Method and system for robot to autonomously recognize grain piles in warehouse

Abstract

The invention provides a method and system for a robot to autonomously recognize grain piles in a warehouse, and belongs to the technical field of intelligent agricultural machinery and mobile robots, the robot comprises a wheel type robot, a double-foot robot or other robots capable of advancing on the grain surface, and grain pile recognition is achieved only through a sensing system carried by the robot. The method does not need to arrange any electronic equipment in a warehouse in advance, and comprises the following steps: acquiring point cloud data of an environment in front of the robot under all poses, and preprocessing the point cloud data; utilizing spatial proximity to separate potential grain pile point cloud data through clustering to obtain candidate point cloud clusters; performing geometric dimension screening and grain pile natural accumulation rule screening on the candidate point cloud clusters to obtain effective grain piles; and according to the coordinate of the effective grain pile vertex in the point cloud coordinate system and the coordinate and the attitude angle of the robot in the warehouse coordinate system, calculating the coordinate of the effective grain pile vertex in the warehouse coordinate system. According to the invention, grain piles in the warehouse can be automatically identified.

Description

Technical Field

[0001] This invention belongs to the field of intelligent agricultural machinery and mobile robot technology, and specifically refers to a method and system for a robot to autonomously identify grain piles in a warehouse. Background Technology

[0002] When grain is received and stored in grain storage departments, grain conveyors are usually used to pour the grain into the grain silos from a height. Because the silos are large, manual movement of the conveyors is required to move the grain into different locations within the silo. This results in a large number of uneven piles and grains inside the silo. Figure 1 As shown. Large amounts of grain piles and grains seriously affect intelligent warehousing operations such as temperature measurement, ventilation, fumigation, and storage measurement after entering the warehouse. At present, the leveling operation (i.e., leveling the grain pile and arranging the grain in the warehouse into a plane) in the grain storage process in my country is generally carried out manually, which has the following serious problems: (1) Low efficiency: Leveling a single standard grain warehouse requires 15 to 20 man-days of work, which is time-consuming and labor-intensive, and there is a shortage of labor; (2) Health and safety hazards: Operators need to work in a high dust environment for a long time, which can easily cause occupational health problems. There are also safety hazards such as slipping and falling into the grain surface and suffocation when working in the grain pile. Summary of the Invention

[0003] To address the technical problems existing in the prior art, the present invention provides a method and system for a robot to autonomously identify grain piles in a warehouse, the technical solution of which is as follows: On the one hand, a method for autonomously identifying grain piles in a warehouse is provided. The robot includes wheeled, bipedal, or other robots capable of traversing the grain surface. It identifies grain piles solely through its onboard sensing system, without requiring any pre-installed electronic equipment within the warehouse. The method includes: S31. Obtain point cloud data of the environment in front of the robot under all poses, and preprocess the point cloud data. S32. Utilize spatial proximity to separate potential grain pile point cloud data through clustering to obtain candidate point cloud clusters; S33. Perform geometric size screening and natural grain pile accumulation law screening on the candidate point cloud clusters to obtain effective grain piles; S34. Based on the coordinates of the effective grain pile vertex in the point cloud coordinate system, and the coordinates and attitude angle of the robot in the warehouse coordinate system, calculate the coordinates of the effective grain pile vertex in the warehouse coordinate system.

[0004] On the other hand, a system for autonomously identifying grain piles in a warehouse is provided, the system comprising: The preprocessing module is used to acquire point cloud data of the environment in front of the robot under all poses and to preprocess the point cloud data. A clustering separation module is configured to separate potential grain pile point cloud data by clustering to obtain candidate point cloud clusters by using spatial proximity; An effective grain pile screening module is configured to perform geometric size screening and grain pile natural accumulation rule screening on the candidate point cloud clusters respectively to obtain effective grain piles. A warehouse coordinate solving module is configured to solve coordinates of the effective grain pile vertices in a warehouse coordinate system according to coordinates of the effective grain pile vertices in a point cloud coordinate system and coordinates and attitude angles of the robot in the warehouse coordinate system.

[0005] In another aspect, an electronic device is provided, which includes a processor and a memory having at least one instruction stored therein, the at least one instruction being loaded and executed by the processor to implement the above-mentioned method for a robot to autonomously identify grain piles in a warehouse.

[0006] In another aspect, a computer-readable storage medium is provided, which has at least one instruction stored therein, the at least one instruction being loaded and executed by a processor to implement the above-mentioned method for a robot to autonomously identify grain piles in a warehouse.

[0007] The technical solution provided by the present application has at least the following beneficial effects: The embodiment of the present application can accurately extract three-dimensional coordinates of grain pile vertices: directly positioning the highest point of each grain pile; not relying on external equipment: only the perception system carried by itself is needed, without a warehouse top radar or camera; based on the natural form of the grain pile and the point cloud rule: using the physical properties of the grain pile formed by the internal friction coefficient of the grain, the identification accuracy is high; strong robustness, suitable for complex environment: through spatial features, geometric features and natural accumulation rules of grain piles, non-grain pile interference is excluded, and the warehouse environment with multiple grain piles and chaotic distribution is adapted; high calculation efficiency, easy to deploy: not relying on deep learning models, real-time, suitable for embedded or vehicle-mounted computing platforms. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows.

[0009] Figure 1 is a schematic diagram of a grain pile in a warehouse provided by the embodiment of the present application; Figure 2 is a structure block diagram of a warehouse flattening robot provided by the embodiment of the present application; Figure 3 is a side view of a warehouse flattening robot provided by the embodiment of the present application; Figure 4 is a front view of a warehouse flattening robot provided by the embodiment of the present application; Figure 5It is a structural schematic diagram of a pipe auger grain transfer system provided by an embodiment of the present application. Figure 6 It is a method flowchart for autonomous flat storage operation using a flat storage robot trolley provided by an embodiment of the present application. Figure 7 It is a warehouse coordinate system schematic diagram provided by an embodiment of the present application. Figure 8 It is a robot trolley point cloud coordinate system schematic diagram provided by an embodiment of the present application. Figure 9 It is a posture angle schematic diagram provided by an embodiment of the present application. Figure 10 It is a yaw angle schematic diagram provided by an embodiment of the present application. Figure 11 It is a pitch angle schematic diagram provided by an embodiment of the present application. Figure 12 It is a roll angle schematic diagram provided by an embodiment of the present application. Figure 13 It is an initial placement schematic diagram of a robot trolley provided by an embodiment of the present application. Figure 14 It is a method flowchart for robot autonomous positioning based on environment cognition provided by an embodiment of the present application. Figure 15 It is a plane schematic diagram obtained by point cloud solving provided by an embodiment of the present application. Figure 16 It is a method flowchart for robot autonomous identification of grain piles in a warehouse provided by an embodiment of the present application. Figure 17 It is a grain pile cross-section schematic diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0010] To make the technical problems, technical solutions and advantages of the present application clearer, specific embodiments will be described in detail below with reference to the accompanying drawings.

[0011] As shown in Figures 2 to 4 An embodiment of the present application provides a flat storage robot trolley, which comprises a spiral wheel driving chassis (1), a pipe auger grain transfer system (2) and a vehicle body (3). The spiral wheel driving chassis (1) is located at the bottom of the flat storage robot trolley and is used for stable travel on loose grain surface. The pipe auger grain transfer system (2) is located at the vehicle body (3) and is used for efficient grain transfer.

[0012] Optionally, the spiral wheel drive chassis (1) is symmetrically equipped with spiral wheels (such as Archimedes spiral wheels or other spiral wheels) on both sides of the chassis to provide spiral propulsion force, prevent the flattening robot trolley from sinking when traveling on the loose grain surface, and adapt to load-bearing and uphill / downhill operations.

[0013] Optionally, such as Figure 5 As shown, the tubular auger grain transfer system (2) is integrated and installed on the vehicle body (3), including: a conical head, a spiral conveying pipe and a scattering outlet; The conical head is located at the head of the trolley and is used to insert into the high-level grain pile; The spiral conveying pipe is connected to the conical head and the scattering outlet, and is used to bring the grain from the grain pile into the pipe; The scattering outlet is located at the rear of the trolley and is used to scatter the grain brought into the pipe to a low area at the rear of the trolley.

[0014] Optionally, the helical wheel drive chassis (1) is equipped with a bracket, which connects helical wheels symmetrically installed on both sides; The tubular auger grain transfer system (2) passes through the vehicle body (3), and the two are fixed as a whole (for example, by bolts). The vehicle body (3) together with the tubular auger grain transfer system (2) can rotate vertically about the bracket as an axis to insert the conical head into the high-level grain pile; or the tubular auger grain transfer system (2) and the vehicle body (3) are connected (for example, by bearings), but the tubular auger grain transfer system (2) can rotate independently to insert the conical head into the high-level grain pile; The pipe is equipped with a rotating auger inside, and the front of the pipe is open. As the rotating auger rotates, the grain in the grain pile is brought into the pipe from the opening.

[0015] Optionally, one end of the rotating auger is fixed to the conical head, and the other end is connected to a motor and driven by the motor.

[0016] The embodiments of the present invention utilize a grain leveling robot trolley equipped with a tubular auger grain transfer system, which can automatically and efficiently transfer grain, significantly improving the efficiency of grain leveling operations.

[0017] Optionally, the flattening robot trolley equipped with the tubular auger grain transfer system in this embodiment of the invention can obtain its own location information and identify grain pile information by installing electronic devices such as lidar on the top of the warehouse, and then use the tubular auger grain transfer system to transfer grain efficiently.

[0018] Optionally, the flattening robot trolley equipped with the tubular auger grain transfer system in this embodiment of the invention can also use its own sensing system to locate and obtain its own position information and then identify the grain pile information, and then use the tubular auger grain transfer system to transfer the grain efficiently, without having to pre-deploy any electronic equipment in the warehouse.

[0019] Optionally, the flattening robot is also equipped with a perception system, which includes a point cloud acquisition device and a three-dimensional posture acquisition device. The point cloud acquisition device (such as a binocular camera or 3D radar) is used to acquire point cloud data in front of the vehicle. The three-dimensional attitude acquisition device (e.g., an IMU or a gyroscope) is used to acquire the three-dimensional attitude data of the vehicle. Optionally, the point cloud acquisition device and the three-dimensional attitude acquisition device are integrated together; or they are installed separately (e.g., they can be integrated at the front or rear of the vehicle, or at a position higher than the vehicle body, so that the point cloud acquisition device can acquire point cloud data of the environment in front of the vehicle; or they can be installed separately, with the point cloud acquisition device installed at the front or rear of the vehicle, or at any position on the vehicle).

[0020] Optionally, the flattening robot car obtains its own position information and information of each grain pile based on the point cloud data and three-dimensional posture data, and then determines the flattening operation sequence of the grain piles, and moves to the grain piles in sequence to perform the flattening operation (for details, please refer to the following content on the method of using a flattening robot car for autonomous flattening operation, the robot autonomous localization method based on environmental cognition, and the robot autonomous identification method of grain piles in the warehouse).

[0021] Optionally, the flattening robot trolley also includes a scraper plate located at the rear of the trolley. After the robot trolley performs high-level transfer of the large grain pile through the tubular auger grain transfer system (2), the scraper plate performs a leveling operation to ensure the flatness of the grain surface.

[0022] Optionally, the scraper plate is installed at the rear of the vehicle body and its lifting is controlled by an electric push rod to scrape the grain surface behind the vehicle body flat when the vehicle is moving.

[0023] like Figure 6 As shown in the figure, this embodiment of the invention also provides a method for autonomous closing operations using a closing robot vehicle, the method comprising: S1. Place the flattening robot trolley on the grain surface inside the warehouse and initialize its state; S2. The flattening robot car autonomously travels to the top area of ​​the grain pile and performs autonomous positioning to determine its location information in the warehouse space. S3. After obtaining location information, the flattening robot car identifies the surrounding grain piles and establishes a flattening operation sequence based on the identified grain pile information. S4. The flattening robot trolley performs the efficient transfer of high-level grain piles in an orderly manner according to the order in the flattening operation sequence, and dynamically updates the flattening operation sequence. S5. When the leveling operation sequence is empty, the leveling robot trolley performs the task of leveling the grain surface in the entire warehouse.

[0024] Let's first introduce some basic concepts: In the method of this embodiment of the invention, the warehouse coordinate system and robot car coordinate system, and the definitions of attitude angles (yaw angle, roll angle, and pitch angle) are explained as follows: (1) Warehouse coordinate system For ease of subsequent description, it is assumed that the warehouse is a south-facing building (the warehouse in this embodiment of the invention can also be a square warehouse with other orientations, or a circular warehouse, without limitation, all of which are within the protection scope of this invention), 50 meters long from east to west and 25 meters wide from north to south. The longitudinal wall (east-west direction) of the warehouse is the X-axis, the transverse wall (north-south direction) is the Z-coordinate, the height of the warehouse is the Y-coordinate, and the origin of the warehouse coordinate system is the southwest corner of the warehouse. Figure 7 As shown, the coordinates of the i-th point in the warehouse coordinate system are represented as follows: ; (2) Point cloud coordinate system of robot car The robotic vehicle is equipped with a point cloud acquisition device (such as a binocular camera or 3D radar), and the coordinate system of the acquired point cloud is as follows: Figure 8 As shown in the figure, the Z_m axis represents the direction of the robot's movement, the Y_m axis represents the robot's height, and the X_m axis represents the robot's horizontal direction. The plane formed by the X_m axis and the Z_m axis is the horizontal plane of the robot's body. The coordinates of the i-th point in the point cloud coordinate system are represented as... .

[0025] (3) Attitude angle: The robot car is equipped with a 3D attitude acquisition device (such as an IMU or gyroscope). The robot car is then placed in the storage space, as shown in the reference. Figure 9 The attitude angles of the robot car at this time are explained as follows: 1) Yaw angle θ y It is the angle between the projection of the Z_m axis (forward direction) in the robot's coordinate system onto the XZ plane (ground) in the warehouse coordinate system and the Z axis of the warehouse coordinate system.

[0026] If the robot car moves entirely along the positive Z-axis of the warehouse space, then the yaw angle θ y It is 0°.

[0027] The yaw angle value is set as follows: Figure 10 As shown.

[0028] 2) Pitch angle θ p It is the angle between the projection of the Z_m axis (forward direction) in the robot's coordinate system onto the YZ plane (width-side wall) in the warehouse coordinate system and the Z axis of the warehouse coordinate system.

[0029] If the robot car moves along the horizontal ground of the warehouse space, then the pitch angle θ p It is 0°.

[0030] Pitch angle θ p The value setting is as follows Figure 11 As shown.

[0031] 2) Roll angle θ r It is the angle between the projection of the X_m axis (horizontal direction of the robot car) in the robot car coordinate system onto the YX plane (longitudinal wall) in the warehouse coordinate system and the X axis of the warehouse coordinate system.

[0032] If the robot car moves along the horizontal ground of the warehouse space, then the roll angle θ r It is 0°.

[0033] Roll angle θ r The value setting is as follows Figure 12 As shown.

[0034] Optionally, S1 specifically includes: Based on the characteristics of warehouse construction, set the length L and width W of the warehouse, and input the leveling line height H. Ref The window size and height (length a × width b, H_win) are set; this setting yields the coordinates of the four walls, windows, and the flattening line in the warehouse coordinate system. Set the maximum slope θ of the grain pile according to the grain variety. P ; By combining storage information and the bulk density of this grain variety, the height of the grain after leveling in the warehouse can be estimated, denoted as H. Final ; The robotic cart enters through the grain condition inspection gate and is placed on the grain pile slope from the grain condition inspection platform (for older grain silos without grain condition inspection gates, the robotic cart can also be placed on the grain pile slope through a window via a lifting platform). Specifically, it can be placed on the grain pile slope with its forward direction perpendicular to the longitudinal wall of the silo (other directions are also possible; this embodiment of the invention does not limit this and all are within the protection scope of this embodiment). The initialization state is then established, and the cart's attitude θ, including the pitch angle θ, is obtained. p Roll angle θ r yaw angle θy And initialize the yaw angle to 0°, see Figure 13 .

[0035] Simultaneously, the robot car activates its point cloud acquisition device to obtain point cloud data of the surrounding environment. The point cloud data read by the robot car is represented as follows:

[0036] Where: N is the total number of points in the point cloud, and the subscript i represents the i-th three-dimensional spatial point; Optionally, S2 specifically includes: The leveling robot trolley goes uphill to the top of the grain pile for autonomous positioning. If the autonomous positioning fails (possibly because the grain pile is too low and the field of vision is limited), it goes downhill and uphill to the top of another grain pile (the other grain pile is higher), climbs to the top, and tries to position itself again. If the autonomous positioning fails, it goes downhill and uphill again to the top of another grain pile until it successfully positions itself and obtains its location information.

[0037] Optionally, the leveling robot trolley ascends the slope to the top area of ​​the grain pile, specifically including: Read the roll angle θ of the car r and pitch angle θ p By controlling steering and movement, the vehicle's posture is made to meet the following requirements: Roll angle | θ r |<θ R To avoid deviating from the top area of ​​the grain pile in the direction of travel; Pitch angle θ p >θ P Keep the trolley on the effective slope and facing the top area of ​​the grain pile; Where θ R θ P The slope-related threshold is set based on the characteristics of the grain pile for specific grain varieties; At this point, it is determined that the trolley is on an effective slope and its direction of travel is towards the top of the grain pile, maintaining a yaw angle θ. y Unchanged, roll angle θ r Unchanged, continuing uphill; As you approach the top of the grain pile, the pitch angle θ p It will gradually decrease, maintaining the direction of the car's movement until the pitch angle θ is reached. p Approximately 0°, at which point it is determined that the trolley has reached the top area of ​​the grain pile and should stop moving forward.

[0038] Optionally, the process of descending and ascending from one grain pile to the top of another grain pile specifically includes: Downhill control: Determine the roll angle θ of the car while it is moving.r and pitch angle θ p By controlling steering and movement, the vehicle's posture is made to meet the following requirements: Roll angle | θ r |< θ R And the pitch angle θ p <-θ min ; Where θ R θ min The slope-related threshold is set based on the characteristics of the grain pile of different grain varieties; At this point, it is determined that the trolley is on an effective slope and its direction of travel is towards the bottom of the grain pile, maintaining a yaw angle θ. y Unchanged, roll angle θ r Unchanged, continuing downhill; Turning uphill: As the trolley continues downhill, upon approaching the bottom of the grain pile, the grain pile in the trolley's direction of travel will be identified as an obstacle, causing the trolley to steer to avoid it. The trolley then adjusts its attitude to achieve a pitch angle θ. p Gradually increase to θ p >0, at this point the uphill climb begins.

[0039] Optionally, the identification and avoidance of the obstacle specifically includes: The obstacle space R in front is defined using the coordinate system of the 3D point cloud data acquired by the flattening robot. o The statistics show that point cloud data acquired by the point cloud acquisition device falls into R. o If the number of points exceeds the preset threshold T o If so, it is determined that there is an obstacle ahead; At this point, the car steers to avoid the obstacle.

[0040] For example, in this embodiment of the invention, the coordinate system of the 3D point cloud data acquired by the robot car can be used as a reference, and a cuboid region with Z-axis (forward direction) of 0.3~1m, X-axis (lateral direction) of 0±1.2m, and Y-axis (vertical direction) of 0~0.5m can be defined as the obstacle space R in front. o Then determine whether each point in the point cloud data acquired by the robot car falls into the obstacle space R. o The method is as follows:

[0041] Count all those falling into R o If the number of points exceeds the preset threshold T o If an obstacle is detected, the car will steer to avoid it and continue its previous journey.

[0042] When the vehicle reverses or turns left or right, the microwave radar at the rear detects nearby obstacles within a range of 0.3 to 1.5 meters behind and to the sides. If an obstacle is encountered, the vehicle adjusts its direction of travel and maneuvers around it until the obstacle is cleared.

[0043] Optionally, step S3 involves establishing a leveling operation sequence based on the identified grain pile information, specifically including: The height of each identified grain pile and the height H of the grain pile after leveling will be determined. Final The H was compared. Final Based on storage information and the bulk density of the grain variety, it is estimated that if the height of a grain pile is higher than H... Final The information of this grain pile is compared with the grain pile information already recorded in the grain pile operation task queue J. If the information of this grain pile is not in the grain pile operation task queue J, then the information of this grain pile is added to the grain pile operation task queue J. Sort the height of the grain pile vertices in the grain pile operation task queue J to generate a "high first, low last" leveling operation sequence.

[0044] Optionally, S4 specifically includes: The grain leveling robot trolley first moves to the top of the highest grain pile in the leveling operation sequence. It then adjusts the conical head of its own tubular auger grain transfer system through motor drive, inserting it into the grain pile. The motor is then started, and the grain flows from the spiral conveying pipe of the tubular auger grain transfer system to the rear of the trolley and is scattered to a lower area below the grain pile. The leveling robot trolley continuously monitors whether grain has entered the spiral conveyor pipe. If no grain has entered, the leveling robot trolley autonomously repositions itself and obtains its own position information. If the height of this position is still higher than H... Final The flattening robot trolley retreats slightly, then re-inserts its conical head into the grain pile to transfer the grain until the height of its position is close to H. Final At that time, the trolley retracts its conical head, completing the leveling operation of this grain pile and clearing the information of this grain pile in J. Then, the leveling robot trolley proceeds sequentially to the top areas of other grain piles in the leveling operation sequence to perform leveling operations on these grain piles until the leveling operation sequence is empty.

[0045] Optionally, S5 specifically includes: After the robotic vehicle completes its efficient transfer task, all the grain piles above the leveling line in the warehouse have been moved to the grain level. However, many grain surfaces still protrude. The vehicle then repositions itself and moves to the southwest corner of the warehouse (X=0, Z=0), adjusting its yaw angle θ. yThe angle is 0°. At this point, the scraper plate is lowered to the grain surface, and the vehicle continues forward until the warehouse wall is detected by point cloud data. Then, the trolley performs a U-turn, maintaining the distance of the original straight path after the U-turn as the width of the scraper plate. After the turn, θ... y To 180°, continue the operation and repeat the step until the trolley scrapes the grain surface of the entire warehouse.

[0046] While the trolley is performing the leveling task, the three-dimensional coordinates of the grain surface can be recorded every 100 milliseconds. After the first round of leveling of the entire grain silo, a three-dimensional height map of the grain surface is generated. Areas above the leveling line are located in this height map. When the trolley reaches these areas, the leveling task is performed again, and the three-dimensional height map of the grain surface is updated, until the final grain surface height matches H. Final Keep it level (error 0.1m).

[0047] The autonomous positioning in S2 and the grain pile identification in S3 of the method for autonomous grain leveling using a grain leveling robot provided in this embodiment of the invention can be performed using existing methods (such as using a lidar installed on the top of the warehouse), or using the robot autonomous positioning method based on environmental cognition and the robot autonomous identification method for grain piles in the warehouse provided in this embodiment of the invention. The robot includes wheeled, bipedal, or other robots that can travel on the grain surface. It achieves autonomous positioning solely through its own onboard sensing system, without the need to pre-deploy any electronic devices in the warehouse. This embodiment of the invention does not limit the methods of autonomous positioning and grain pile identification, and all of them are within the protection scope of this embodiment of the invention.

[0048] In this embodiment of the invention, a leveling robot trolley autonomously travels to the top area of ​​the grain pile, performs autonomous positioning, identifies information about the surrounding grain piles, establishes a leveling operation sequence based on the identified information, and performs the efficient transfer of high-level grain piles in an orderly manner according to the order in the leveling operation sequence. Then, the grain surface is leveled for the entire warehouse. This not only enables orderly operation and efficient transfer of high-level grain, but also allows for large-scale deployment, achieving truly autonomous and efficient intelligent leveling operations.

[0049] More specifically, the robot car obtains three-dimensional posture data through its own three-dimensional posture acquisition device, guiding the car to autonomously climb from its initial position within the warehouse spatial coordinate system. After regaining its field of vision, it performs autonomous positioning (if positioning fails due to a low grain pile, it goes downhill and then uphill to the top of a higher grain pile to try to regain its field of vision, and then completes its self-positioning after the field of vision is restored), and then obtains information about the grain pile, and performs leveling operations based on the information about the grain pile.

[0050] like Figure 14As shown, this embodiment of the invention also provides a robot autonomous localization method based on environmental cognition. The robot includes wheeled, bipedal, or other robots capable of traversing grain surfaces (the robot can be the aforementioned leveling robot vehicle, or other robots). It achieves autonomous localization solely through its onboard sensing system, without requiring any pre-deployed electronic devices within the warehouse. The method includes: S21. When the robot is located at the top of the grain pile, it acquires point cloud data of the environment in front of the robot in a certain pose through its own point cloud acquisition device. Based on the point cloud data, it fits multiple candidate planes and selects the effective candidate planes. S22. Optimize the effective candidate planes and transform the poses to obtain a set of all effective candidate planes under all poses; S23. Transform the set of all valid candidate planes under all poses into the warehouse coordinate system, and merge the same planes; S24. Identify the plane in the warehouse spatial coordinate system to obtain two mutually perpendicular walls; S25. Based on the identified positional relationship between the wall and the robot, calculate the horizontal x and z coordinates of the robot; S26. Using the windows or leveling lines of the warehouse, calculate the vertical Y coordinate of the robot.

[0051] This invention aims to provide a method for autonomous robot localization based on the structural characteristics of a warehouse and its grain storage features. It integrates 3D posture data acquired by a 3D posture acquisition device and point cloud data acquired by a point cloud acquisition device, addressing the problems of heavy reliance on auxiliary positioning equipment and inaccurate localization in existing technologies. Specific objectives include: achieving autonomous localization without the need for external auxiliary equipment, reducing deployment costs and maintenance complexity; and constructing a scenario for autonomous robot localization within a warehouse. The core idea is to abandon reliance on top radar or auxiliary positioning systems, instead utilizing known structural features of the warehouse (length, width, leveling line, windows) as a reference. Under visible conditions, the method uses 3D posture data acquired by a 3D posture acquisition device and point cloud data acquired by a point cloud acquisition device to detect and fit stable structural planes in the environment, thereby identifying the warehouse walls. Then, combining the warehouse's length and width, leveling line, or windows, and through conversion between the point cloud coordinate system and the warehouse coordinate system, the robot's position coordinates within the warehouse space are calculated.

[0052] Optionally, S21 specifically includes: When the robot is located in the top area of ​​the grain pile (the robot reads the pitch angle θ) p Roll angle θ R Yaw angle θ y And by controlling its own movement, the pitch angle θ is achieved. p<θ L θ L A preset threshold close to 0 is used to keep the robot as horizontal as possible in the top area. Using its onboard point cloud acquisition device, the robot acquires point cloud data of the environment in front of it in a given pose. Based on this point cloud data, multiple candidate planes are fitted, and valid candidate planes are selected, including: S21-1, Planar Model Generation and Interior Point Statistics The random sampling consensus algorithm is used, and multiple iterations are performed. In each iteration k: From the acquired point cloud data By randomly selecting three non-collinear points, an initial planar model is obtained. Its equation is: ; Its normal vector It is a unit vector; Calculate each point in the point cloud To the initial planar model distance :

[0053] Statistical analysis of the initial planar model interior point set and their quantity :

[0054] in This is a preset distance threshold; S21-2, Multi-plane selection strategy After a predetermined number of iterations, select candidate planes from all fitted planes by the number of interior points. Descending order filtering: For each candidate plane, determine whether it satisfies the following:

[0055] in Let be the threshold for the number of interior points. All planes that meet the condition are retained as valid candidate planes, denoted as set . ,in The total number of valid candidate planes; For each valid candidate plane Record its plane equation parameters Interior point set Unit normal vector The robot's yaw angle θ y , the θ yThe attitude is acquired through the robot's own onboard 3D attitude acquisition device.

[0056] Optionally, the optimization of the effective candidate plane in S22 specifically includes: S22-1. Perform density verification on each valid candidate plane to eliminate false planes caused by sparse point clouds or noise, including: For valid candidate planes and its interior point set : (1) Calculate the area of ​​the projected bounding box: interior point set Projecting the points onto its own planar model forms a two-dimensional point set. Calculate the area of ​​the smallest bounding rectangle of this two-dimensional point set. as follows:

[0057] in , and These are the minimum and maximum values ​​of the coordinates of the projection point in the local coordinate system of the plane, respectively. (2) Calculate the point cloud surface density: Calculate the effective candidate plane Point cloud surface density as follows:

[0058] in It is an interior point set The number of points; (3) Filtering by density threshold: If the effective candidate plane surface density satisfy , If the surface density is within the preset threshold, it will be retained; otherwise, it will be discarded as a false detection plane. After density filtering, a set of candidate planes is obtained.

[0059] Optionally, optimizing the effective candidate plane in step S22 further includes: S22-2, For the candidate plane set The planes in the data are merged to resolve the issue of the same physical plane being fragmented into multiple discontinuous segments due to occlusion, including: S22-2-1. Detect whether the planes belong to the same plane. Iteratively detect the candidate plane set Any two plane unit normal vectors and The included angle as follows:

[0060] If the two planes satisfy If they belong to the same physical plane, then they are determined to be in the same physical plane. The preset merging angle threshold; S22-2-2, Merging Planes Plane pairs that meet the merging conditions Merge the interior point sets: ; Use the merged interior point set A new planar model is refitted using the least squares method. ; from Remove from and and add new merge planes ; S22-2-3, Final Plane Set Output After the density filtering and merging processes described above, the optimized set of effective candidate planes is obtained:

[0061] in ; for For each plane in the equation, record its final plane equation parameters, set of interior points, and unit normal vector.

[0062] Optionally, the pose transformation in step S22 to obtain the set of all valid candidate planes under all poses specifically includes: The robot rotates horizontally (for example, it can be controlled to rotate horizontally in fixed angle steps Δθ (e.g., 30°) to obtain a set of valid candidate planes within the field of view of other poses, and records each pose. The set of all valid candidate planes under )

[0063] If the robot still cannot obtain a set of valid candidate planes after performing a 360° horizontal rotation, it means that the robot is currently in a low position (the view is blocked and the warehouse walls cannot be seen). It needs to continue to climb higher to expand its field of vision. At this time, the robot goes downhill from this grain pile and then uphill to another higher grain pile, and tries again until it obtains a valid candidate plane.

[0064] Optionally, S23 specifically includes: For each yaw angle θ y The following records the valid candidate planes The rotation matrix corresponding to the warehouse coordinate system is:

[0065] Transform the set of valid candidate planes based on the robot's point cloud coordinate system to the warehouse coordinate system:

[0066] Since the car may scan the same wall in different poses, the same plane is merged: Check whether any two valid candidate planes satisfy the following condition: normal vector similarity: If the condition is met, the two planes are merged into one plane, resulting in a set of merged planes: .

[0067] The planes in the plane set obtained by point cloud solving through the above steps are as follows: Figure 15 As shown.

[0068] Optionally, S24 specifically includes: Assuming the warehouse is oriented north-south, with east-west as its length and north-south as its width, and based on the warehouse coordinate system and the robot's yaw angle for entering the warehouse, the normal vectors of the four wall planes are as follows: East wall, normal vector approximately... The west wall has a normal vector close to... North wall, normal vector close to The south wall has a normal vector close to... (This is an assumption made for ease of description only, and does not limit the information of the warehouse, the warehouse coordinate system, the normal vector, etc., all of which are within the protection scope of this invention.) By comparing the normal vectors of the planes in the obtained warehouse space coordinate system with the normal vectors of the four wall planes, two mutually perpendicular wall surfaces are obtained.

[0069] If at least two mutually perpendicular wall planes cannot be found at this point, it is determined that the robot's field of vision is limited here, and it needs to go down the slope again and climb up to another, higher grain pile to try.

[0070] Optionally, S25 specifically includes: Assuming the two perpendicular walls are the east wall and the north wall, the horizontal x and z coordinates of the robot are calculated based on the identified relationship between the walls and the robot's position as follows: ; .

[0071] Other wall combinations can be solved similarly.

[0072] Optionally, S26 specifically includes: Using the warehouse window: The robot turns to face the wall with windows (such as ventilation windows installed on the long wall of a warehouse) and identifies the window's location (for example, by using RGB information (two-dimensional image information) and depth information from a binocular depth camera). Given a fixed height H_win of the bottom edge of the window from the ground, combined with the y-coordinate of the bottom edge of the window detected in the point cloud coordinate system... c Using the coordinates, calculate the height Y of the car above the ground: Y = H_win –y c The height is the vertical Y-coordinate of the robot; Alternatively, you can use the liquidation line in the warehouse: Identify the red leveling line painted on the warehouse wall (e.g., by using RGB information (two-dimensional image information) from a binocular depth camera), given the height H. Ref ; By reading the y-coordinate of the liquidation line from the point cloud data, the absolute height Y = H of the car above the ground can be calculated. Ref -y, where the height is the robot's vertical Y-coordinate.

[0073] This invention provides a robot autonomous localization method based on environmental cognition, which does not rely on external positioning equipment: it does not require top radar, UWB, GPS, etc., and is suitable for signal-constrained warehouse environments; it utilizes prior structural knowledge: based on the known length, width, leveling lines, windows, and other features of the warehouse, it achieves low-cost and practical positioning; it integrates data from its own onboard multi-sensor system: an IMU or gyroscope provides three-dimensional attitude, and a binocular depth camera or 3D radar provides point clouds and images, enabling geometric reasoning and visual localization; it is robust and adaptable to complex grain piles and obstacle environments; it is scalable and reusable: suitable for robot autonomous localization tasks in structured enclosed environments such as grain silos, warehouses, and indoor factories.

[0074] like Figure 16 As shown, this embodiment of the invention also provides a method for a robot to autonomously identify grain piles in a warehouse. The robot includes wheeled, bipedal, or other robots capable of traversing the grain surface (the robot can be the aforementioned leveling robot vehicle, or other robots). It identifies grain piles solely through its onboard sensing system, without requiring any pre-installed electronic equipment within the warehouse. The method includes: S31. Obtain point cloud data of the environment in front of the robot under all poses, and preprocess the point cloud data. S32. Utilize spatial proximity to separate potential grain pile point cloud data through clustering to obtain candidate point cloud clusters; S33. Perform geometric size screening and natural grain pile accumulation law screening on the candidate point cloud clusters to obtain effective grain piles; S34. Based on the coordinates of the effective grain pile vertex in the point cloud coordinate system, and the coordinates and attitude angle of the robot in the warehouse coordinate system, calculate the coordinates of the effective grain pile vertex in the warehouse coordinate system.

[0075] Optionally, obtaining point cloud data of the robot's front environment in all poses in step S31 specifically includes: When the robot is located at the top of the grain pile, it obtains the point cloud data of the environment in front of the robot in its current pose through its onboard point cloud acquisition device. The robot rotates horizontally to obtain point cloud data of the environment in front of the robot in other poses.

[0076] This invention provides a method for a robot to autonomously identify grain piles in a warehouse. It relies solely on point cloud data and three-dimensional posture data acquired by its own onboard perception system. By analyzing spatial features, geometric features, and the natural stacking patterns of grain piles, it automatically identifies grain piles and extracts the three-dimensional coordinates of their tops. This provides key input for subsequent intelligent operations such as leveling grain piles. Furthermore, it does not rely on equipment on the top of the warehouse, requires no human intervention, and is suitable for complex grain pile environments.

[0077] Optionally, the preprocessing of the point cloud data in step S31 specifically includes: Dust filtration noise: N frames of point cloud data are continuously acquired, where N is a preset interval, to obtain a set of time-series point cloud data {P1, P2, ..., P_N}; For each point in each frame, its neighboring points are found in a multi-frame sequence. If a point has stable neighboring points in most frames, it is retained; otherwise, it is discarded, resulting in stable point cloud data. ; Filter outliers: For each point Calculate the average distance between it and its k nearest neighbors. k is a preset parameter:

[0078] Calculate all of the point cloud mean and standard deviation :

[0079] Set distance threshold The calculation formula is as follows:

[0080] in These are adjustable parameters; like Then the point is considered It identifies outliers and removes them; The preprocessed point cloud data is denoted as:

[0081] Optionally, S32 specifically includes: S32-1, Creating a Kd-tree and key parameters For point clouds Construct a Kd-Tree data structure. Kd-Tree can greatly improve the efficiency of finding the neighborhood of a point by recursively partitioning the space. The following key parameters are preset according to the characteristics of the grain: Neighborhood search radius The distance threshold is set based on the sparsity of the point cloud and the surface characteristics of the grain pile of a specific grain variety. If the distance between two points is less than this distance threshold, they will be classified into the same cluster. Minimum number of points in a single cluster With the maximum number of points Used to determine whether a cluster is effective; clusters with too few points may be noise or residual matrix debris; clusters with too many points may be multiple grain piles incorrectly connected together. S32-2, Cluster Region Growth Initialize an empty list of clusters. And a list Lvisit that records whether a point has been visited; Combine the aforementioned list Lvisit to traverse the point cloud. Each point in For each visit: if Not yet visited, mark it first. Visited, then... Create a new point cloud cluster for the seed point. Perform region growth: Search radius in the neighborhood Inside, search All neighboring points; For each unvisited neighboring point Mark it as visited and join ; Recursively using the newly added point as the seed, repeat the above neighbor search and addition process until no new point can be added to the cluster; When the region growth ends, check Points : if If it is a valid candidate grain cluster, it is added to the clustering list. ; Otherwise, consider it an invalid cluster and discard it; Ultimately, multiple independent candidate point cloud clusters are obtained, each of which is considered an independent potential grain pile:

[0082] in, The number of potential grain piles identified.

[0083] Optionally, step S33 involves geometric size filtering of the candidate point cloud clusters, specifically including: For each candidate point cloud cluster, calculate its three-dimensional bounding box to obtain the height and horizontal span of the candidate grain pile; Filtering is performed according to pre-set height and horizontal span thresholds to retain point cloud clusters that meet the criteria.

[0084] Optionally, step S33 involves screening the candidate point cloud clusters based on the natural accumulation pattern of grain piles, specifically including: Different grain varieties have different internal friction coefficients, which determines the different slopes of the grain piles formed after the grain conveyor throws the grain from a height. The slope value of the grain pile is preset according to different grain varieties. The candidate point cloud clusters are screened according to the natural accumulation pattern of grain piles, including: Different grain varieties have different internal friction coefficients, which determines the different slopes of the grain piles formed after the grain conveyor throws the grain from a height. The slope value of the grain pile is preset according to different grain varieties. The candidate point cloud clusters are screened according to the natural accumulation pattern of grain piles, including: a. such as Figure 17 As shown on the left, each candidate point cloud cluster is divided into several layers (e.g., 4 layers) along the Z-axis direction (longitudinal depth section) in the XY plane. In the point cloud data of each layer, the coordinates of the vertex and the lower corner point of the section are obtained, and the slope of each corner is obtained by using the tangent function. It is verified whether it is close to the preset grain pile slope value. If the difference is too large, the candidate point cloud cluster is removed. b. Figure 17 As shown on the right, each candidate point cloud cluster is divided into several layers (e.g., 5 layers) along the Y-axis direction (horizontal tangent) using the XZ plane. In the point cloud data of each layer, the edge points of the horizontal tangent and the vertices of the relative point cloud clusters are obtained. The slope of each corner is calculated using the tangent function to verify whether it is close to the preset grain pile slope value. If the difference is too large, the candidate point cloud cluster is removed.

[0085] Optionally, S34 specifically includes: Extract the coordinates (Xa, Ya, Ya) of each effective grain pile vertex in the point cloud coordinate system, and combine them with the coordinates (Xm, Ym, Zm) of the vehicle in the warehouse coordinate system obtained by the robot's autonomous localization, and the attitude angle: pitch angle θ. p Roll angle θ r yaw angle θ y The coordinates (X, Y, Z) of the grain pile vertex in the warehouse coordinate system are calculated by following these steps: Since the robot is currently positioned at the top of the grain pile and looking straight ahead to identify it, a pitch angle θ is set. p and roll angle θ r Since both are 0°, the rotation matrix simplifies to a rotation matrix around the y-axis as follows: ; The coordinates (X, Y, Z) of the grain pile vertex in the warehouse space are obtained through rotation and translation, and the calculation formula is as follows: ; ; ;

[0086] Transform each vertex of the effective grain pile one by one to obtain the coordinates (X, Y, Z) of all effective grain pile vertices in the warehouse coordinate system.

[0087] The invention's embodiments can accurately extract the three-dimensional coordinates of the grain pile apex: directly locating the highest point of each grain pile; it does not rely on external equipment: only its own onboard sensing system is needed, without the need for radar or cameras on the top of the warehouse; it is based on the natural shape and point cloud patterns of the grain pile: utilizing the physical characteristics of the grain pile formed by the internal friction coefficient of the grain, the recognition accuracy is high; it is robust and adaptable to complex environments: through spatial features, geometric features, and the natural stacking patterns of the grain pile, it eliminates interference from non-grain piles and adapts to warehouse environments with a large number of grain piles and chaotic distribution; it has high computational efficiency and is easy to deploy: it does not rely on deep learning models, has strong real-time performance, and is suitable for embedded or vehicle-mounted computing platforms.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for a robot to autonomously identify grain piles in a warehouse, characterized in that, The robot includes a wheeled, biped or other robot that can travel on the grain surface, and the grain pile recognition is realized only by the perception system carried by the robot itself without the need for any pre-installed electronic equipment in the warehouse, and the method comprises the following steps: S31, acquiring point cloud data of the environment in front of the robot in all poses, and preprocessing the point cloud data; S32, using spatial proximity, separating potential grain pile point cloud data by clustering to obtain candidate point cloud clusters; S33, respectively performing geometric size screening and grain pile natural accumulation rule screening on the candidate point cloud clusters to obtain effective grain piles; S34, according to the coordinates of the effective grain pile vertexes in the point cloud coordinate system and the coordinates and attitude angle of the robot in the warehouse coordinate system, calculating the coordinates of the effective grain pile vertexes in the warehouse coordinate system.

2. The method of claim 1, wherein, In the S31, the point cloud data of the environment in front of the robot in all poses is acquired, specifically comprising the following steps: When the robot is located at the top region of the grain pile, the point cloud data of the environment in front of the robot in the current pose is obtained by the point cloud acquisition device carried by the robot itself; The robot is horizontally rotated to obtain the point cloud data of the environment in front of the robot in other poses.

3. The method of claim 1, wherein, In the S31, the point cloud data is preprocessed, specifically comprising the following steps: Filtering dust noise: Continuously collecting N frames of point cloud data, N being a preset interval, to obtain a group of time series point cloud data {P1, P2,..., P_N}; For each point in each frame, find its neighboring points in multiple frames, if the point has stable neighboring points in most frames, keep it, otherwise, remove it, to obtain stable point cloud data: }； Filtering outliers: For each point , compute its average distance to the k nearest neighbors , k is a pre-set parameter: ； Compute the mean and standard deviation of all points in the entire point cloud : ； Setting a distance threshold , which is calculated as follows: ； wherein is an adjustable parameter; If , then the point is considered an outlier and is removed. The pre-processed point cloud data is denoted as: .

4. The method of claim 3, wherein, In the S32, specifically comprising the following steps: S32-1, creating a Kd-tree and key parameters; For point cloud A Kd-Tree data structure is constructed, and the Kd-Tree can greatly improve the efficiency of finding the neighborhood of points by recursively dividing the space. According to the grain characteristics, the following key parameters are set in advance: Neighborhood search radius: a distance threshold value set according to the sparsity of the point cloud and the grain pile surface characteristics of the specific grain variety, if the distance between two points is less than the distance threshold value, they will be classified into the same cluster; Single cluster minimum number of points With maximum number of points : used to determine if a cluster is valid, a cluster with too few points may be noise or a residual base fragment; a cluster with too many points may be multiple grain piles incorrectly connected together; S32-2, region growing clustering; Initialize an empty list of clusters and a list Lvisit that records whether a point has been visited or not. In combination with the list Lvisit, each point in the point cloud is traversed for each visit: If Not yet visited, mark as visited Visited, then create a new point cloud cluster with Seed point, perform region growing : Within the neighborhood search radius , all neighboring points of are searched for; For each unvisited neighboring point mark it as visited and add ; Recursively taking the newly added point as a seed, repeating the above neighboring point searching and adding process until there is no new point that can be added to the cluster; When the region growing is finished, check the number of points : If then consider this a valid candidate grain cluster and add it to the list of clusters ; Otherwise, it is regarded as an invalid cluster and discarded; Finally, a plurality of independent candidate point cloud clusters are obtained, and each candidate point cloud cluster is regarded as an independent potential grain pile: ； wherein, is the number of identified potential grain piles.

5. The method of claim 1, wherein, In the S33, the candidate point cloud clusters are screened by geometric size, specifically comprising the following steps: For each candidate point cloud cluster, calculate its three-dimensional bounding box to obtain the height and horizontal span of the candidate grain pile; According to the pre-set height threshold and horizontal span threshold, filter and retain the point cloud clusters that meet the conditions.

6. The method of claim 1, wherein, In the S33, the candidate point cloud clusters are screened by the natural accumulation rule of the grain pile, specifically comprising the following steps: Different grain varieties have different internal friction coefficients, which determines the different slopes of the grain piles formed after the grain is thrown from a high place by a grain conveying machine, and the grain pile slope value is preset according to different grain varieties; The candidate point cloud clusters are screened according to the natural accumulation rule of the grain pile, including: a. Split the point cloud into several layers along the Z-axis direction in the XY plane, and obtain the coordinates of the vertex and the lower corner point in each layer of point cloud data. Use the tangent function to obtain the slope of each corner, and verify whether it is close to the preset grain pile slope value. If the difference is too large, the candidate point cloud cluster is removed; b. Split the point cloud into several layers along the Y-axis direction in the XZ plane, and obtain the edge point and the vertex of the relative point cloud cluster in the horizontal section of each layer of point cloud data. Use the tangent function to obtain the slope of each corner, and verify whether it is close to the preset grain pile slope value. If the difference is too large, the candidate point cloud cluster is removed.

7. The method of claim 1, wherein, The S34 specifically comprises: The coordinates (Xa, Ya, Ya) of each effective grain pile vertex in the point cloud coordinate system are extracted, and the coordinates (Xm, Ym, Zm) and the attitude angles: the pitch angle θ p , the roll angle θ r , and the yaw angle θ y of the trolley in the warehouse coordinate system obtained by the robot autonomous positioning are combined to calculate the coordinates (X, Y, Z) of the grain pile vertex in the warehouse coordinate system according to the following steps: Since the robot is located at the top of the grain pile at this time, the pitch angle θ is set to 0° p and the roll angle θ r is set to 0°, so the rotation matrix is simplified to a rotation matrix around the y-axis as follows: ； The coordinates (X, Y, Z) of the grain pile vertex in the warehouse space are obtained by rotation and translation, and the calculation formula is as follows: ； ； ； Each grain pile vertex in the effective grain pile is converted to obtain the coordinates (X, Y, Z) of all effective grain pile vertices in the warehouse coordinate system.

8. A robot-driven autonomous identification system for grain piles in a warehouse, characterized in that, The system comprises: An acquisition preprocessing module is configured to acquire point cloud data of an environment in front of the robot in all poses, and to pre-process the point cloud data; A clustering separation module is configured to separate potential grain pile point cloud data through clustering to obtain candidate point cloud clusters by using spatial proximity; An effective grain pile screening module is configured to perform geometric size screening and grain pile natural accumulation rule screening on the candidate point cloud clusters to obtain effective grain piles; A warehouse coordinate calculation module is configured to calculate the coordinates of the effective grain pile vertices in the warehouse coordinate system according to the coordinates of the effective grain pile vertices in the point cloud coordinate system, and the coordinates and attitude angles of the robot in the warehouse coordinate system. 9.An electronic device comprising a processor and a memory in which at least one instruction is stored, wherein, The at least one instruction is loaded and executed by the processor to implement the method for the robot to autonomously identify grain piles in a warehouse according to any one of claims 1-7.

10. A computer-readable storage medium having stored therein at least one instruction, wherein The at least one instruction is loaded and executed by the processor to implement the method for the robot to autonomously identify grain piles in a warehouse according to any one of claims 1-7.

Images