generated by the local path planning module, taking the position of the AGV driving to the first reference point as the starting point and the optimal lateral edge distance calculated by the intelligent edge calculation module as the constraint. The warehouse entry navigation comprises in-place rotation and reverse entry of the AGV; The in-place rotation is executed based on a state confirmation module, a rotation module and a warehouse entry path planning module; The state confirmation module performs position alignment confirmation, attitude alignment confirmation and vehicle static confirmation, which together constitute the core prerequisite condition for the AGV to enter the fixed-point self-rotation and attitude calibration logic; The position alignment confirmation is used to verify whether the AGV is accurately parked at the second reference point; The attitude alignment confirmation is used to verify whether the parallelism of the AGV body center line and the target storage location and the channel meets the operation requirements; The vehicle static confirmation ensures that the vehicle body is completely stable without displacement or shaking; The rotation module performs a fixed-angle in-place rotation to adjust the AGV body from a driving attitude parallel to the channel to an operation attitude perpendicular to the channel and facing the warehouse door, according to the absolute orientation of the target storage location, only after the position alignment, attitude alignment and vehicle static confirmation are all confirmed. The warehouse entry path planning module is used to plan the reverse entry of the AGV into the target storage location, and a smooth reverse path with curvature is generated by the local planner, taking the accurate pose of the AGV after rotation as the starting point and the picking point inside the storage location as the end point.

2. The AGV narrow passage precise picking method based on multi-level reference point series control according to claim 1, characterized in that: The command receiving module is the core hub and data transmission bridge for the AGV to interact with the upper dispatch center, has high real-time and high-reliability instruction capture and analysis capabilities, and can capture and analyze instruction information including picking coordinates, cargo number, operation priority and related safety specifications in real time.

3. The AGV narrow passage precise picking method based on multi-level reference point series control according to claim 1, characterized in that: The point cloud map information calling module is a key support unit for the AGV to realize accurate positioning and operation navigation, provides accurate data support for subsequent path planning and operation execution, adapts to dynamic operation requirements in complex scenarios, and improves the reliability and efficiency of AGV automatic operation.

4. The AGV narrow passage precise picking method based on multi-level reference point series control according to claim 3, characterized in that: The global path planning module and the local path planning module plan paths through global planner and local planner respectively, and the two form a cooperation mode of global direction and local safety, which not only ensures the overall efficiency of the operation, but also improves the path adaptability and driving safety in complex environment, thereby guaranteeing the efficient and accurate driving of the AGV.

5. The AGV narrow passage precise picking method based on multi-level reference point series control according to claim 1, characterized in that: The calculation formula logic of the intelligent side calculation module is that the lateral side distance = (passage width-AGV car width) / 2-preset safety margin, and this calculation ensures that the AGV car has sufficient safety margin on both sides when driving in the passage center or performing subsequent operations, avoiding scratching with the shelf.

6. The AGV narrow aisle precise picking method based on multi-level reference point series control according to claim 5, characterized in that: The AGV car pose standard on the second reference point is as follows: Parallel direction, the AGV car body center line is completely parallel to the direction of the target passage or the location, ensuring that the car does not deviate obliquely when driving in the passage, avoiding scratching with obstacles, and this parallel posture can make the AGV car strictly adhere to the planned path, reduce the time loss caused by direction correction, and improve the operation continuity; Distance intelligence, the AGV car body maintains a fixed distance from one side of the passage, and the fixed distance is dynamically calculated from the passage width, rather than a fixed value, so that it can respond to the changes in the passage width in real time, and combine the AGV car's own safety operation envelope and the preset safety margin to accurately output the fixed distance that adapts to the current passage, so as to adapt to passages of different widths; Directly facing the location, the AGV car body is parked in front of the location entrance in the direction of the extension line of the location entrance, and the specific distance is optimized according to the location depth and the minimum turning radius of the vehicle, to ensure that the subsequent self-rotation and reverse into the location have the best operation space, avoid collision with the edge of the location when turning due to parking too close, or increase the invalid driving distance due to parking too far, reduce the path correction frequency, and improve the efficiency and smoothness of the reverse into the location.

7. The AGV narrow aisle precise picking method based on multi-level reference point series control according to claim 6, characterized in that: After the rotation module completes the rotation of the AGV car body, the sensor module including the encoder confirms that the yaw angle has accurately reached the target value, and completes the final pose self-calibration. The sensor module synchronously fuses the attitude data of the gyroscope and the displacement feedback of the encoder, and through multi-source data cross verification, constructs a three-fold guarantee mechanism of data consistency verification, outlier rejection and error compensation, further improves the accuracy and reliability of the yaw angle detection, and avoids calibration errors caused by single sensor failure or environmental interference.

Images