AGV safety navigation and obstacle avoidance system based on intelligent storage environment and control method

Abstract

The invention provides an AGV safety navigation and obstacle avoidance system based on an intelligent storage environment and a control method. The method comprises the following steps: acquiring three-dimensional point data of an intelligent storage environment through a laser radar, and generating a three-dimensional grid chart after processing and removing interference points; and acquiring the distance between the obstacle and the AGV by using a millimeter wave radar, integrating the distance with the three-dimensional grid diagram, updating the space relationship between the obstacle and the AGV by combining the position of the AGV, and outputting a dynamic environment state quantity with a timestamp. The dynamic environment state quantity is input into a reinforcement learning path planner, a goods shelf dense environment is recognized, and a collision-free steering angle and speed instruction is generated; and the steering flexibility and the braking force of the AGV are dynamically adjusted according to an instruction through a fuzzy PID controller, so that safe navigation and obstacle avoidance in a goods shelf dense environment are realized. According to the invention, the safe navigation and obstacle avoidance capabilities of the AGV in an environment with dense intelligent storage shelves are improved.

Description

Technical Field

[0001] This application relates to the field of AGV safety navigation and obstacle avoidance system technology, and in particular to an AGV safety navigation and obstacle avoidance system and control method based on an intelligent warehousing environment. Background Technology

[0002] In intelligent warehousing environments, Automated Guided Vehicles (AGVs), as core equipment for cargo transfer, need to achieve autonomous navigation in scenarios with densely arranged shelves, narrow aisle spaces, multiple AGVs operating simultaneously, and dynamic obstacles such as temporarily stacked goods. Their core technological requirements focus on three aspects: first, accurately identifying the real-time positions of fixed shelves and moving obstacles to ensure centimeter-level spatial perception accuracy; second, when facing sudden obstacles, coordinating steering and braking adjustments within 0.5 seconds to avoid collisions; and third, balancing steering flexibility and driving stability in complex sections such as shelf corners and aisle intersections to prevent goods from swaying or tipping over.

[0003] Currently, the mainstream solution for addressing the aforementioned needs is a technology system based on LiDAR SLAM navigation combined with proportional-integral-derivative (PI-DE) control. This solution uses LiDAR to collect 3D point cloud data of the environment, constructing a real-time map of the warehouse scene and locating the relative position of the AGV to obstacles in real time. Simultaneously, the PI-DE controller outputs steering angle and speed commands based on the deviation between the preset path and the actual position, driving the AGV to complete navigation and obstacle avoidance actions, achieving basic autonomous operation in most standardized warehouse scenarios.

[0004] However, this solution has significant drawbacks in dense, dynamic environments: First, the lidar is susceptible to glare from metal shelves and obstructions from goods in the warehouse environment, resulting in a 10-30 cm deviation in obstacle location identification, which can easily lead to collision risks in narrow aisles; Second, the proportional-integral-derivative controller relies on fixed parameters, and when the speed or direction of a dynamic obstacle changes abruptly, the parameters cannot be adapted in real time, resulting in a lag in steering response; Third, the control logic is not optimized for the aisle characteristics of densely shelf-saturated scenarios, and when approaching the edge of the shelf, it is prone to scraping the shelf due to oversteering or insufficient braking force. Summary of the Invention

[0005] This application provides an AGV safety navigation and obstacle avoidance system and control method based on an intelligent warehousing environment, in order to solve the problems of poor safety navigation and obstacle avoidance capabilities of AGVs in dense intelligent warehouse racking environments caused by identification deviation, poor parameter adaptation, and unoptimized control logic in the prior art.

[0006] Firstly, this application provides a method for safe navigation and obstacle avoidance control of AGVs in an intelligent warehousing environment, including:

[0007] The 3D point data of the intelligent warehousing environment where the AGV is located is collected by LiDAR, and the 3D point data is processed by a 3D filtering algorithm to remove noise and impurities and generate a 3D mesh map.

[0008] Obstacle distance data between obstacles and AGVs is acquired by millimeter-wave radar. The three-dimensional mesh map and the obstacle distance data are integrated. Based on the integrated result and the position of the AGV, the spatial distribution relationship between the AGV and obstacles is updated, and a dynamic environmental state quantity with a timestamp is output.

[0009] The dynamic environmental state variables are input into the path planner built based on the reinforcement learning network to determine that the intelligent warehouse environment is a dense shelving environment, and to generate the turning angle and speed commands for the AGV to pass through the dense shelving environment without collision.

[0010] Based on a fuzzy PID controller, the steering stiffness and braking force of the AGV are dynamically adjusted according to the steering angle and the speed command to achieve safe navigation and obstacle avoidance control of the AGV in dense shelving environments.

[0011] Optionally, the step of inputting the dynamic environment state variables into a path planner constructed based on a reinforcement learning network, determining that the intelligent warehousing environment is a densely racked environment, and generating steering angle and speed commands for the AGV to pass through the densely racked environment without collision includes:

[0012] The dynamic environment state is input into the path planner built based on the reinforcement learning network. The decomposition module of the path planner decomposes the dynamic environment state into the AGV's current coordinates, obstacle distribution data and timestamp information.

[0013] The obstacle distribution density and channel space parameters are extracted from the obstacle distribution data using the extraction module of the path planner.

[0014] The path planner's determination module identifies smart warehouse environments where the obstacle distribution density is greater than a preset density value and the channel space parameter is less than a preset width value as dense shelving environments.

[0015] In the dense shelving environment, the mapping module of the path planner maps the current coordinates of the AGV and the timestamp information into multiple candidate action combinations containing turning angle and speed parameters.

[0016] The path planner's selection module selects a candidate action combination that matches the obstacle distribution data from the multiple candidate action combinations, which serves as the steering angle and speed command for the AGV to pass through densely packed shelving environments without collisions.

[0017] Optionally, selecting the candidate action combination that matches the obstacle distribution data from the plurality of candidate action combinations as the steering angle and speed command for the AGV to pass through densely packed shelving environments without collisions includes:

[0018] Based on the obstacle distribution data, a predicted travel trajectory is determined for each candidate action combination. The predicted travel trajectory includes a series of future position points of the AGV when it executes the candidate action combination.

[0019] For each predicted passage trajectory, the distance between each future position point and the nearest fixed obstacle is calculated as a fixed obstacle safety parameter; the distance between each future position point and the nearest moving obstacle is calculated as a moving obstacle safety parameter; the degree to which each future position point deviates from the center of the passage is calculated as a passage centering parameter; and the change in the turning angle of adjacent future position points is calculated as a turning smoothness parameter.

[0020] The fixed obstacle safety parameters, the moving obstacle safety parameters, the passage centering parameters, and the steering smoothness parameters are integrated to generate a comprehensive evaluation value for each candidate action combination;

[0021] The candidate action combination with the best comprehensive evaluation value is selected as the steering angle and speed command for the AGV to pass through densely packed shelving environments without collisions.

[0022] Optionally, the step of collecting three-dimensional point data of the intelligent warehousing environment where the AGV is located using LiDAR, and processing the three-dimensional point data using a three-dimensional filtering algorithm to remove noise and artifacts and generate a three-dimensional mesh map includes:

[0023] Remove data that exceeds a preset distance range from the three-dimensional point data, and retain the initial three-dimensional point data within the effective sensing area around the AGV;

[0024] The initial three-dimensional point data is filtered to obtain target three-dimensional point data after removing noise and speculative points;

[0025] The target three-dimensional point data is divided into multiple spatial grid units according to a preset size, and the spatial interval between adjacent three-dimensional points in the target three-dimensional point data is calculated to obtain three-dimensional points whose spatial interval conforms to the preset spatial interval.

[0026] Each grid cell records the number of the three-dimensional points distributed within the intelligent warehousing environment;

[0027] The grid state is determined based on the distribution quantity, and a three-dimensional grid map containing obstacle areas and passage areas is generated based on the grid state.

[0028] Optionally, the step of acquiring obstacle distance data between obstacles and AGV via millimeter-wave radar, integrating the 3D mesh map and the obstacle distance data, updating the spatial distribution relationship between the AGV and obstacles based on the integrated result and the AGV's position, and outputting a timestamped dynamic environmental state quantity includes:

[0029] Millimeter-wave radar is used to detect fixed and moving obstacles around the AGV, and the straight-line distance between each obstacle and the AGV is recorded as obstacle distance data;

[0030] Based on the obstacle distance data, the positions of each obstacle are marked in the corresponding grid cells of the three-dimensional grid map, so that the corresponding grid cells simultaneously contain three-dimensional point distribution information and obstacle distance information;

[0031] Based on the position of the AGV, calculate the orientation and spacing of obstacles in the corresponding grid cell relative to the AGV;

[0032] Based on the orientation and the spacing, adjust the three-dimensional point distribution information and obstacle distance information of each obstacle to generate an updated spatial distribution relationship between the AGV and the obstacles;

[0033] Add corresponding time information to the spatial distribution relationship and output dynamic environmental state quantities with timestamps.

[0034] Optionally, the step of adjusting the three-dimensional point distribution information and obstacle distance information of each obstacle according to the orientation and the spacing to generate an updated spatial distribution relationship between the AGV and the obstacles includes:

[0035] The category to which the obstacle belongs is determined to distinguish between fixed obstacles and moving obstacles;

[0036] For each of the fixed obstacles, when the position corresponding to the three-dimensional point distribution information of the obstacle is inconsistent with the position derived based on the orientation and the spacing, the coordinate data in the three-dimensional point distribution information of the obstacle is corrected to obtain the corrected three-dimensional point distribution information of the obstacle.

[0037] For each of the moving obstacles, when there is a difference between the obstacle distance information and the spacing, the obstacle distance information is updated according to the spacing to obtain the updated obstacle distance information;

[0038] Based on the corrected 3D point distribution information of the obstacles and the updated obstacle distance information, combined with the orientation of the fixed obstacles and the moving obstacles, the position parameters of each obstacle in 3D space are determined.

[0039] Based on the position parameters, the relative positional relationships between the obstacles are determined;

[0040] By integrating the relative positional relationship with the current coordinates of the AGV, an updated spatial distribution relationship between the AGV and obstacles is generated.

[0041] Optionally, the step of dynamically adjusting the steering stiffness and braking force of the AGV based on the fuzzy PID controller according to the steering angle and the speed command to achieve safe navigation and obstacle avoidance control of the AGV in densely packed shelving environments includes:

[0042] Extract the actual steering deviation corresponding to the steering angle and the actual speed deviation corresponding to the speed command;

[0043] Based on the actual steering deviation and the actual speed deviation, and combined with the aisle width characteristics in a densely packed shelving environment, the adjustment ranges corresponding to the steering stiffness and braking force of the AGV are determined.

[0044] Based on the adjustment range, calculate the first change in steering stiffness and the second change in braking force;

[0045] The first change is input to the steering component of the AGV to obtain steering operation parameters, and the second change is input to the braking component of the AGV to obtain braking operation parameters.

[0046] The real-time distance between obstacles and the AGV in the dynamic environment state is obtained. Based on the steering operation parameters and the braking operation parameters, and combined with the real-time distance, the adjustment range is corrected to achieve safe navigation and obstacle avoidance control of the AGV in densely packed shelving environments.

[0047] Secondly, this application provides an AGV safety navigation and obstacle avoidance system based on an intelligent warehousing environment, including:

[0048] The first generation module is used to collect three-dimensional point data of the intelligent warehousing environment where the AGV is located through LiDAR, and process the three-dimensional point data using a three-dimensional filtering algorithm to remove noise and impurities and generate a three-dimensional mesh map.

[0049] The integration module is used to acquire obstacle distance data between obstacles and AGV through millimeter-wave radar, integrate the three-dimensional mesh map and the obstacle distance data, and based on the integrated result, combined with the position of AGV, update the spatial distribution relationship between AGV and obstacles, and output dynamic environmental state quantities with timestamps.

[0050] The second generation module is used to input the dynamic environment state quantity into the path planner built based on the reinforcement learning network, determine that the intelligent warehouse environment is a dense shelf environment, and generate the turning angle and speed command for the AGV to pass through the dense shelf environment without collision.

[0051] The adjustment module is used to dynamically adjust the steering stiffness and braking force of the AGV based on the fuzzy PID controller and the steering angle and speed command, so as to achieve safe navigation and obstacle avoidance control of the AGV in dense shelving environments.

[0052] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement the AGV safety navigation and obstacle avoidance control method based on an intelligent warehousing environment as described in any of the first aspects.

[0053] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a method for safe navigation and obstacle avoidance control of AGVs in an intelligent warehousing environment as described in any of the first aspects.

[0054] This application provides a method for safe navigation and obstacle avoidance control of AGVs in an intelligent warehousing environment. The method includes: acquiring three-dimensional point data of the intelligent warehousing environment where the AGV is located using LiDAR; processing the three-dimensional point data using a three-dimensional filtering algorithm to remove noise and artifacts, generating a three-dimensional mesh map; acquiring obstacle distance data between obstacles and the AGV using millimeter-wave radar; integrating the three-dimensional mesh map and obstacle distance data; updating the spatial distribution relationship between the AGV and obstacles based on the integrated result and the AGV's position, outputting a timestamped dynamic environmental state quantity; inputting the dynamic environmental state quantity into a path planner constructed based on a reinforcement learning network to determine that the intelligent warehousing environment is a densely shelved environment, and generating steering angle and speed commands for collision-free passage of the AGV in the densely shelved environment; and dynamically adjusting the steering stiffness and braking force of the AGV based on a fuzzy PID controller according to the steering angle and speed commands to achieve safe navigation and obstacle avoidance control of the AGV in the densely shelved environment.

[0055] This application has the following advantages: It acquires 3D point data using LiDAR and generates a 3D mesh map through 3D filtering, removing noise and artifacts to obtain accurate 3D structural information about the AGV's environment; it acquires obstacle distance data using millimeter-wave radar, integrates the 3D mesh map, and updates the spatial distribution relationship with the AGV's position, outputting a timestamped dynamic environmental state quantity, which can fuse multi-source sensor data to reflect the dynamic spatial relationship between the AGV and obstacles in real time and accurately; it processes the dynamic environmental state quantity using a path planner constructed through a reinforcement learning network, generating steering angle and speed commands, which can accurately identify densely packed shelving environments and output collision-free action commands adapted to these environments; and it adjusts steering stiffness and braking force according to commands using a fuzzy PID controller, dynamically adapting to the complexities of densely packed shelving environments to achieve safe navigation and obstacle avoidance control for the AGV.

[0056] Furthermore, the path planner first decomposes the dynamic environmental state variables, extracts obstacle distribution density and channel space parameters to determine the dense shelving environment, and then maps and generates multiple candidate action combinations. By predicting the trajectory, calculating fixed obstacle safety parameters, and other multi-dimensional parameters, and integrating and evaluating them, the optimal combination is selected as the turning angle and speed command. The corresponding technical effects are: by refining state decomposition, environmental judgment, candidate action generation, and multi-dimensional evaluation and selection, it can more accurately adapt to the characteristics of dense shelving environments. The generated turning angle and speed commands ensure collision-free passage while also taking into account passage smoothness and centering, thus improving the navigation safety and stability of AGVs in dense shelving environments.

[0057] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart of an AGV safety navigation and obstacle avoidance control method based on an intelligent warehousing environment is provided for embodiments of this application;

[0060] Figure 2 A schematic diagram of the structure of an AGV safety navigation and obstacle avoidance system based on an intelligent warehousing environment is provided in this application embodiment;

[0061] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0062] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0063] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 11, 12, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0064] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] To address the issues of identification bias, poor parameter adaptation, and unoptimized control logic in existing technologies, this application provides a method for safe navigation and obstacle avoidance control of AGVs in an intelligent warehousing environment. This method employs the following concept: Three-dimensional environmental data is acquired using LiDAR and processed to generate a three-dimensional mesh map; simultaneously, distance data between obstacles and the AGV is acquired using millimeter-wave radar. These two types of data are integrated with the AGV's own position to update the spatial relationship between the AGV and obstacles in real time and record the time. Then, an intelligent planning method processes the above information to identify densely shelved environments and generate suitable turning and speed commands. Finally, the AGV's turning flexibility and braking force are dynamically adjusted according to the commands. This approach enables accurate environmental perception to reduce positional judgment bias, rapid response to dynamic obstacles to avoid reaction lag, and optimized control in densely shelved areas to prevent collisions with shelving, thereby achieving safe navigation and obstacle avoidance for AGVs in an intelligent warehousing environment.

[0066] Figure 1 A flowchart of an AGV safety navigation and obstacle avoidance control method based on an intelligent warehousing environment is provided for embodiments of this application, such as... Figure 1 As shown, the method includes:

[0067] S11. Collect three-dimensional point data of the intelligent warehousing environment where the AGV is located using LiDAR, and process the three-dimensional point data using a three-dimensional filtering algorithm to remove noise and impurities, and generate a three-dimensional mesh map.

[0068] Among them, LiDAR is a device used to scan the intelligent warehousing environment around AGV and obtain a large number of three-dimensional position information. The three-dimensional point data is a set of point coordinates that reflect the position of the surface of objects in the environment, obtained by the device. The three-dimensional filtering algorithm is a method to filter these three-dimensional point data to remove useless points (i.e., noise) and scattered redundant points (i.e., noisy points) caused by interference. The three-dimensional mesh map is a graphic that converts the processed effective three-dimensional point data into a graph that divides the environmental space into multiple small squares and marks whether there are objects in each square.

[0069] In this embodiment, the LiDAR on the AGV first scans the smart warehouse environment to collect three-dimensional point data of surrounding objects. Then, a three-dimensional filtering algorithm is used to process these data to remove noise and specks. Finally, the processed valid data is converted into a three-dimensional mesh map, which divides the environment into multiple small squares and marks whether there are objects in each square. For example, in the smart warehouse environment, the LiDAR scan obtains 10,000 three-dimensional point data. After three-dimensional filtering, 2,000 noise and specks are removed, and the remaining 8,000 valid points are used to generate a three-dimensional mesh map. Each square represents a 0.5m × 0.5m area, and squares with shelves and aisles without objects are marked.

[0070] S12. Obtain obstacle distance data between obstacles and AGV through millimeter-wave radar, integrate the 3D mesh map and obstacle distance data, and based on the integrated result, combined with the position of AGV, update the spatial distribution relationship between AGV and obstacles, and output dynamic environmental state quantities with timestamps.

[0071] Among them, millimeter-wave radar is a device used to measure the straight-line distance between AGV and surrounding obstacles. Obstacle distance data is the specific distance value measured by this device. Integration is the process of merging the three-dimensional grid map with the obstacle distance data. The position of AGV refers to the current coordinates of AGV in the warehouse environment. The spatial distribution relationship between AGV and obstacles refers to the relative position of AGV and all surrounding obstacles. The dynamic environmental state quantity with timestamp is a data set that contains current time information and can reflect the positional relationship between AGV and obstacles in real time.

[0072] In this embodiment, firstly, the millimeter-wave radar on the AGV continuously measures the distance between itself and surrounding obstacles to obtain obstacle distance data. For example, it measures the distance between the front shelf and the moving AGV every 0.1 seconds to obtain data such as 10 meters and 3 meters. Next, these distance data are merged with the previously generated three-dimensional mesh map, that is, the distance information is mapped to the corresponding position on the mesh map. Then, combined with the current coordinates of the AGV, such as X = 5 meters and Y = 3 meters, the specific position of the obstacle on the mesh map is calculated, and the relative positional relationship between the AGV and the obstacle is updated. Finally, the current time, such as 10:05:30, is added to the updated positional relationship to form a dynamic environmental state quantity and output it.

[0073] S13. Input the dynamic environment state variables into the path planner built based on the reinforcement learning network, determine that the intelligent warehouse environment is a dense shelf environment, and generate the turning angle and speed instructions for the AGV to pass through the dense shelf environment without collision.

[0074] Among them, the path planner built based on reinforcement learning network is a tool that can learn and plan the AGV driving path autonomously according to environmental information. The dense shelf environment refers to the storage area with a large number of shelves, closely arranged and narrow aisle width. The turning angle is the rotation angle when the AGV changes the driving direction. The speed command is the command to control the speed of the AGV. Collision-free passage means that the AGV does not collide with surrounding objects during the driving process.

[0075] In this embodiment, firstly, the dynamic environmental state data with timestamps is input into the path planner, for example, the input includes data on the positional relationship between the AGV and the shelves and the moving AGV at 14:30:25; secondly, the planner identifies whether the current environment is a dense shelf environment based on the density of the shelves and the width of the aisles in the grid diagram, such as if the aisle width is <1.5 meters. Then, after determining that it is a dense environment, the planner generates multiple possible combinations of turning angles (e.g., 5 degrees, 10 degrees) and speeds (e.g., 0.5 m / s, 1 m / s); finally, by judging whether these combinations will cause the AGV to collide with obstacles, a safe turning angle and speed command is selected. For example, when the moving AGV is 3 meters away, a combination of a turning angle of 5 degrees and a speed of 0.5 m / s is selected to ensure that no collision occurs.

[0076] S14. Based on the fuzzy PID controller, the steering stiffness and braking force of the AGV are dynamically adjusted according to the steering angle and speed commands to achieve safe navigation and obstacle avoidance control of the AGV in dense rack environments.

[0077] Among them, the fuzzy PID controller is a device that can flexibly adjust control parameters according to input instructions; steering stiffness refers to the ease or difficulty of AGV steering (higher stiffness means more stable steering, lower stiffness means more flexible steering); braking force refers to the force of AGV braking; and safe navigation and obstacle avoidance control refers to the control process that ensures AGV safe driving and avoidance of obstacles.

[0078] In this embodiment, firstly, the fuzzy PID controller receives steering angle and speed commands generated by the path planner, such as a steering angle of 8 degrees and a speed of 0.4 m / s. Secondly, it adjusts the steering stiffness according to the steering angle, reducing stiffness to increase flexibility when the angle is large and increasing stiffness to maintain stability when the angle is small. Simultaneously, it adjusts the braking force according to the speed command, increasing the force to shorten the braking distance when the speed is high and reducing the force to avoid sudden braking when the speed is low. Finally, through these two adjustments, the AGV is made to drive safely according to the commands. For example, when the steering angle is 8 degrees, the steering stiffness is reduced to allow the AGV to turn flexibly, and when the speed is 0.4 m / s, a moderate braking force is maintained for easy adjustment.

[0079] This application provides the following specific example: In warehouse A, after the AGV starts, its onboard LiDAR scans the surrounding environment 360 degrees, acquiring 15,000 three-dimensional point data points including shelves, columns, and other objects within 10 seconds. After three-dimensional filtering to remove 3,000 noise and artifacts, a three-dimensional grid map with a grid size of 0.5m × 0.5m × 0.5m is generated, clearly marking the positions of shelves and aisles. Subsequently, millimeter-wave radar measures the distance to surrounding obstacles every 0.1 seconds, obtaining obstacle distance data of 8 meters for the left shelf and 4 meters for the right moving AGV. This data is then integrated with the three-dimensional grid map. Based on the AGV's current coordinates (X = 10 meters, Y = 2 meters), the relative position relationship is updated and the time 14:30:25 is marked, generating a dynamic environmental state quantity. After this state quantity is input into the path planner, the planner detects that the channel width is 1.2 meters (determined to be a densely packed shelf environment), generates 5 sets of turning angle and speed combinations, and after simulation, a safe command of 8 degrees turning and 0.4 meters / second speed is selected. Finally, the fuzzy PID controller adjusts the turning stiffness from 5 to 3 according to the command (to increase flexibility), and keeps the braking force at a moderate level, so that the AGV can turn smoothly between dense shelves without collision.

[0080] By executing S11 to S14, this embodiment of the application, through the collaborative perception, data integration, and dynamic updating of LiDAR and millimeter-wave radar, combined with the synergistic effect of reinforcement learning path planning and fuzzy PID control, can accurately capture the three-dimensional structure and obstacle dynamics of the intelligent warehousing environment, accurately identify dense shelving scenarios and generate appropriate driving commands, dynamically adjust the steering flexibility and braking force of the AGV, and ultimately achieve reliable control of the AGV in complex warehousing environments from environmental perception to safe driving, ensuring collision-free passage and stable navigation.

[0081] In one possible embodiment, S13, the dynamic environment state variables are input into the path planner constructed based on the reinforcement learning network, the intelligent warehousing environment is determined to be a dense shelving environment, and steering angle and speed commands for the AGV to pass through the dense shelving environment without collision are generated, including:

[0082] Step 131: Input the dynamic environment state variables into the path planner built based on the reinforcement learning network. Through the decomposition module of the path planner, the dynamic environment state variables are decomposed into the current coordinates of the AGV, obstacle distribution data and timestamp information.

[0083] Among them, the dynamic environment state quantity is a data set containing the positional relationship between the AGV and obstacles and the time. The path planner built based on the reinforcement learning network is a tool that can autonomously plan paths. The decomposition module is the part of the path planner that splits the dynamic environment state quantity. The current coordinates of the AGV are the location information of the AGV in the warehouse. The obstacle distribution data is the information reflecting the location and quantity of obstacles. The timestamp information is a marker that records the time when the data was generated.

[0084] In this embodiment of the application, the dynamic environmental state quantity is input into the path planner, and the decomposition module of the path planner processes it to extract the current coordinates of the AGV, obstacle distribution data and timestamp information. For example, after inputting data containing the position and time of the AGV and surrounding objects, the distribution of the AGV at the position of X=15 meters and Y=6 meters, with 4 shelves and 2 mobile AGVs around it, as well as the time 14:30:00, is extracted.

[0085] Step 132: Extract the obstacle distribution density and channel spatial parameters from the obstacle distribution data using the extraction module of the path planner.

[0086] The extraction module is the part of the path planner that extracts specific information from obstacle distribution data. Obstacle distribution density is the number of obstacles per unit space, and channel space parameters are information reflecting the channel width.

[0087] In this embodiment of the application, the extraction module receives obstacle distribution data, counts the number of obstacles in a unit area to calculate the obstacle distribution density. For example, if there are 20 obstacles in a 10m×10m area, the density is calculated by dividing 20 by 100 (i.e., 10×10) to get 0.2 obstacles per square meter. At the same time, the width of the gap between obstacles is measured to obtain the aisle space parameters. For example, the aisle width is obtained by measuring the distance between shelves. Finally, these two parameters are output.

[0088] Step 133: The path planner's judgment module determines intelligent warehousing environments with obstacle distribution density greater than the preset density value and channel space parameters less than the preset width value as dense shelving environments.

[0089] The determination module is the part of the path planner that determines the environment type. The preset density value is a reference value for determining whether obstacles are dense, and the preset width value is a reference value for determining whether the passage is narrow. A dense shelf environment is an environment with many obstacles and narrow passages.

[0090] In this embodiment, the determination module receives the obstacle distribution density and the channel space parameters, compares the obstacle distribution density with a preset density value, and compares the channel space parameters with a preset width value. When the obstacle distribution density is greater than the preset density value and the channel space parameters are less than the preset width value, it is determined to be a dense shelving environment. For example, if the preset density value is 0.15 per square meter and the preset width value is 1.6 meters, and the actual density is 0.2 per square meter and the width is 1.4 meters, it is determined to be a dense environment.

[0091] Step 134: In a densely packed shelving environment, the current coordinates and timestamp information of the AGV are mapped into multiple candidate action combinations containing steering angle and speed parameters through the mapping module of the path planner.

[0092] The mapping module is the part of the path planner that converts AGV coordinates and timestamp information, and the candidate action combination is a set of multiple driving schemes that include steering angle and speed parameters.

[0093] In this embodiment of the application, in a densely packed shelving environment, the mapping module receives the current coordinates and timestamp information of the AGV and converts them into multiple candidate action combinations. For example, based on the AGV's position at X=15 meters, Y=6 meters and the time 14:30:00, combinations such as turning 4 degrees with a speed of 0.6 m / s and turning 6 degrees with a speed of 0.5 m / s are generated.

[0094] Step 135: Through the selection module of the path planner, select the candidate action combination that matches the obstacle distribution data from multiple candidate action combinations, and use it as the steering angle and speed command for the AGV to pass through the dense shelf environment without collision.

[0095] Among them, the selection module is the part of the path planner that filters candidate action combinations, matching means that the candidate action combination is adapted to the obstacle distribution data, and collision-free passage means that the AGV does not collide with obstacles when it travels.

[0096] In this embodiment, the selection module receives multiple candidate action combinations and obstacle distribution data, analyzes whether the driving route corresponding to each combination avoids obstacles, and selects the combination that matches the obstacle distribution data as the steering angle and speed command. For example, the analysis finds that the combination of steering at 6 degrees and speed at 0.5 m / s can avoid all obstacles, so this combination is selected.

[0097] This application provides the following specific example: In warehouse A, after the dynamic environmental state variables are input into the path planner, the decomposition module breaks them down into the current coordinates of the AGV (X=10 meters, Y=4 meters), obstacle distribution data including 5 shelves and 3 mobile AGVs, and the timestamp 10:30:00; the extraction module processes the obstacle distribution data, counts 20 obstacles in a 10m×10m area, calculates an obstacle distribution density of 0.2 obstacles per square meter, and measures the shelf gaps to obtain a passage space parameter of 1.3 meters; the judgment module compares the obstacle distribution density of 0.2 obstacles / square meter with the preset density value of 0.15 obstacles / square meter, and the passage space parameter of 1.3 meters with the preset width value of 1.6 meters, and judges it as a dense shelf environment; the mapping module generates multiple candidate action combinations such as turning 5 degrees with a speed of 0.6 m / s and turning 7 degrees with a speed of 0.4 m / s based on the current coordinates and timestamp of the AGV; the selection module analyzes and finds that the combination of turning 7 degrees and speed of 0.4 m / s can avoid all obstacles, and finally uses it as the turning angle and speed command.

[0098] By executing steps 131 to 135, this embodiment of the application obtains key information by decomposing dynamic environmental state quantities, extracts obstacle distribution density and channel space parameters to determine dense shelf environments, maps and generates multiple candidate action combinations, and selects a scheme that matches the obstacle distribution. It can accurately identify dense shelf scenarios from complex environmental data, generate appropriate steering angle and speed commands, provide reliable action basis for AGV to drive safely in dense shelf environments, and ensure that the driving process is adapted to the obstacle distribution.

[0099] In one possible embodiment, step 135, selecting a candidate action combination that matches the obstacle distribution data from multiple candidate action combinations, as the steering angle and speed command for the AGV to pass through a densely packed shelf environment without collision, includes:

[0100] a1. Based on obstacle distribution data, determine the predicted travel trajectory corresponding to each candidate action combination. The predicted travel trajectory includes a series of future position points of the AGV when executing the candidate action combination.

[0101] The obstacle distribution data reflects the location and type of surrounding obstacles, including the positions of fixed objects like shelves and other moving objects like AGVs; the candidate action combination is a driving plan that includes steering angle and speed; the predicted travel trajectory is the route that the AGV may take when executing a certain candidate action combination; and the future location point is the specific location that the AGV will reach at different times on this route. These location points together constitute the predicted travel trajectory.

[0102] In this embodiment, based on obstacle distribution data, the AGV's travel route is simulated for each candidate action combination to determine the corresponding predicted travel trajectory. Specifically, the AGV's position at each moment in the future is calculated according to the turning angle and speed in the candidate action combination. For example, a candidate action combination is a 5-degree turn and a speed of 0.5 m / s. After 1 second, the AGV travels 0.5 meters along this turning angle to reach X=5 meters and Y=3 meters. After 2 seconds, it travels another 0.5 meters to reach X=5.5 meters and Y=3.2 meters. After 3 seconds, it reaches X=6 meters and Y=3.4 meters. These position points constitute the predicted travel trajectory of this combination.

[0103] a2. For each predicted travel trajectory, calculate the distance between each future location point and the nearest fixed obstacle as a fixed obstacle safety parameter; calculate the distance between each future location point and the nearest moving obstacle as a moving obstacle safety parameter; calculate the degree to which each future location point deviates from the center of the travel channel as a travel centering parameter; and calculate the change in the turning angle of adjacent future location points as a turning smoothness parameter.

[0104] Among them, the fixed obstacle safety parameter is the distance between the future location point and the nearest fixed obstacle; the moving obstacle safety parameter is the distance between the future location point and the nearest moving obstacle; the passage centering parameter is the deviation between the future location point and the center position of the passage; and the steering smoothness parameter is the difference between the steering angles corresponding to two adjacent future location points.

[0105] In this embodiment, for each predicted travel trajectory, various parameters of the future location point are calculated one by one. The distance to the nearest fixed obstacle is calculated as a fixed obstacle safety parameter, for example, the distance from a location point to the nearest shelf is measured with a ruler and found to be 1.5 meters; the distance to the nearest moving obstacle is calculated as a moving obstacle safety parameter, for example, the distance to another AGV is measured and found to be 2 meters; the deviation of the location point from the center of the passage is measured as a travel centering parameter, for example, the deviation from the center is 0.2 meters; the difference in turning angle between adjacent location points is calculated as a turning smoothness parameter, for example, if the previous location turns 5 degrees and the next location turns 5.5 degrees, the difference of 0.5 degrees is the parameter.

[0106] a3. Integrate the safety parameters of fixed obstacles, moving obstacles, passage centering parameters, and turning smoothness parameters to generate a comprehensive evaluation value for each candidate action combination.

[0107] Integration is the process of combining fixed obstacle safety parameters, moving obstacle safety parameters, passage centering parameters, and turning smoothness parameters; the comprehensive evaluation value is a numerical value obtained by combining these parameters, reflecting the overall quality of a candidate action combination.

[0108] In this embodiment, the four parameters corresponding to each candidate action combination are comprehensively processed. First, a scoring standard is set for each parameter: the closer the distance, the lower the score; the smaller the deviation and the smaller the angle change, the higher the score. Then, the parameter scores of all position points are added together to obtain the comprehensive evaluation value of the combination. For example, the total score for the fixed obstacle safety parameter of a certain combination is 20 points, the total score for the moving obstacle safety parameter is 18 points, the total score for the passage centering parameter is 15 points, and the total score for the turning smooth parameter is 16 points. The sum of these four scores gives a comprehensive evaluation value of 39 points.

[0109] a4. Select the candidate action combination with the best comprehensive evaluation value as the steering angle and speed command for AGV to pass through densely packed shelving environments without collision.

[0110] Among them, the best comprehensive evaluation value refers to the one with the highest comprehensive evaluation value among all candidate action combinations; collision-free passage means that the AGV will not collide with surrounding obstacles when it is traveling; steering angle and speed commands are the final steering and speed parameters of the AGV when it is traveling.

[0111] In this embodiment, the comprehensive evaluation values ​​of all candidate action combinations are compared, and the combination with the highest value is selected. The steering angle and speed contained in this combination are used as the driving command for the AGV. For example, if there are three candidate combinations with comprehensive evaluation values ​​of 35, 39, and 37 points respectively, and 39 points is the highest, then the combination corresponding to this score is selected as the final command.

[0112] This application provides the following specific example: In warehouse A, based on obstacle distribution data including 3 shelves and 2 mobile AGVs, a simulated and predicted passage trajectory is generated for candidate action combinations such as turning 5 degrees and speed of 0.5 m / s. One particular combination's trajectory includes position points from 1 to 4 seconds in the future: after 1 second, reaching X=7 meters, Y=4 meters; after 2 seconds, reaching X=7.5 meters, Y=4.2 meters; after 3 seconds, reaching X=8 meters, Y=4.4 meters; and after 4 seconds, reaching X=8.5 meters, Y=4.6 meters. The parameters for this trajectory are calculated as follows: fixed obstacle safety parameters score 21 points, moving obstacle safety parameters score 18 points, passage centering parameters score 14 points, and turning smoothness parameters score 15 points, for a total of 21 + 18 + 14 + 15 = 68 points. Among the other combinations, the combination with a steering angle of 6 degrees and a speed of 0.4 m / s has a comprehensive evaluation score of 75 points, while the combination with a steering angle of 7 degrees and a speed of 0.3 m / s has a score of 70 points. 75 points is the highest, so the combination corresponding to this score is selected as the steering angle and speed command.

[0113] By executing a1 to a4, this embodiment of the application generates a predicted travel trajectory for each candidate action combination, calculates parameters from three dimensions of safety, centering, and stability, and comprehensively evaluates them. This allows for a comprehensive consideration of the applicability of each combination, and finally selects the solution with the best overall performance as the driving instruction. This ensures that the AGV can avoid fixed and moving obstacles in densely packed shelving environments, while maintaining the stability and centering of its travel, thus achieving safe and reliable passage.

[0114] In one possible embodiment, S11, the 3D point data of the intelligent warehousing environment where the AGV is located is collected by LiDAR, and the 3D point data is processed using a 3D filtering algorithm to remove noise and artifacts, generating a 3D mesh map, including:

[0115] Step 111: Remove data that exceeds the preset distance range from the 3D point data, and retain the initial 3D point data within the effective sensing area around the AGV.

[0116] Among them, the three-dimensional point data is the location information of a large number of points obtained by LiDAR scanning the intelligent warehouse environment, the preset distance range is the surrounding space distance that the AGV needs to pay attention to, the effective perception area is the space within the preset distance range, and the initial three-dimensional point data is the data that is filtered from the three-dimensional point data and located within the effective perception area.

[0117] In this embodiment, points within a preset distance range from the AGV are identified from the three-dimensional point data acquired by the LiDAR, and points exceeding this range are removed to obtain initial three-dimensional point data. For example, if the preset distance range is 0 to 10 meters, then points within 10 meters of the AGV are selected from all three-dimensional points, and points beyond 10 meters are removed. These points within 10 meters constitute the initial three-dimensional point data.

[0118] Step 112: Filter the initial 3D point data to obtain the target 3D point data after removing noise and speckles.

[0119] The initial 3D point data is the data obtained in step 111. The filtering process is a method to remove useless interference points from the data. Noise is random points caused by environmental interference. Random points are scattered points that do not reflect the actual surface of the object. The target 3D point data is 3D point data that has been filtered to remove noise and random points and can accurately reflect the environmental objects.

[0120] In this embodiment of the application, the initial three-dimensional point data obtained in step 111 is filtered to remove noise and specks, thereby obtaining the target three-dimensional point data. For example, the initial data may contain random points caused by light reflection and scattered points formed by dust. After filtering, these points are removed, and the remaining points can clearly show the surfaces of objects such as shelves and the ground.

[0121] Step 113: Divide the target 3D point data into multiple spatial grid units according to a preset size, and calculate the spatial interval between adjacent 3D points in the target 3D point data to obtain 3D points with spatial intervals that conform to the preset spatial intervals.

[0122] Among them, the target 3D point data is the data processed in step 112, the preset size is the size of the spatial grid, the spatial grid cell is a small space divided according to the preset size, the spatial interval is the distance between two adjacent 3D points, and the preset spatial interval is a pre-set distance standard for judging whether a point is valid.

[0123] In this embodiment, the target 3D point data obtained in step 112 is divided into multiple spatial grid cells according to a preset size. The distance (spatial interval) between adjacent 3D points within each grid is calculated, and 3D points with spatial intervals conforming to the preset spatial interval are retained. For example, if the preset size is 0.5m × 0.5m × 0.5m and the preset spatial interval is 0.1 to 0.3m, after dividing the grid, the distance between adjacent points within each grid is calculated, and points with a distance between 0.1 and 0.3 meters are retained.

[0124] Step 114: Record the number of 3D points distributed within the intelligent warehousing environment in each grid cell.

[0125] Among them, the spatial grid cell is a small space divided in step 113, and the number of distributions is the number of three-dimensional points contained in each spatial grid cell. The three-dimensional points in the intelligent warehousing environment refer to the points on the surface of objects in the environment.

[0126] In this embodiment of the application, the number of three-dimensional points contained in each spatial grid cell is counted for the three-dimensional points retained after processing in step 113, and the distribution quantity of each grid cell is recorded. For example, if there are 8 three-dimensional points in a spatial grid cell, the distribution quantity of that grid cell is recorded as 8.

[0127] Step 115: Determine the grid state based on the distribution quantity, and generate a 3D grid map containing obstacle areas and passage areas based on the grid state.

[0128] Among them, the number of distributions is the number of three-dimensional points in each grid cell recorded in step 114, the grid state refers to whether there are objects (such as obstacles such as shelves) in the grid cell, the obstacle area is the area composed of grid cells with objects, the passable area is the area composed of grid cells without objects, and the three-dimensional grid map is a graphic that marks the obstacle area and the passable area.

[0129] In this embodiment, based on the distribution quantity of each grid cell recorded in step 114, the grid state is determined. Grid cells with a large distribution quantity are identified as having objects (i.e., belonging to obstacle areas), while those with a small distribution quantity are identified as having no objects (i.e., belonging to passage areas). These grid cells are then combined to generate a three-dimensional grid map containing obstacle areas and passage areas. For example, grid cells with a distribution quantity greater than 6 are defined as obstacle areas, and those with a quantity less than or equal to 6 are defined as passage areas. After being marked in this way, they are combined to form a three-dimensional grid map.

[0130] This application provides the following specific example: In warehouse A, from the 3D point data obtained by LiDAR scanning, points within a range of 0 to 10 meters from the AGV are first selected, and points beyond 10 meters are removed to obtain initial 3D point data; the initial data is filtered to remove interference points caused by reflection and dust, resulting in target 3D point data; the target data is divided into spatial grid cells of 0.5m × 0.5m × 0.5m, and the distance between adjacent points in each grid is calculated, retaining points with an interval between 0.1 and 0.3 meters; the number of 3D points in each grid cell is counted, with 12 points for the grid corresponding to the shelves and 3 points for the grid corresponding to the aisles; finally, a 3D grid map that clearly distinguishes between shelves and aisles is generated according to the standard that areas with more than 6 points are obstacle areas and areas with less than or equal to 6 points are passage areas.

[0131] By executing steps 111 to 115, this embodiment of the application filters the 3D point data of the effective area around the AGV, removes interfering points and standardizes the distribution of points, and then distinguishes obstacles and passable areas by counting the number of points within the grid, ultimately generating an intuitive 3D grid map. This process can transform raw environmental data into clear environmental graphic information, accurately reflecting the location of obstacles and passable areas in intelligent warehousing, providing reliable basic information for AGV navigation in complex environments, and ensuring that the AGV can clearly perceive the surrounding environmental structure.

[0132] In one possible embodiment, S12, obstacle distance data between the obstacle and the AGV is acquired via millimeter-wave radar; the 3D mesh map and obstacle distance data are integrated; based on the integrated result and the position of the AGV, the spatial distribution relationship between the AGV and the obstacle is updated; and a dynamic environmental state quantity with a timestamp is output, including:

[0133] Step 121: Use millimeter-wave radar to detect fixed and moving obstacles around the AGV, and record the straight-line distance between each obstacle and the AGV as obstacle distance data.

[0134] Among them, millimeter-wave radar is a device used to detect objects around AGVs and can measure the straight-line distance between objects and AGVs; fixed obstacles refer to objects in the warehouse whose position does not change, such as shelves; moving obstacles refer to objects whose position changes, such as other AGVs; obstacle distance data is the straight-line distance information between each obstacle and AGV recorded by millimeter-wave radar.

[0135] In this embodiment, millimeter-wave radar is used to scan the area around the AGV to locate fixed and moving obstacles. The straight-line distance between each obstacle and the AGV is measured and recorded. This recorded distance information constitutes the obstacle distance data. For example, if the right-side shelf is detected to be 6 meters away from the AGV and the moving AGV behind it is 2 meters away, these 6-meter and 2-meter values ​​constitute the obstacle distance data.

[0136] Step 122: Based on the obstacle distance data, mark the location of each obstacle in the corresponding grid cell of the 3D mesh map so that the corresponding grid cell contains both 3D point distribution information and obstacle distance information.

[0137] Among them, obstacle distance data is the distance information obtained in step 121; the three-dimensional mesh map is a graphic that marks objects and passable areas; the mesh cell is a small square in the three-dimensional mesh map; the three-dimensional point distribution information is the distribution of points on the surface of the object in the mesh cell; and the obstacle distance information is the distance data between obstacles and AGV in the mesh cell.

[0138] In this embodiment, the position of each obstacle in the 3D mesh map is determined based on obstacle distance data, and they are marked in the corresponding mesh cells. These mesh cells contain both the original 3D point distribution information and the new obstacle distance information. For example, if a moving AGV is 4 meters away from another AGV, it corresponds to a mesh cell in the 3D mesh map with X=5 meters and Y=8 meters. This cell simultaneously contains the 3D point distribution of the AGV and the distance information of 4 meters.

[0139] Step 123: Based on the position of the AGV, calculate the orientation and spacing of obstacles in the corresponding grid cell relative to the AGV.

[0140] Here, the AGV's position is its current coordinates in the warehouse; the orientation is the direction of the obstacle relative to the AGV, such as in front or to the right; the spacing is the straight-line distance between the obstacle and the AGV; and the corresponding grid cell is the grid cell marked with the obstacle.

[0141] In this embodiment, by combining the current coordinates of the AGV and examining the coordinates of the grid cell containing the obstacle, the direction (orientation) of the obstacle on the AGV and the distance (spacing) between them can be calculated. For example, if the AGV is in a grid cell with X=7 meters and Y=3 meters, and the obstacle is in a grid cell with X=7 meters and Y=6 meters, it can be concluded that the obstacle is directly in front of the AGV, and the spacing is 3 meters (6-3=3).

[0142] Step 124: Based on the orientation and spacing, adjust the three-dimensional point distribution information and obstacle distance information of each obstacle to generate the updated spatial distribution relationship between the AGV and the obstacles.

[0143] Among them, orientation is the direction of the obstacle relative to the AGV; spacing is the distance between the obstacle and the AGV; three-dimensional point distribution information is the distribution of points on the surface of the object in the grid cell; obstacle distance information is the distance data between the obstacle and the AGV in the grid cell; the spatial distribution relationship between the AGV and the obstacles refers to the relative position of the AGV and all obstacles.

[0144] In this embodiment, based on the orientation and spacing obtained in step 123, the three-dimensional point distribution information and obstacle distance information corresponding to the obstacle are adjusted, and the relative positional relationship between the AGV and the obstacle is updated to obtain a new spatial distribution relationship. For example, if the orientation of the moving obstacle changes from left to front left and the spacing changes from 5 meters to 4 meters, its information in the grid is adjusted according to these changes.

[0145] Step 125: Add corresponding time information to the spatial distribution relationship and output the dynamic environment state quantity with timestamp.

[0146] Among them, spatial distribution relationship refers to the relative position of AGV and obstacles; time information refers to the specific time of data recording; timestamp is a label that marks the time when the data was generated; and dynamic environmental state quantity with timestamp is data containing time information that reflects the real-time positional relationship between AGV and obstacles.

[0147] In this embodiment, the spatial distribution relationship updated in step 124 is supplemented with corresponding time information, such as time down to the second, to form a timestamped dynamic environment state quantity and output it. For example, after the spatial distribution relationship is updated, the current time is marked as 09:45:10, and this information including time is the timestamped dynamic environment state quantity.

[0148] This application provides the following specific example: In warehouse A, millimeter-wave radar detects two shelves and one mobile AGV, recording their distances to the AGV as 5 meters, 7 meters, and 3 meters, respectively, forming obstacle distance data; based on these distances, the shelves and mobile AGV are marked in the corresponding grid cells of the 3D mesh map, so that these cells simultaneously contain 3D point distribution and distance information; combined with the current position of the AGV (X=9 meters, Y=4 meters), it is calculated that the 5-meter shelf is directly to the left, and the 3-meter mobile AGV is directly in front; based on the orientation and spacing adjustment information, the spatial distribution relationship between the AGV and them is updated; finally, the time 11:30:05 is added to this relationship, and the dynamic environmental state quantity with timestamp is output.

[0149] By executing steps 121 to 125, this embodiment of the application acquires obstacle distances using millimeter-wave radar, integrates the distance information with a 3D mesh map, determines the direction and distance of obstacles based on the AGV's position, updates spatial relationships, and adds time markers. This allows the AGV to monitor the position and changes of surrounding obstacles in real time. This process integrates various environmental information, providing timely and accurate data for the AGV's subsequent path planning, and helping the AGV to operate safely in dynamic environments.

[0150] In one possible embodiment, step 124, adjusting the three-dimensional point distribution information and obstacle distance information of each obstacle according to its orientation and spacing, and generating an updated spatial distribution relationship between the AGV and the obstacles, includes:

[0151] b1. Determine the category of the obstacle to distinguish between fixed and moving obstacles.

[0152] The category of an obstacle refers to whether it is fixed or mobile. Fixed obstacles are objects that remain in place, such as shelves, while mobile obstacles are objects that change position, such as other AGVs. The distinction between the two is to allow for different handling methods for different types of obstacles.

[0153] In this embodiment of the application, the category of an obstacle is determined by observing its movement state. Obstacles whose positions remain unchanged are identified as fixed obstacles, while those whose positions change are identified as moving obstacles. For example, a shelf that is always in the same position is identified as a fixed obstacle, while another AGV that is moving is identified as a moving obstacle.

[0154] b2. For each fixed obstacle, when the position corresponding to the three-dimensional point distribution information of the obstacle is inconsistent with the position derived based on orientation and spacing, the coordinate data in the three-dimensional point distribution information of the obstacle is corrected to obtain the corrected three-dimensional point distribution information of the obstacle.

[0155] Among them, fixed obstacles are obstacles whose positions remain unchanged, three-dimensional point distribution information is the position data of points on the surface of the obstacle, orientation is the direction of the obstacle relative to the AGV, spacing is the distance between the obstacle and the AGV, the position derived from orientation and spacing is the position of the obstacle calculated based on the direction and distance, and the corrected three-dimensional point distribution information is the point data that accurately reflects the position of the obstacle after adjustment.

[0156] In this embodiment, for each fixed obstacle, the position corresponding to its three-dimensional point distribution information is compared with the position calculated based on orientation and spacing. If they are inconsistent, the coordinate data in the three-dimensional point distribution information is adjusted to obtain the corrected information. For example, if the three-dimensional point distribution information of a shelf is displayed at X=5 meters, but the orientation and spacing indicate that it should be at X=6 meters, then the X coordinate in the three-dimensional point distribution information is changed to 6 meters.

[0157] b3. For each moving obstacle, when there is a difference between the obstacle distance information and the spacing, the obstacle distance information is updated according to the spacing to obtain the updated obstacle distance information.

[0158] Among them, moving obstacles are obstacles whose positions change, obstacle distance information is the distance between the obstacle and the AGV recorded by millimeter-wave radar, spacing is the distance calculated based on orientation and AGV position, difference refers to the difference between the obstacle distance information and the spacing value, and the updated obstacle distance information is the adjusted distance data consistent with the spacing.

[0159] In this embodiment of the application, for each moving obstacle, its obstacle distance information is compared with the spacing. If there is a difference, the obstacle distance information is adjusted according to the spacing to obtain updated information. For example, if the obstacle distance information of a moving obstacle is 4 meters and the calculated spacing is 3.5 meters, the obstacle distance information is updated to 3.5 meters.

[0160] b4. Based on the corrected 3D point distribution information of obstacles and the updated obstacle distance information, combined with the orientation of fixed obstacles and moving obstacles, determine the position parameters of each obstacle in 3D space.

[0161] Among them, the corrected 3D point distribution information is the accurate location point data of the fixed obstacle, the updated obstacle distance information is the accurate distance data of the moving obstacle, the orientation is the direction of the obstacle relative to the AGV, and the position parameter is the coordinate data of the obstacle in 3D space.

[0162] In this embodiment, the coordinates of each obstacle in three-dimensional space are calculated as position parameters by combining the corrected three-dimensional point distribution information and orientation of fixed obstacles, and the updated obstacle distance information and orientation of moving obstacles. For example, the corrected information of a fixed obstacle, combined with its orientation to the left of the AGV, yields coordinates of X = 2 meters and Y = 5 meters.

[0163] b5. Based on position parameters, determine the relative positional relationships between obstacles.

[0164] Among them, the position parameter is the coordinate data of the obstacle in three-dimensional space, and the relative position relationship is the orientation and distance between each obstacle.

[0165] In this embodiment of the application, the orientation and distance between each obstacle are calculated based on the position parameters of each obstacle to determine their relative positional relationship. For example, if obstacle A is at X=5 meters and Y=3 meters, and obstacle B is at X=8 meters and Y=3 meters, it is calculated that A is directly to the left of B and at a distance of 3 meters.

[0166] b6. Integrate the relative positional relationship with the current coordinates of the AGV to generate an updated spatial distribution relationship between the AGV and obstacles.

[0167] Among them, the relative positional relationship refers to the positional situation between each obstacle, the current coordinates of the AGV is the position of the AGV in the warehouse, and the updated spatial distribution relationship between the AGV and the obstacles is a comprehensive information that includes the position and mutual relationship of the AGV and all obstacles.

[0168] In this embodiment, the relative positional relationships between obstacles are combined with the current coordinates of the AGV to integrate the positional relationships between the AGV and each obstacle, as well as the positional relationships between obstacles, forming an updated spatial distribution relationship. For example, if the AGV is at X=6 meters and Y=4 meters, the distance and direction between the AGV and each obstacle, as well as the positional relationships between obstacles, are determined by combining the relative positions between obstacles.

[0169] This application provides the following specific example: In warehouse A, obstacles are first divided into two fixed obstacles (shelves) and one mobile obstacle (other AGVs). For the fixed obstacle, its three-dimensional point distribution information corresponds to the position X = 8 meters and Y = 3 meters. Based on the orientation and spacing, it should be X = 9 meters and Y = 3 meters. The X coordinate is then corrected to 9 meters. For the mobile obstacle, its obstacle distance information is 3 meters and the spacing is 2.8 meters. The spacing is then updated to 2.8 meters. Combining the corrected and updated information and orientation, the position parameters of the fixed obstacle are calculated to be X = 9 meters and Y = 3 meters, and the mobile obstacle is X = 10 meters and Y = 5 meters. It is then determined that the fixed obstacle is 1 meter to the left of the mobile obstacle. Finally, combined with the current coordinates of the AGV X = 7 meters and Y = 4 meters, the spatial distribution relationship between the AGV and the obstacles, as well as between the obstacles, is integrated.

[0170] By executing steps b1 to b6, this embodiment of the application distinguishes obstacle types, corrects the position data of fixed obstacles and updates the distance information of moving obstacles, then determines the position of each obstacle and their interrelationships, ultimately integrating a complete spatial distribution relationship including the AGV. This process improves the accuracy of obstacle position information, comprehensively reflects the positional relationships between the AGV and obstacles, and provides accurate environmental information for AGV safe navigation.

[0171] In one possible embodiment, S14, based on a fuzzy PID controller, dynamically adjusts the steering stiffness and braking force of the AGV according to the steering angle and speed commands to achieve safe navigation and obstacle avoidance control of the AGV in densely packed shelving environments, including:

[0172] Step 141: Extract the actual steering deviation corresponding to the steering angle and the actual speed deviation corresponding to the speed command.

[0173] Among them, the steering angle is the angle that the AGV should turn given by the path planner; the actual steering deviation is the difference between the angle that the AGV actually turns and the steering angle; the speed command is the speed that the AGV should travel given by the path planner; and the actual speed deviation is the difference between the actual travel speed of the AGV and the speed command.

[0174] In this embodiment, we first look at the actual angle the AGV turns and the steering angle, and calculate the difference between the two to get the actual steering deviation; then we look at the actual speed of the AGV and the speed command, and calculate the difference between the two to get the actual speed deviation. For example, if the steering angle is 8 degrees, but the AGV only turns 7.6 degrees, the actual steering deviation is 0.4 degrees; if the speed command is 0.5 meters per second, but the AGV actually travels 0.46 meters per second, the actual speed deviation is 0.04 meters per second.

[0175] Step 142: Based on the actual steering deviation and actual speed deviation, and combined with the aisle width characteristics in a densely packed rack environment, determine the corresponding adjustment ranges for the steering stiffness and braking force of the AGV.

[0176] Among them, the actual steering deviation is the steering difference calculated in step 141; the actual speed deviation is the speed difference calculated in step 141; the aisle width characteristic is the width of the aisle in a densely packed rack environment; the steering stiffness is the ease or difficulty of the AGV turning; the braking force is the force of the AGV braking; and the adjustment range is the range within which the steering stiffness and braking force can be adjusted.

[0177] In this embodiment, the adjustment range of steering stiffness and braking force is determined by combining the actual steering deviation and actual speed deviation, and taking into account the width of the aisle in a densely packed shelving environment (for example, a narrow aisle means more caution is required). For example, if the aisle width is 1.3 meters, the actual steering deviation is 0.3 degrees, and the speed deviation is 0.03 meters / second, the adjustment range of steering stiffness is set to [2.6, 4.6], and the adjustment range of braking force is set to [1.7, 3.7].

[0178] Step 143: Calculate the first change in steering stiffness and the second change in braking force based on the adjustment range.

[0179] The adjustment range is the adjustable range of steering stiffness and braking force determined in step 142; the first change is the specific amount of steering stiffness that needs to be adjusted; and the second change is the specific amount of braking force that needs to be adjusted.

[0180] In this embodiment of the application, based on the adjustment range determined in step 142, the steering stiffness is adjusted by a certain amount (first change) in conjunction with the actual steering deviation, and the braking force is adjusted by a certain amount (second change) in conjunction with the actual speed deviation. For example, if the steering stiffness adjustment range is [2.6, 4.6], and the actual steering deviation is 0.3 degrees, the first change is calculated to be 0.6; if the braking force range is [1.7, 3.7], and the speed deviation is 0.03 m / s, the second change is calculated to be 0.5.

[0181] Step 144: Input the first change amount to the steering component of the AGV to obtain steering operation parameters, and input the second change amount to the braking component of the AGV to obtain braking operation parameters.

[0182] Wherein, the first change is the steering stiffness adjustment range calculated in step 143; the second change is the braking force adjustment range calculated in step 143; the steering component is the device responsible for steering the AGV; the braking component is the device responsible for braking the AGV; the steering operation parameters are the specific steering parameters that the steering component can execute; and the braking operation parameters are the specific braking parameters that the braking component can execute.

[0183] In this embodiment, the first change is input to the steering component of the AGV, and the steering component is adjusted according to this magnitude to obtain specific steering operation parameters (e.g., the steering stiffness is adjusted to 3.2); the second change is input to the braking component, and the braking component is adjusted to obtain specific braking operation parameters (e.g., the braking force is adjusted to 2.2).

[0184] Step 145: Obtain the real-time distance between obstacles and AGV in the dynamic environment state variables. Based on the steering operation parameters and braking operation parameters, and combined with the real-time distance, correct and adjust the range to achieve safe navigation and obstacle avoidance control of AGV in densely packed rack environments.

[0185] Among them, the dynamic environment state quantity is the data that records the real-time positional relationship between the obstacle and the AGV; the real-time distance is the current distance between the obstacle and the AGV; the steering operation parameter is the steering parameter obtained in step 144; the braking operation parameter is the braking parameter obtained in step 144; and the correction adjustment range is the optimization of the adjustment range in step 142 based on the real-time distance.

[0186] In this embodiment, the real-time distance between the obstacle and the AGV is found from the dynamic environmental state variables. Combined with the steering and braking operation parameters, if the distance is relatively close (e.g., 1.5 meters), the adjustment range is narrowed (to make steering and braking more stable); if the distance is far (e.g., 3 meters), the range can be appropriately widened. By correcting the adjustment range in this way, the AGV can navigate safely and avoid obstacles.

[0187] This application provides the following specific example: In warehouse A, the AGV's steering angle is 7 degrees, the actual steering angle is 6.8 degrees, and the actual steering deviation is 0.2 degrees; the speed command is 0.4 m / s, the actual speed is 0.38 m / s, and the actual speed deviation is 0.02 m / s. Combining these deviations with a 1.2-meter aisle width, the steering stiffness adjustment range is determined to be [2.5, 4.5], and the braking force adjustment range is [1.6, 3.6]. Based on these ranges, a first change of 0.5 and a second change of 0.4 are calculated. The first change is input into the steering component to obtain the steering operation parameter "steering stiffness adjusted to 3.0"; the second change is input into the braking component to obtain the braking operation parameter "braking force adjusted to 2.0". At this time, the real-time distance between the obstacle and the AGV is obtained as 1.6 meters. Combining the operation parameters, the steering stiffness adjustment range is corrected to [2.8, 4.2], and the braking force adjustment range is corrected to [1.9, 3.3], ensuring the safe operation of the AGV.

[0188] By executing steps 141 to 145, this embodiment of the application extracts the deviation between actual driving and instructions, combines the channel characteristics of densely packed shelving environments to determine the adjustment range of steering and braking, calculates the specific adjustment range and converts it into parameters that the components can execute, and then optimizes the adjustment range based on the real-time distance to obstacles. This process enables the steering and braking adjustments of the AGV to be more precise, adapt to the narrow channels of dense environments, ensure a safe distance from obstacles, and achieve stable and safe navigation and obstacle avoidance.

[0189] Figure 2 A schematic diagram of a safety navigation and obstacle avoidance system for AGVs in an intelligent warehousing environment, provided as an embodiment of this application, is shown below. Figure 2 As shown, the system includes:

[0190] The first generation module 21 is used to collect three-dimensional point data of the intelligent warehousing environment where the AGV is located through LiDAR, and process the three-dimensional point data using a three-dimensional filtering algorithm to remove noise and impurities and generate a three-dimensional mesh map.

[0191] The integration module 22 is used to acquire obstacle distance data between obstacles and AGV through millimeter-wave radar, integrate the three-dimensional mesh map and obstacle distance data, and based on the integrated result, combined with the position of AGV, update the spatial distribution relationship between AGV and obstacles, and output dynamic environmental state quantities with timestamps.

[0192] The second generation module 23 is used to input the dynamic environment state variables into the path planner built based on the reinforcement learning network, determine that the intelligent warehouse environment is a dense shelf environment, and generate the turning angle and speed instructions for the AGV to pass through the dense shelf environment without collision.

[0193] The adjustment module 24 is used to dynamically adjust the steering stiffness and braking force of the AGV based on the fuzzy PID controller and the steering angle and speed commands, so as to realize the safe navigation and obstacle avoidance control of the AGV in the case of dense rack environment.

[0194] Figure 2 The aforementioned AGV safety navigation and obstacle avoidance system based on an intelligent warehousing environment can perform... Figure 1 The implementation principle and technical effects of the AGV safety navigation and obstacle avoidance control method in the intelligent warehousing environment described in the above embodiment will not be repeated here. The specific operation methods of each module and unit in the AGV safety navigation and obstacle avoidance system in the intelligent warehousing environment described in the above embodiment have been described in detail in the embodiments of the relevant method, and will not be elaborated upon here.

[0195] In one possible design, Figure 2 The illustrated embodiment of an AGV safety navigation and obstacle avoidance system in an intelligent warehousing environment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32.

[0196] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0197] The processing component 32 is used to execute the following processes: It acquires 3D point data of the intelligent warehousing environment where the AGV is located using LiDAR, processes the 3D point data using a 3D filtering algorithm to remove noise and artifacts, and generates a 3D mesh map; it acquires obstacle distance data between obstacles and the AGV using millimeter-wave radar, integrates the 3D mesh map and obstacle distance data, updates the spatial distribution relationship between the AGV and obstacles based on the integrated result and the AGV's position, and outputs a timestamped dynamic environment state quantity; it inputs the dynamic environment state quantity into a path planner built based on a reinforcement learning network, determines that the intelligent warehousing environment is a densely shelved environment, and generates steering angle and speed commands for the AGV to pass through the densely shelved environment without collision; based on a fuzzy PID controller, it dynamically adjusts the steering stiffness and braking force of the AGV according to the steering angle and speed commands to achieve safe navigation and obstacle avoidance control of the AGV in the densely shelved environment.

[0198] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0199] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Random Access Memory (RAM), Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0200] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0201] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0202] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0203] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0204] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a method for safe navigation and obstacle avoidance control of AGVs in an intelligent warehousing environment.

[0205] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0206] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0207] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0208] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for safe navigation and obstacle avoidance control of AGVs in an intelligent warehousing environment, characterized in that, include: The 3D point data of the intelligent warehousing environment where the AGV is located is collected by LiDAR, and the 3D point data is processed by a 3D filtering algorithm to remove noise and impurities and generate a 3D mesh map. Obstacle distance data between obstacles and AGVs is acquired by millimeter-wave radar. The three-dimensional mesh map and the obstacle distance data are integrated. Based on the integrated result and the position of the AGV, the spatial distribution relationship between the AGV and obstacles is updated, and a dynamic environmental state quantity with a timestamp is output. The dynamic environmental state variables are input into the path planner built based on the reinforcement learning network to determine that the intelligent warehouse environment is a dense shelving environment, and to generate the turning angle and speed commands for the AGV to pass through the dense shelving environment without collision. Based on a fuzzy PID controller, the steering stiffness and braking force of the AGV are dynamically adjusted according to the steering angle and the speed command to achieve safe navigation and obstacle avoidance control of the AGV in dense shelving environments.

2. The method according to claim 1, characterized in that, The step of inputting the dynamic environment state variables into a path planner built based on a reinforcement learning network, determining that the intelligent warehousing environment is a densely racked environment, and generating steering angle and speed commands for the AGV to pass through the densely racked environment without collision includes: The dynamic environment state is input into the path planner built based on the reinforcement learning network. The decomposition module of the path planner decomposes the dynamic environment state into the AGV's current coordinates, obstacle distribution data and timestamp information. The obstacle distribution density and channel space parameters are extracted from the obstacle distribution data using the extraction module of the path planner. The path planner's determination module identifies smart warehouse environments where the obstacle distribution density is greater than a preset density value and the channel space parameter is less than a preset width value as dense shelving environments. In the dense shelving environment, the mapping module of the path planner maps the current coordinates of the AGV and the timestamp information into multiple candidate action combinations containing turning angle and speed parameters. The path planner's selection module selects a candidate action combination that matches the obstacle distribution data from the multiple candidate action combinations, which serves as the steering angle and speed command for the AGV to pass through densely packed shelving environments without collisions.

3. The method according to claim 2, characterized in that, The step of selecting a candidate action combination that matches the obstacle distribution data from the plurality of candidate action combinations as the turning angle and speed command for the AGV to pass through densely packed shelving environments without collisions includes: Based on the obstacle distribution data, a predicted travel trajectory is determined for each candidate action combination. The predicted travel trajectory includes a series of future position points of the AGV when it executes the candidate action combination. For each predicted passage trajectory, the distance between each future position point and the nearest fixed obstacle is calculated as a fixed obstacle safety parameter; the distance between each future position point and the nearest moving obstacle is calculated as a moving obstacle safety parameter; the degree to which each future position point deviates from the center of the passage is calculated as a passage centering parameter; and the change in the turning angle of adjacent future position points is calculated as a turning smoothness parameter. The fixed obstacle safety parameters, the moving obstacle safety parameters, the passage centering parameters, and the steering smoothness parameters are integrated to generate a comprehensive evaluation value for each candidate action combination; The candidate action combination with the best comprehensive evaluation value is selected as the steering angle and speed command for the AGV to pass through densely packed shelving environments without collisions.

4. The method according to claim 1, characterized in that, The process of acquiring 3D point data of the intelligent warehousing environment where the AGV is located via LiDAR, and processing the 3D point data using a 3D filtering algorithm to remove noise and artifacts to generate a 3D mesh map includes: Remove data that exceeds a preset distance range from the three-dimensional point data, and retain the initial three-dimensional point data within the effective sensing area around the AGV; The initial three-dimensional point data is filtered to obtain target three-dimensional point data after removing noise and speculative points; The target three-dimensional point data is divided into multiple spatial grid units according to a preset size, and the spatial interval between adjacent three-dimensional points in the target three-dimensional point data is calculated to obtain three-dimensional points whose spatial interval conforms to the preset spatial interval. Each grid cell records the number of the three-dimensional points distributed within the intelligent warehousing environment; The grid state is determined based on the distribution quantity, and a three-dimensional grid map containing obstacle areas and passage areas is generated based on the grid state.

5. The method according to claim 1, characterized in that, The process involves acquiring obstacle distance data between obstacles and the AGV using millimeter-wave radar, integrating the 3D mesh map and the obstacle distance data, updating the spatial distribution relationship between the AGV and obstacles based on the integrated result and the AGV's position, and outputting a timestamped dynamic environmental state quantity, including: Millimeter-wave radar is used to detect fixed and moving obstacles around the AGV, and the straight-line distance between each obstacle and the AGV is recorded as obstacle distance data; Based on the obstacle distance data, the positions of each obstacle are marked in the corresponding grid cells of the three-dimensional grid map, so that the corresponding grid cells simultaneously contain three-dimensional point distribution information and obstacle distance information; Based on the position of the AGV, calculate the orientation and spacing of obstacles in the corresponding grid cell relative to the AGV; Based on the orientation and the spacing, adjust the three-dimensional point distribution information and obstacle distance information of each obstacle to generate an updated spatial distribution relationship between the AGV and the obstacles; Add corresponding time information to the spatial distribution relationship and output dynamic environmental state quantities with timestamps.

6. The method according to claim 5, characterized in that, The step of adjusting the three-dimensional point distribution information and obstacle distance information of each obstacle according to the orientation and the spacing to generate an updated spatial distribution relationship between the AGV and the obstacles includes: The category to which the obstacle belongs is determined to distinguish between fixed obstacles and moving obstacles; For each of the fixed obstacles, when the position corresponding to the three-dimensional point distribution information of the obstacle is inconsistent with the position derived based on the orientation and the spacing, the coordinate data in the three-dimensional point distribution information of the obstacle is corrected to obtain the corrected three-dimensional point distribution information of the obstacle. For each of the moving obstacles, when there is a difference between the obstacle distance information and the spacing, the obstacle distance information is updated according to the spacing to obtain the updated obstacle distance information; Based on the corrected 3D point distribution information of the obstacles and the updated obstacle distance information, combined with the orientation of the fixed obstacles and the moving obstacles, the position parameters of each obstacle in 3D space are determined. Based on the position parameters, the relative positional relationships between the obstacles are determined; By integrating the relative positional relationship with the current coordinates of the AGV, an updated spatial distribution relationship between the AGV and obstacles is generated.

7. The method according to claim 1, characterized in that, The fuzzy PID controller dynamically adjusts the steering stiffness and braking force of the AGV according to the steering angle and speed command to achieve safe navigation and obstacle avoidance control of the AGV in densely packed shelving environments, including: Extract the actual steering deviation corresponding to the steering angle and the actual speed deviation corresponding to the speed command; Based on the actual steering deviation and the actual speed deviation, and combined with the aisle width characteristics in a densely packed shelving environment, the adjustment ranges corresponding to the steering stiffness and braking force of the AGV are determined. Based on the adjustment range, calculate the first change in steering stiffness and the second change in braking force; The first change is input to the steering component of the AGV to obtain steering operation parameters, and the second change is input to the braking component of the AGV to obtain braking operation parameters. The real-time distance between obstacles and the AGV in the dynamic environment state is obtained. Based on the steering operation parameters and the braking operation parameters, and combined with the real-time distance, the adjustment range is corrected to achieve safe navigation and obstacle avoidance control of the AGV in densely packed shelving environments.

8. A safe navigation and obstacle avoidance system for AGVs in an intelligent warehousing environment, characterized in that, include: The first generation module is used to collect three-dimensional point data of the intelligent warehousing environment where the AGV is located through LiDAR, and process the three-dimensional point data using a three-dimensional filtering algorithm to remove noise and impurities and generate a three-dimensional mesh map. The integration module is used to acquire obstacle distance data between obstacles and AGV through millimeter-wave radar, integrate the three-dimensional mesh map and the obstacle distance data, and based on the integrated result, combined with the position of AGV, update the spatial distribution relationship between AGV and obstacles, and output dynamic environmental state quantities with timestamps. The second generation module is used to input the dynamic environment state quantity into the path planner built based on the reinforcement learning network, determine that the intelligent warehouse environment is a dense shelf environment, and generate the turning angle and speed command for the AGV to pass through the dense shelf environment without collision. The adjustment module is used to dynamically adjust the steering stiffness and braking force of the AGV based on the fuzzy PID controller and the steering angle and speed command, so as to achieve safe navigation and obstacle avoidance control of the AGV in dense shelving environments.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the AGV safety navigation and obstacle avoidance control method based on any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements the AGV safety navigation and obstacle avoidance control method as described in any one of claims 1 to 7 in an intelligent warehousing environment.

Images