3D scanning laser radar, laser radar scanning method and unmanned forklift

Abstract

The invention discloses a 3D scanning laser radar, a laser radar scanning method and an unmanned forklift. According to the method, a mathematical model of a rotation speed relation and point cloud space angle distribution is established, a main shaft rotation speed r1 is fixed, a high-speed rotation speed r2 is adjusted to form non-integer proportional motion, and non-repeated scanning is realized to generate uniform point cloud. And calculating a point cloud uniformity index delta S based on the model, generating a characteristic spectrum, and selecting an optimal rotating speed to support uniform, line-by-line or repeated scanning mode switching. Furthermore, through pre-calculated gyroscopic effect compensation and bidirectional laser arrangement, a crossed oblique line grid is formed, and the boundary recognition precision is improved; and adjusting the ratio of r1 to r2 to control the grid angle so as to realize an acute-angle, right-angle or obtuse-angle grid. The problems that the resolution of a traditional radar is limited by the number of modules, the coverage speed is low, and control is complex are solved, through open-loop control and dynamic adjustment, cost and power consumption are reduced, the point cloud density and task adaptability are improved, and the method is suitable for the fields of automatic driving, robots and the like.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional perception technology, and in particular to a 3D lidar scanning method for multiple scanning units, which aims to generate non-repetitive point clouds suitable for applications such as autonomous driving, robotics, and mapping. Background Art

[0002] In the prior art, 3D laser radar is widely used in the fields of robot navigation, obstacle avoidance and target recognition. It detects the three-dimensional coordinates of space targets in real time and organizes them into point clouds for output. According to the organization form of point clouds, it can be divided into repetitive scanning and non-repetitive scanning. Repetitive scanning usually uses a spindle motor to drive multiple transmitting and receiving modules to rotate to form a standard spatial grid coverage, such as using multiple light sources to complete longitudinal or transverse field of view coverage, and then using a rotating mechanism to achieve grid filling in other directions. For example, Chinese patent CN113687387B discloses a laser radar scanning device and method, which includes a first laser emitter, a first galvanometer, a second laser emitter and a second galvanometer. By controlling the first laser emitter to emit a first laser and controlling the rotation of the first galvanometer, the first laser is scanned along the transverse direction of the target object. At the same time, the second laser emitter is controlled to emit a second laser and the rotation of the second galvanometer is controlled to cause the second laser to scan along the longitudinal direction of the target object. This method emphasizes the coordination of dual galvanometers to achieve two-dimensional scanning, and the point cloud is organized into a repetitive grid to improve scanning efficiency. This technology structurally relies on multiple laser emitters and galvanometer assemblies to achieve grid coverage of space. However, resolution is limited by the number of transmitter / receiver modules. Higher resolution requires additional modules, increasing cost and power consumption. Currently, the highest line count available on the market is only 128, which lags behind visual resolution. The patented scanning process relies on precise rotational control of the galvanometer, and stability in highly dynamic environments relies on a closed-loop feedback mechanism. The algorithm is complex and consumes significant computing resources.

[0003] Non-repetitive scanning uses fewer transmitting and receiving modules, and forms a non-repetitive point cloud by controlling the scanning mechanism to adjust the phase in a pseudo-random manner in two orthogonal directions. As the integration time increases, the coverage area increases, and the equivalent resolution is improved. For example, Chinese patent CN114782651A discloses a method for automatic calibration of external parameters of a non-repetitive scanning 3D laser radar and a thermal camera. The method includes heating or cooling a calibration plate, aligning the laser radar and camera with the calibration plate and moving it, collecting multiple sets of point clouds and image data, extracting feature points of the calibration plate, and solving external parameters through coordinate pairing and nonlinear optimization. This technology utilizes the non-repetitive scanning characteristics to achieve high-density coverage by accumulating multiple frames of point clouds, and considers the non-repetitive trajectory of the laser radar during the calibration process, but the calibration relies on external auxiliary equipment and multiple sets of data collection, and is suitable for static calibration scenarios. In point cloud generation, this method emphasizes that non-repetitive paths fill the preceding gaps to improve long-term resolution, but the short-term point cloud is sparse and the repetition rate is less than 200kHz, resulting in a longer time to reach effective resolution. The calibration process of this patent involves multi-step optimization calculations and relies on camera assistance, which increases the complexity of the system and has low calibration efficiency in actual deployment.

[0004] Another type of technology focuses on compensating for scanning deviations to optimize point cloud quality. For example, international patent WO2021226763A1 discloses a lidar system based on non-integer speed ratios and closed-loop synchronous control. Through a master-slave control architecture, it feeds back the position information of the slave scanning module in real time, generates a characteristic signal describing the position difference (i.e., phase error), and adjusts the motion parameters of the slave module accordingly to accurately maintain the preset non-integer speed ratio, thereby achieving stable non-repetitive scanning. Although this method can optimize the scanning trajectory, it relies on complex feedback loops and algorithms, consumes a lot of computing resources, increases system cost and power consumption, and is difficult to support dynamic adjustment in low-power scenarios. In addition, most existing scanning point clouds pass through objects in one direction, and edges parallel to the scanning direction are easily missed or measured inaccurately, making it impossible to form an ideal orthogonal or inclined cross grid, resulting in incomplete boundary recognition of small objects. The vertical field of view angle is generally limited to 0° to 90° or -15° to 30°, which makes it difficult to cover the top and bottom space of a robot with height. These shortcomings limit the application of lidar in smart devices, and a method is needed that can simplify control logic, reduce costs and improve scanning flexibility.

[0005] The resolution of existing repetitive scanning is limited by the number of modules, and its high cost and power consumption hinder its development, making it unable to meet the needs of robots for high-precision interaction. Although non-repetitive scanning can improve resolution through integration, its spatial coverage is slow. In addition, existing scanning point clouds are mostly one-way through the object, which easily leads to missed detection or inaccurate measurement of edges parallel to the scanning direction. This prevents the formation of an ideal orthogonal or oblique cross-grid, resulting in incomplete boundary recognition of small objects. The vertical field of view is generally limited to 0° to 90° or -15° to 30°, making it difficult to cover the space above and below the height of the robot. Existing control methods are mostly closed-loop mechanisms. While these closed-loop methods can achieve precise trajectory optimization, they rely on complex feedback loops and algorithms, increasing system cost and power consumption, and making it difficult to support dynamic adjustment in low-power scenarios. These shortcomings limit the application of lidar in smart devices, and a method is needed that can simplify control logic, reduce costs, and improve scanning flexibility. Summary of the Invention

[0006] The present invention provides a laser radar scanning method, which solves the problems in the prior art of repeated scanning resolution being limited by the number of modules, slow coverage speed, inaccurate boundary recognition, and high cost caused by complex closed-loop control.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A 3D laser radar scanning method is applied to a 3D perception system including a first execution unit and a second execution unit capable of relative motion. The method generates a non-repeatedly scanned 3D point cloud based on a non-integer ratio motion relationship between the two execution units. The method includes:

[0009] a) establishing a mathematical model describing the relationship between the rotational speed relationship between the first execution unit and the second execution unit and the final point cloud spatial angular distribution, wherein the first execution unit is a spindle execution unit with a fixed rotational speed r1, and the second execution unit is a high-speed execution unit with an adjustable rotational speed r2; based on the mathematical model, obtaining a distribution array S by performing histogram statistics on the point cloud spatial angular distribution, and calculating the standard deviation δ of the distribution array S s , to quantify the spatial distribution uniformity of the point cloud;

[0010] b) Repeat the calculation of step a) for a series of different second execution unit motor speeds r2, thereby generating the uniformity index δ s A characteristic spectrum that changes with the speed r2 of the motor of the second execution unit;

[0011] c) analyzing the feature maps based on an expected point cloud shape, and selecting a set of motor speeds corresponding to the expected point cloud shape, wherein the expected point cloud shape includes a uniformly scanned point cloud, a line-by-line scanned point cloud, or a repeatedly scanned point cloud;

[0012] d) instructing the first execution unit and the second execution unit to operate at the motor speed corresponding to the optimal operating point, wherein the speed r1 of the first execution unit remains unchanged, and the expected point cloud shape is achieved by adjusting the speed r2 of the second execution unit, and supporting real-time adjustment of the speed during operation to switch between different point cloud shapes.

[0013] Furthermore, the rotation periods of the first execution unit and the second execution unit are calculated; a mathematical model is established based on the rotation speeds r1 and r2, the output angle is associated with the rotation speeds r1 and r2, and a sequence f(n) representing the distribution of all scan line angles is generated by recursion; a histogram statistics is performed on the sequence f(n) to obtain a distribution array S, and the standard deviation δ of the distribution array S is calculated. s , to quantify the spatial distribution uniformity of the point cloud.

[0014] Furthermore, the calculation of the rotation period of the first execution unit and the second execution unit specifically includes: calculating the rotation period T1 = 60 / r1 of the first execution unit, where r1 is the spindle motor speed; calculating the scanning period T2 = 60 / (nr2) of each prism of the second execution unit, where knife is the number of prism edges; in the mathematical model, the distribution angle of each prism in space is calculated as θ = mod(360(T2 / T1)*k, 360), where k is an integer index, and all angle distribution sequences f(n) are generated by recursion with a given initial phase; performing histogram statistics on the sequence f(n) to obtain a distribution array S, and calculating the standard deviation δ of the distribution array S. s , to quantify the spatial distribution uniformity of the point cloud.

[0015] Furthermore, according to the rotation speed r2 value and the uniformity index δ s Generate the characteristic map, wherein the characteristic map includes multiple curves corresponding to different integration times, and the curves are periodic waveforms, which are used to identify the rotational speed points at which all curves coincide with valley values ​​to achieve a point cloud that is uniformly encrypted over time, or to identify the rotational speed points at which the short integration curve is at a peak value and the long integration curve is at a valley value to achieve line-by-line scanning of the point cloud.

[0016] Furthermore, by adjusting the rotation speed r2 of the second execution unit, three scanning states are switched: repeated scanning point cloud state, in which all scanning lines coincide; uniform non-repetitive scanning point cloud state, in which the point cloud is evenly distributed and gradually encrypted; line-by-line scanning point cloud state, in which the point cloud is filled line by line along a specific direction.

[0017] Furthermore, the method for generating the gyroscopic effect by the rotation of the second execution unit includes: calculating the ratio of the moment of inertia J1 of the driving part of the second execution unit to the moment of inertia J2 of the compensation part; adjusting the rotational speed rb of the compensation part in real time according to the ratio so that it satisfies ra / rb=J2 / J1 with the rotational speed ra of the driving part of the second execution unit; and correcting the rotational speed rb in real time by controlling the driving signal to offset the gyroscopic effect.

[0018] Furthermore, the method includes controlling the scanning trajectory by adjusting the rotational speed ratio of the first execution unit and the second execution unit to generate a bidirectional cross-slash grid to replace the unidirectional parallel diagonal scanning, thereby scanning the object boundary from two directions; gradually refining the cross-grid as the integration time increases, from a coarse grid state to a dense state, wherein the scanning lines cross to fill the blank area; scanning a square object, wherein the cross-slash scanning hits the side boundary of the object from two directions, and gradually encrypts the point cloud coverage under non-repetitive scanning conditions to achieve complete capture of the object boundary.

[0019] Furthermore, the method includes controlling the grid crossing angle by adjusting the ratio of the rotation period T1 of the first execution unit to the effective scanning period T2 of the second execution unit, wherein an acute-angle crossing grid is generated when the T2 / T1 ratio is small, a right-angle crossing grid is generated when the T2 / T1 ratio is moderate, and an obtuse-angle crossing grid is generated when the T2 / T1 ratio is large.

[0020] Furthermore, step a) further includes: setting the scanning period T1 of the first execution unit and the rotation period T2 of the second execution unit, controlling T1 / T2 to be a non-integer ratio to ensure non-repetitive scanning; setting the target non-integer ratio T1 / T2 = N±O+ε, where N is a positive integer, 0<ε<1, and O is a small irrational number.

[0021] Furthermore, the step c) further comprises: analyzing the uniformity index δ of the characteristic spectrum s The curve is selected to determine the rotation speed r2 corresponding to the expected point cloud shape, wherein when the rotation speed r2 at which all curves coincide with the valley value is selected, a uniform scanning point cloud is generated; when the rotation speed r2 at which the short integration time curve is at the peak value and the long integration time curve is at the valley value is selected, a line-by-line scanning point cloud is generated; and when the rotation speed r2 at which all curves coincide with the peak value is selected, a repeated scanning point cloud is generated.

[0022] A 3D laser radar, comprising:

[0023] a main scanning unit, configured to perform a first scanning motion with a first motion parameter;

[0024] an auxiliary scanning unit, configured to perform a second scanning motion with a second motion parameter;

[0025] a processor electrically connected to the main scanning unit and the auxiliary scanning unit;

[0026] The processor is configured to: establish a mathematical model describing the relationship between the rotational speed relationship of the main scanning unit and the auxiliary scanning unit and the spatial angular distribution of the point cloud, control the first rotational speed r1 to be fixed, and adjust the second rotational speed r2 to form a non-integer ratio motion relationship; based on the mathematical model, obtain a distribution array S by performing histogram statistics on the spatial angular distribution of the point cloud, and calculate the standard deviation δ of the distribution array S s For a series of different second speeds r2, the uniformity index δ is repeatedly calculated s , generating δ s A characteristic map that changes with r2; analyzing the characteristic map according to an expected point cloud shape to determine a rotational speed r1 and r2 combination corresponding to a target working point, wherein the expected point cloud shape includes a uniformly scanned point cloud, a line-by-line scanned point cloud, or a repeatedly scanned point cloud; instructing the main scanning unit and the auxiliary scanning unit to operate at a rotational speed corresponding to the target working point, and supporting real-time adjustment of the second rotational speed r2 to switch the point cloud shape.

[0027] Furthermore, the main scanning unit includes a first motor and a main reflection mirror, and the auxiliary scanning unit includes a second motor and a polygonal reflection prism.

[0028] Furthermore, it includes an inertia compensation unit, which is coaxially arranged with the rotation axis of the auxiliary scanning unit, and the processor controls the inertia compensation unit and the auxiliary scanning unit to move at the same angular velocity and in opposite rotation directions.

[0029] Furthermore, the inertia compensation unit includes a third motor and a counterweight.

[0030] Furthermore, the auxiliary scanning unit includes at least two laser transceiver modules, which are arranged in a manner to emit laser beams in opposite directions in the same scanning plane to generate a bidirectional cross-slash grid; the processor controls the grid cross angle to generate an acute angle, right angle or obtuse angle cross grid by adjusting the ratio of the first rotation speed r1 to the second rotation speed r2.

[0031] Furthermore, the laser transceiver module includes a laser emitting part and a laser echo receiving part, and at least one internal reflector is provided inside the laser emitting part and / or the laser echo receiving part for folding the optical path so that the laser emitting optical path or the echo receiving optical path is L-shaped or U-shaped.

[0032] Furthermore, the polygonal reflecting prism is a hollow structure, the second motor is an external rotor motor, and its rotor is arranged inside the hollow polygonal reflecting prism and coaxially connected to it; the main reflector is configured to support a spatial field of view angle range of -20° to 90° or -45° to 60° through position adjustment.

[0033] Furthermore, the processor stores a plurality of preset scanning mode configuration files, each configuration file corresponds to a point cloud distribution characteristic, and the processor can load different configuration files according to external instructions to switch the scanning mode.

[0034] An unmanned forklift, comprising a vehicle body, a control system, and a 3D laser radar for environmental perception, wherein the 3D laser radar is any of the 3D laser radars described above, and the control system is configured as follows:

[0035] When performing large-scale path planning or autonomous navigation tasks, control the 3D lidar to operate in a globally uniform mode;

[0036] When approaching shelves or avoiding dynamic obstacles, the 3D LiDAR is controlled to switch to orthogonal obstacle avoidance mode.

[0037] When performing cargo identification or fork alignment with pallet holes, the 3D lidar is controlled to switch to local encryption mode.

[0038] The present invention provides a 3D laser radar scanning method, a 3D laser radar, and an unmanned forklift, which have the following beneficial effects:

[0039] First, the present invention forms a specific non-integer proportional motion relationship between a fixed-speed spindle execution unit r1 in the system and a high-speed execution unit with adjustable speed, such as a polygonal reflective prism, whose speed r2 can be dynamically adjusted, to achieve efficient non-repetitive scanning, thereby effectively breaking through the bottleneck of traditional laser radar resolution being limited by the number of physical laser modules. By keeping r1 constant and precisely adjusting r2, this method ensures that the scanning trajectory fills the blank areas in the field of view continuously and without overlap in time. As the integration time accumulates, the equivalent line number and spatial density of the point cloud can continue to increase. This enables systems with less laser transceiver hardware to generate high-resolution, high-density point clouds comparable to or even exceeding high-line-count traditional radars, fundamentally solving the problem that improving resolution in the existing technology must rely on increasing the amount of hardware, which in turn leads to a sharp increase in cost, power consumption and volume.

[0040] Secondly, the present invention does not simply achieve non-repetitive scanning, but rather quantitatively analyzes and optimizes the uniformity of the point cloud by pre-establishing a mathematical model that associates the relationship between the rotational speed of constant r1 and adjustable r2 and the spatial angle distribution of the final point cloud. This method specifically performs histogram statistics on the spatial angles of the scanned point cloud to obtain a distribution array, and calculates its standard deviation to quantify the uniformity. The characteristic map generated based on this model can clearly show the law of change of the point cloud uniformity index with the r2 rotational speed. By selecting the r2 rotational speed of the valley point in the corresponding map as the optimal working point, it is ensured that the point cloud always maintains a high degree of spatial distribution balance during the encryption process. This controlled and uniform encryption method avoids the local point cloud clustering or sparseness problems that may be caused by random scanning by fixing r1 and dynamically adjusting r2, and can obtain effective, dead-angle-free coverage of the entire perception area in a shorter time.

[0041] Furthermore, the present invention adopts an advanced open-loop control strategy. Unlike the existing technology that generally relies on complex closed-loop feedback mechanisms, the control system of the present invention is based on a preset mathematical model, and controls the speed by fixing r1 and dynamically adjusting r2, without the need for expensive external sensors. In the pre-compensation of physical interference, for example, by setting an inertia compensation unit that rotates coaxially and counter-rotating with the high-speed execution unit, such as including a third motor and a counterweight, the gyroscopic effect generated by the high-speed rotation is actively offset, thereby ensuring the stability of the scan at the physical level. This design eliminates complex real-time feedback loops and algorithms, greatly reduces the system's computing load and potential failure points, and enhances overall reliability.

[0042] Another significant advantage brought about by this is the high cost-effectiveness of the equipment. By eliminating the precision feedback sensors and complex drive compensation devices required for closed-loop control, and adopting compact designs such as a hollow multi-prism reflective prism with an embedded external rotor motor, the present invention significantly reduces the bill of materials cost and manufacturing cost of the lidar. At the same time, the laser transceiver module can fold the optical path by setting an internal reflector to achieve an L-shaped or U-shaped optical path, further reducing the size of the device. The simplified system architecture also means lower operating power consumption, making it easy to integrate into application platforms that have strict requirements on cost, energy consumption and space, such as battery-powered unmanned forklifts, and has broad market application prospects.

[0043] In addition, the present invention gives the laser radar unprecedented real-time adjustability and task adaptability. The processor can store a variety of preset scanning mode configuration files corresponding to different point cloud distribution characteristics, and different configurations can be loaded according to external instructions. By adjusting the r2 speed in real time, it is possible to seamlessly switch between multiple modes such as uniform non-repetitive scanning to select the valley r2 speed where all integral time curves coincide, line-by-line scanning to select the r2 speed at the peak value of the short-time curve and the valley value of the long-time curve, and repeated scanning to select the r2 speed at the peak value where all curves coincide, without interrupting the operation of the system. This optimized perception strategy through dynamic adjustment of r2 enables equipment such as unmanned forklifts to flexibly adjust point cloud and grid characteristics according to specific tasks such as large-scale navigation, obstacle avoidance by racks, and cargo identification, greatly improving operational efficiency and intelligence.

[0044] Finally, the present invention significantly improves the ability to identify the boundaries of objects through a special hardware layout and control method. By arranging at least two laser transceiver modules in opposite emission directions within the same scanning plane, combined with a control method in which r1 is constant and r2 is adjustable, the system can generate a bidirectional cross-grid scanning trajectory. This cross grid ensures that any object, especially its contour boundary, can be detected by laser beams from two different directions. What is more valuable is that by adjusting the r2 rotation speed to change the rotation speed ratio of the first execution unit and the second execution unit, the intersection angle of the grid can also be precisely controlled to generate acute-angle, right-angle or obtuse-angle grids to accommodate different detection targets. This method fundamentally solves the problem of missed detection caused by the scanning line being parallel to the edge of the object during one-way scanning, can capture a more complete and accurate three-dimensional form of the object, and significantly improves the reliability of the robot's autonomous obstacle avoidance and safe interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Attachment Figure 1a : 3D laser radar system framework logic diagram in the present invention;

[0046] Attachment Figure 1b : Schematic diagram of the structure of the 3D laser radar in the present invention;

[0047] Attachment Figure 2a : Schematic diagram of the structure of the scanning unit and the inertia compensation unit in the present invention;

[0048] Attachment Figure 2b : Schematic diagram of the structure of the scanning unit and the inertia compensation unit in the present invention;

[0049] Attachment Figure 2c : Schematic diagram of the relationship between the moment of inertia of the inertia compensation unit in the present invention;

[0050] Attachment Figure 3 : Schematic diagram of the optical path folding of the laser transceiver module in the present invention;

[0051] Attachment Figure 4a 、 4b 4c: Point cloud distribution at time points t1, t2, and t3 in the present invention;

[0052] Attachment Figure 5a 、 5b , 5c: in the present invention, respectively corresponding to Figure 4a 、 4b , projection of 4c on a two-dimensional plane;

[0053] Attachment Figure 6 : Demonstrates that the 3D LiDAR scanning method generates a uniformly tilted dense dot pattern in the angle sequence distribution;

[0054] Attachment Figure 7 :for Figure 6 Two-dimensional histogram statistics of ;

[0055] Attachment Figure 8 : is a cluster of curves showing the change of the point cloud uniformity index δS with the rotation speed of the auxiliary execution unit in the present invention;

[0056] Attachment Figure 9 :for Figure 8 Schematic diagram after partial enlargement;

[0057] Attachment Figure 10 : A flow chart of generating non-periodic disturbance in the present invention;

[0058] Attachment Figure 11 : Flowchart of generating grid structure adjustment amount in the present invention;

[0059] Attachment Figure 12 : Schematic diagram of the dynamic adjustment principle of the point cloud grid intersection angle in the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] In order to make the objectives, technical solutions and advantages of the present invention clearer and more complete, the following will provide a comprehensive, in-depth and detailed description of a 3D scanning laser radar, a laser radar scanning method and an unmanned forklift using the laser radar disclosed in the present invention in combination with the accompanying drawings and specific embodiments.

[0062] 3D scanning laser weighs:

[0063] Figure 1aThe system logic architecture diagram of the 3D laser radar of the present invention is macroscopically displayed. The figure mainly includes three core parts: a main scanning unit 100, an auxiliary scanning unit 101, an inertia compensation unit 102, and a processor 103 as the brain of the system. The main scanning unit and the auxiliary scanning unit jointly realize the scanning of the laser beam in three-dimensional space through relative motion. The inertia compensation unit and the auxiliary scanning unit are coaxial and rotate in opposite directions, and are used to offset the gyroscopic effect generated during the movement through pre-calculated static balance to ensure the stability of the scan. As the control core, the processor is electrically connected to each unit, responsible for executing the core scanning control algorithm, selecting the rotation speed and sending instructions based on a preset mathematical model, and supporting dynamic adjustment to switch the scanning mode.

[0064] Figure 1b A structural schematic diagram of a specific embodiment of the present invention is shown.

[0065] The main scanning unit 100 is one of the core moving components for spatial scanning in the 3D lidar system of the present invention. Its primary function is to perform a first-dimensional scanning motion according to preset first motion parameters. In a typical embodiment of the present invention, the main scanning unit performs a stable, uniform rotational motion, specifically a 360° horizontal scan. This scanning method ensures extensive coverage of the scanned area.

[0066] In this specific embodiment, the functions of the main scanning unit 100 are implemented by the horizontal turntable 2 and its drive mechanism. The horizontal turntable 2 structurally comprises a spindle motor 21 and a horizontal turntable 22. The spindle motor 21 consists of a spindle motor stator 211 and a spindle motor rotor 212. In the installation orientation, the spindle motor 21 is vertically positioned, with its stator 211 securely fixed to the radar base 1. The spindle motor rotor 212 is connected to the horizontal turntable 22. Based on this structure, the spindle motor 21, through the rotation of its rotor 212, can drive the horizontal turntable 22 to rotate stably in the horizontal direction, thereby completing the scanning task defined by the main scanning unit.

[0067] The auxiliary scanning unit 101 is another core motion component, responsible for executing a second-dimensional scanning motion that is orthogonal to or at a predetermined angle to the motion of the main scanning unit. This unit works in conjunction with the main scanning unit according to preset second motion parameters to achieve comprehensive scanning coverage of the three-dimensional space. The core control method of the present invention precisely adjusts the auxiliary scanning unit's motion parameters, particularly during dynamic adjustment and multi-mode switching, ensuring the system's adaptability to diverse operating environments.

[0068] In this specific embodiment, the function of the auxiliary scanning unit is realized by the rotating mirror module 5. The rotating mirror module 5 structurally includes a high-speed motor bracket 51, a high-speed motor 52 and a reflecting prism 53. The high-speed motor 52 is composed of a high-speed motor stator 521 and a high-speed motor rotor 522. In the installation orientation, the high-speed motor 52 is placed horizontally, and its high-speed motor stator 521 is firmly fixed on the horizontal turntable 22 through the high-speed motor bracket 51. In this preferred embodiment, the high-speed motor rotor 522 is an outer rotor structure and is coaxially connected to the reflecting prism 53. Its function is to drive the reflecting prism 53 to perform high-speed horizontal axial rotation above the laser transceiver module 4, thereby completing the scanning task defined by the auxiliary scanning unit.

[0069] In addition to the above two motion units, the 3D laser radar also includes key optical components for realizing light path transmission and reception. They are carried on the horizontal turntable 22 of the main scanning unit and rotate accordingly.

[0070] Specifically, a laser transceiver module 4 and a polygonal reflector 3 are directly fixed to the horizontal turntable 22. In this preferred embodiment, the two are arranged axially symmetrically on the horizontal turntable 22 to ensure dynamic balance during rotation. The laser transceiver module 4 is located in the central area of ​​the horizontal turntable 22, while the two reflectors 3 are located on the two laser transceiver sides of the laser transceiver module 4. The function of the reflector 3 is to reflect the laser beam emitted from the laser transceiver module 4 upward at a preset angle onto the reflective surface of the auxiliary scanning unit's reflective prism 53. Ultimately, the scanning laser is reflected again by the reflective prism 53, and then emitted into the external environment.

[0071] The processor 103 is the core control unit of the lidar system of the present invention, responsible for the management and scheduling of the entire system. It establishes electrical connections with key units such as the main scanning unit and auxiliary scanning unit through an internal bus or interface circuit to ensure the coordinated operation of each unit. The main responsibilities of the processor include:

[0072] Execute the core scanning control algorithm disclosed in the present invention;

[0073] Based on a preset mathematical model, setting and maintaining a non-integer motion speed ratio between the main scanning unit and the auxiliary scanning unit;

[0074] Load the pre-stored scan mode configuration file according to external instructions and select the corresponding speed combination;

[0075] Send fixed speed instructions to control the operation of the main scanning unit and auxiliary scanning unit;

[0076] Supports dynamic speed adjustment based on external instructions to achieve switching between different scanning modes;

[0077] Interact with upper-level application systems for data and instructions.

[0078] Figure 2a 、 2b Figures 2c and 2c detail the core components of the multi-motor compound rotational angular momentum cancellation structure, mounted on the horizontal turntable 22, in an embodiment of the present invention. This structure is designed to proactively eliminate the gyroscopic effect generated by the compound rotation of the multiple motors through pre-calculated static balance, thereby ensuring operational stability and measurement accuracy of the 3D LiDAR.

[0079] The structure primarily comprises a rotating mirror module 5 and a set of counter-rotating modules, both of which are secured to the horizontal turntable 22 via a motor bracket 51. The rotating mirror module 5 includes a high-speed motor 52 and a reflective prism 53, enabling vertical scanning. However, this rotation, combined with the horizontal rotation of the spindle motor 21, produces a compound gyroscopic effect, leading to device vibration.

[0080] To eliminate this vibration, one of the features of this solution is the provision of a counter-rotation module. This module includes a third motor 6 and a counterweight load 7. Its key structural feature is that the third motor 6 is coaxially arranged with the high-speed motor 52. The operating principle of the compensation mechanism is that the third motor 6 drives the counterweight load 7 to rotate in the opposite direction, generating angular momentum that can compensate for the rotation of the high-speed motor 52.

[0081] Composite rotational angular momentum balance equation:

[0082] ra·J1=rb·J2

[0083] in:

[0084] ra: the configured speed of the high-speed motor 52;

[0085] J1: the total moment of inertia of the high-speed motor 52 and the reflective prism 53 fixed thereon;

[0086] rb: the rotation speed of the third motor 6;

[0087] J2: is the total moment of inertia of the third motor 6 and the counterweight load 7 fixed thereon.

[0088] During the design phase, processor 103 sets the speed of third motor 6 at rb = (J1 / J2)·ra based on the precalculated moment of inertia ratio J1 / J2, and controls the operation of the counter-rotating modules via a fixed drive signal. When this equilibrium equation holds, the angular momentum generated by the two counter-rotating modules is equal in magnitude and opposite in direction, theoretically canceling out completely, ensuring smooth operation of the LiDAR. Thanks to the stability of the high-precision motor and the pre-calibrated mechanical design, the present invention maintains long-term stability without the need for real-time vibration monitoring or dynamic speed adjustment.

[0089] Figure 3 The schematic diagram intuitively illustrates the vertical scanning process of a 3D laser radar in an embodiment of the present invention. This diagram clearly illustrates how the rotation of the rotating mirror module deflects the laser beam emitted by the laser transceiver module 4 to different elevation angles, thereby achieving scanning coverage of the vertical dimension of space.

[0090] The principle behind vertical scanning is that when the high-speed motor 522 in the rotating mirror module drives the reflective prism 53 to rotate at high speed, the angle between the incident laser light and the different reflective surfaces of the reflective prism 53 changes continuously and periodically. According to the law of light reflection, this angle change directly causes the laser beam, after being reflected by the reflective prism 53 and ultimately entering the external environment, to continuously change its exit angle in the vertical plane, thereby achieving vertical scanning.

[0091] Figure 3 The specific radar scanning optical path process in this preferred embodiment is also detailed. The laser transceiver module 4 is mounted on the horizontal turntable 22 and emits a laser beam from its laser transceiver side. This laser beam is first reflected by the reflector 3 and then irradiated upward at a preset tilt angle onto a reflective surface of the reflective prism 53. The reflective prism 53 is supported by a high-speed motor bracket 51. The reflective surface of the reflective prism 53 then reflects the laser beam again, causing it to be emitted from the optical cover 6 of the laser radar into the external environment.

[0092] A significant feature of this solution is that the laser transceiver module 4 has two laser transceiver sides, so that it can synchronously transmit and process two scanning lasers on different optical paths during operation. Figure 3As shown, the two laser beams are reflected obliquely upward by their respective reflectors 3, then reflected again by two different cylindrical surfaces of the rotating reflective prism 53 before ultimately exiting the radar. This embodiment meticulously designs the optical paths of the two scanning laser beams so that, after striking different cylindrical surfaces of the reflective prism, the pitch angle of one laser beam relative to the environment increases as the prism rotates, while the pitch angle of the other laser beam relative to the environment decreases accordingly. In this way, the present invention achieves simultaneous vertical scanning at both ends, significantly improving vertical scanning efficiency during 3D scanning.

[0093] 3D scanning lidar scanning method:

[0094] Based on the aforementioned structure of a 3D laser radar in this embodiment, the present invention further discloses an innovative 3D laser radar scanning method. The hardware structure provides a solid physical platform for implementing this method, which fully utilizes the hardware's potential through open-loop control to generate high-quality 3D point cloud data.

[0095] The scanning method of the 3D scanning laser radar in this embodiment is applied to a three-dimensional perception system including a first execution unit (spindle motor 21) and a second execution unit (high-speed motor 52) that can move relative to each other. The core of this method is to adjust the rotational speeds of the spindle motor 21 and the high-speed motor 52 through a preset mathematical model so that their rotational speed ratio is a non-integer multiple, so as to achieve non-repetitive 3D scanning and obtain a pseudo-random point cloud. The density of the point cloud continues to increase with the increase in scanning time, avoiding the drawbacks of traditional periodic scanning, such as point cloud overlap or persistent blank areas. The method includes establishing a mathematical model of the relationship between the rotational speed and the spatial angle distribution of the point cloud, fixing the rotational speed of the first execution unit unchanged, and adjusting the rotational speed of the second execution unit to form a non-integer proportional motion; performing histogram statistics on the point cloud angle distribution based on the model to obtain a distribution array S, and calculating its standard deviation δ s Quantify uniformity; Repeat the calculation to generate δ for different second execution unit speeds s Characteristic maps that change with speed; select the optimal speed based on the expected point cloud shape analysis map; the instruction execution unit runs at the optimal speed and supports real-time adjustment of the switching shape at the second speed.

[0096] Figure 4a 、 4b 4c intuitively shows a three-dimensional diagram of the evolution of the non-repetitive scanning trajectory generated by the scanning method of the present invention over time. Figure 5a 、 5b , 5c correspond to Figure 4a 、 4b, 4c is projected onto a two-dimensional plane. This paper aims to illustrate how the present invention avoids the drawbacks of traditional periodic scanning through a specific open-loop control strategy, thereby obtaining a three-dimensional point cloud whose point cloud density continuously and evenly increases over time.

[0097] like Figure 4a As shown, the three-dimensional image of the point cloud presented at the minimum single-circle coverage time of the scan (t1: 33ms). Figure 5a for Figure 4a Projected onto a plane, the image shows a series of parallel and slightly tilted lines, forming a striped pattern with uniform spacing but noticeable gaps. The 3D image captures the volumetric curve scan path (suitable for visualization of complete spatial density for object boundary detection, such as in unmanned forklifts), while the 2D projection simplifies it into a linear representation, emphasizing angular uniformity and the potential for increasing density through integration time. The tilt effect is particularly pronounced in the central region (x≈150-200), where the lines converge into a dense vertical band, reflecting the focus of the scan concentration, indicating low point cloud coverage, and the pseudo-random trajectory characteristics generated by the non-integer speed ratio, which ensures that the gaps are subsequently filled.

[0098] At time t1, the point cloud generated by the 3D lidar exhibits a sparse distribution, similar to a low-density grid. This stage represents the initial state of non-repetitive scanning. The main scanning unit 100 drives the horizontal turntable 22 at a fixed speed to perform a 360° horizontal scan, while the auxiliary scanning unit 101 drives the reflective prism 53 at a high speed to achieve rapid vertical scanning. Because the speed ratio is set to a non-integer, the angular distribution of the scan lines in space exhibits a pseudo-random characteristic, avoiding the periodic overlap that occurs with traditional repetitive scanning.

[0099] The sparsity of the point cloud is determined by the non-integer ratio motion, the gaps between the scan lines are large, and the coverage is low. The mathematical model generates the initial scanning trajectory sequence by calculating the spatial distribution angle of each prism, and the histogram statistics obtain the distribution array S, whose standard deviation δ s A higher value indicates that the point cloud has lower uniformity.

[0100] This stage is suitable for quickly covering a large area, such as the initial environment perception of an unmanned forklift when performing large-scale path planning. The point cloud is sparse but sufficient to provide preliminary spatial contour information.

[0101] like Figure 4b As shown, the three-dimensional image of the point cloud presented at the intermediate time (t2: 500ms). Figure 5b for Figure 4bThe 2D projection shows interlaced, wavy lines with significantly reduced gaps, increased point cloud density, and improved coverage. The 3D image demonstrates the further densification of the spiral scanning path, with the dynamic speed adjustment of the auxiliary scanning unit optimizing spatial coverage. The wavy pattern in the 2D projection reflects the initial cross-effect of the bidirectional laser arrangement, enhancing boundary recognition. The standard deviation δS at this stage is smaller than at t1, with improved uniformity, making it suitable for medium-resolution scenarios, such as the orthogonal obstacle avoidance mode used by unmanned forklifts approaching shelves or avoiding dynamic obstacles.

[0102] The increase in point cloud density is due to the cumulative effect of non-integer scale motion. The mathematical model generates a new angle distribution sequence through recursion. The distribution array S of the histogram statistics shows a denser distribution, and the standard deviation δ s Compared to time t1, it is smaller, indicating that the uniformity has improved. The characteristic spectrum (the curve of δs changing with the speed) can be used to select the optimal speed point to optimize the filling effect.

[0103] This stage is suitable for scenarios requiring medium resolution, such as when an unmanned forklift approaches a shelf or avoids a dynamic obstacle. Orthogonal obstacle avoidance (a right-angled grid) is used to enhance boundary recognition accuracy. Gap filling in the point cloud improves the ability to capture object contours.

[0104] like Figure 4c As shown, the three-dimensional image of the point cloud presented at the final moment (t3: 1500ms). Figure 5c for Figure 4c The two-dimensional projection shows a highly uniform, dense mesh with no noticeable gaps, demonstrating near-complete coverage. The three-dimensional projection demonstrates complete spherical or ellipsoidal coverage, with spiral lines forming a complex intersecting pattern. The mesh structure in the two-dimensional projection highlights the fine boundary capture capability of the bidirectional cross-hatch mesh. The uniformity metric δS at this stage is near its minimum, reflecting the optimization of the high-density point cloud, making it particularly suitable for high-precision tasks such as the localized encryption pattern used by unmanned forklifts for cargo identification or fork alignment with pallet receptacles.

[0105] The uniformity and density of the point cloud are optimized by a mathematical model, and the standard deviation δ of the distribution array S is s The minimum value reflects the high uniformity of spatial distribution. The bidirectional cross-angle grid (controlling the sharp angle, right angle or obtuse angle grid by adjusting the rotation speed ratio) further enhances the boundary recognition ability and is particularly suitable for capturing the side boundaries of square objects.

[0106] This phase is suitable for high-precision tasks, such as the local encryption pattern used by unmanned forklifts to identify cargo or align forks with pallet receptacles. High-density point clouds ensure complete capture of the object's 3D form, improving operational efficiency and safety, especially in complex environments like densely packed racks.

[0107] The processor 103 sets and controls the movement speed ratio of the spindle motor 21 and the high-speed motor 52 to be a non-integer multiple based on a preset mathematical model. The purpose of this is to ensure that after the spindle motor 21 drives the entire scanning head to rotate one circle, the next circle of scanning lines generated by the high-speed motor 52 driving the reflective prism 53 will not completely overlap with the trajectory of the previous circle, but will accurately fill the previous gap, thereby achieving a non-repetitive scanning effect in which the point cloud density accumulates over time. The non-integer ratio is set in the following way: the scanning period of the first execution unit and the rotation period of the second execution unit are preset as a non-integer ratio to ensure non-repetitive scanning; the speed fluctuation correction value is pre-calculated, and the deviation is calculated according to the target non-integer ratio. This method constructs an open-loop control system based on a pre-calculated mathematical model. The system uses an accurate mathematical model as its target trajectory for operation:

[0108]

[0109] In this model:

[0110] T1 is the scanning period of the spindle motor 21, which is fixed and usually set to T1 = 60 / r1, where r1 is the spindle speed (unit: rpm). For example, when it is fixed at 3000 rpm, T1 = 0.02 seconds.

[0111] T2 is the scanning period of the high-speed motor 52, which can be dynamically changed by adjusting the speed. T2 = 60 / (nr2), where n is the number of prisms of the reflective prism 53 (e.g., 4 or 6) and r2 is the high-speed speed (ranging from 5000 to 10000 rpm). For example, when n = 4 and r2 = 6123 rpm, T2 = 60 / (4 × 6123) = 0.0025 seconds.

[0112] N: As a positive integer, it represents the basic integer multiple relationship between two periods. For example, N=4 means that the basic ratio is close to 4 times.

[0113] O: is a preset offset coefficient used to optimize the distribution of scan lines, controlling the position of subsequent scan tracks relative to the gap between previous scan tracks to improve coverage uniformity. It is usually set to a rational number between 0.1 and 0.5, for example, O = 0.314.

[0114] ε: A small adjustment used to ensure long-term non-repeatability to combat the instability of T1 and T2. It is pre-generated based on an irrational constant, such as ε = π / 100 ≈ 0.0314.

[0115] Figure 4a 、 4b, 4c show that at three consecutive time points t1, t2, and t3, the distribution of the scan point cloud presents a dynamic filling characteristic. At the initial moment t1, the scan line distribution is relatively sparse. Since the speed ratio is preset to a non-integer, at the subsequent moment t2, the new scan line is guided into the gap between the previous scan lines. As time passes to t3, the scan line continues to increase the blank area in the field of view, so that the coverage and density of the point cloud increase significantly and uniformly. Ultimately, this method generates a dense three-dimensional point cloud that presents a pseudo-random distribution at the macro level and is controlled by an accurate mathematical model at the micro level. The point cloud is generated as follows: pre-calculate the rotation period of the first and second execution units; associate the output angles based on the model, and generate a scan line angle sequence by recursion; obtain S from the sequence histogram statistics, and calculate δ s Quantify uniformity; spatial distribution angle is

[0116]

[0117] The point cloud generation process is as follows: First, we precalculate the rotation period T1 of the first execution unit (spindle motor 21) (such as when r1 = 3000 rpm, T1 = 0.02 seconds) and the rotation period T2 of the second execution unit (high-speed motor 52) under a given speed ratio (n is the number of prisms, such as n = 4, when r2 = 6000 rpm, T2 ≈ 0.0025 seconds); Based on the model, the output angle is associated with the speed ratio, and the scanning line angle sequence is generated by recursion, for example, θk = θ0 + k·(360° / n)·(T1 / T2) mod 360°, where θ0 = 0° initial phase, k is the scanning index (such as The first 20 k generate an angle sequence: f(n)[0.0, 73.644, 147.288, 220.932, 294.576, 8.22, 81.864, 155.508, 229.152, 302.796, 16.44, 90.084, 163.728, 237.372, 311.016, 24.66, 98.304, 171.948, 245.592, 319.236]); perform histogram statistics on the sequence (for example, divide the 360° space into grids of 1° each) to obtain the distribution array S and calculate its standard deviation δ s , to quantify uniformity.

[0118] like Figure 6 The figure shows the numerical distribution of the first 6140 angle sequences of the above distribution sequence θk within 3000ms. The horizontal axis is the ordinal number generated by the above method, and the vertical axis is the corresponding scan line angle (0°-360°)

[0119] Perform histogram statistics on the array set to obtain Figure 7This image is a 2D histogram showing the angular distribution of a 3D lidar point cloud at a fixed rotational speed and a 3000ms integration time. The vertical bars and the black dots at the top reflect density fluctuations. The distribution array S = [S1, S2, …, sn] is generated by counting the first 6140 angle sequences. The standard deviation δs quantifies uniformity: a small δS indicates uniform distribution with high non-repeatability, while a large δS indicates uneven distribution with increased repeatability.

[0120] It represents the number of points in the θk sequence in each unit divided into 1 degree within 360 degrees. The resulting array is recorded as S = [S1, S2, S3, ..., Sn]. The standard deviation of the array S is:

[0121]

[0122] So far, we have obtained the uniformity of the point cloud distribution under a certain fixed speed state (r1, r2 are determined) and within a certain fixed time (integration time is determined, 3000ms). s The smaller the value, the better the uniformity of the point cloud and the greater the non-repeatability, and vice versa, the greater the repeatability.

[0123] Determine the r1 speed, change the r2 speed, and then plot the point cloud uniformity value (δ s ) The cluster of curves that changes with r2 is as follows Figure 8 As shown in the figure, selecting the working point where all curves in the spectrum are at valleys will produce a point cloud that is uniformly encrypted over time. Selecting peaks will produce repetitive point clouds, while selecting short valleys and long peaks will produce a fast, line-by-line scanning point cloud.

[0124] After local magnification Figure 9 As shown, the curve details are clearer, the boundaries between peaks and valleys are more obvious, and the short integration time curve (such as 100ms) shows a sharp rise at the peak, reflecting the significant impact of speed fine-tuning on early distribution, while the long integration time curve (such as 3000ms) tends to be flat at the valley, indicating the stabilizing effect of time accumulation on the uniformity of the point cloud. The zoomed-in view highlights the key points within a specific speed range. For example, the valley overlap area is suitable for the global uniform mode, the peak area supports the repeated scanning mode, and the transition area of ​​short valleys and long peaks optimizes the row-by-row filling effect, further verifying the flexibility of the patented method to achieve multi-mode adaptation by dynamically adjusting r2. Selecting the peak value can obtain a repetitive point cloud, and selecting a short valley time and a long peak time can obtain a fast row-by-row scanning point cloud.

[0125] To ensure scanning stability in highly dynamic working environments, this embodiment includes a pre-calculated gyroscopic effect compensation step. Gyroscopic effects in a composite axis rotation system can cause equipment vibration, especially at high rotation speeds (e.g., r2 > 8000 rpm), leading to point cloud distortion. To this end, processor 103 sets the compensation portion's rotational speed to satisfy ra / rb = J2 / J1 based on the pre-calculated moment of inertia ratio J1 / J2, and controls the operation of third motor 6 via a fixed drive signal to offset the effect.

[0126] The combined rotation of the main shaft and high-speed shaft produces a feed-forward effect. This embodiment uses counter-rotation compensation: J1 is the inertia of the high-speed motor and prism, and J2 is the inertia of the compensation unit (third motor 6 and counterweight 7). The ratio J1 / J2 is precalculated (e.g., 1.2), resulting in rb = 1.2ra. Open-loop control is achieved using fixed signals, eliminating the need for real-time sensor fusion.

[0127] The advanced nature of this scanning method lies in its high flexibility and task adaptability, achieved through two pre-generated adjustments: aperiodic perturbation and grid structure adjustment. These adjustments are based on pre-calculated mathematical models and ensure that the system can dynamically optimize point cloud generation in different scenarios (such as unmanned forklift navigation, obstacle avoidance, and cargo identification) to improve coverage, uniformity, and task adaptability. The following details the implementation mechanisms of these two adjustments:

[0128] First, non-periodic disturbance amount: In order to break the potential long-term periodicity, the processor 103 pre-generates a set of non-periodic disturbance sequences through an algorithm before operation and stores them in a configuration file. Figure 10 The generation process is shown: parameter initialization, setting the maximum perturbation amplitude (such as 0.01) and irrational constant (such as π), pre-calculating the perturbation value, and calculating frac for each scan cycle i. i =modf(i·π / 100)[0], where,

[0129] frac i : The fractional part of the i-th scanning period, used as the intermediate value for generating the disturbance sequence.

[0130] i: scan cycle index, a positive integer (eg, i=1, 2, 3, ...), indicating consecutive scan iterations.

[0131] π / 100: A constant based on the irrational number π (approximately 3.14159), divided by 100 to obtain a smaller value.

[0132] i·π / 100: The product of the period index i and the constant π / 100 generates a linearly growing sequence with a non-repeating fractional part.

[0133] [0]: Extract the fractional part from the modf output and ignore the integer part.

[0134] The sequence is stored in the processor 103 and loaded into the speed ratio model in a periodic order during operation, and the speed of the high-speed motor 52 is adjusted by a fixed instruction. The specific formula example is as follows:

[0135] For period i=1, frac1=modf(0.0314)=0.0314, disturbance1=0.01·(2·0.0314-1)≈-0.00937

[0136] For period i=2, frac2=modf(0.0628)=0.0628, disturbance2≈-0.00874.

[0137] For period i=3, frac3=modf(0.0942)=0.0942, disturbance3≈-0.00812.

[0138] …

[0139] In this embodiment, modf is a mathematical function

[0140] Used to decompose a floating point number into an integer part and a decimal part. It returns two values: the decimal part and the integer part.

[0141] Input: 0.0314 (i.e., π / 100≈0.0314)

[0142] Output: modf(0.0314) = (0.0314, 0)

[0143] Decimal part: 0.0314 (because the input value is less than 1, there is no integer part).

[0144] Integer part: 0.

[0145] frac1=modf(0.0314)[0]=0.0314

[0146] This step ensures that the perturbation value is based on the non-periodic nature of irrational numbers.

[0147] The generated disturbance sequence (e.g., [0.00937, 0.00874, 0.00812, ...]) is used to fine-tune the speed ratio. For example, if the target ratio T1 / T2 = 4.314 and the actual ratio is 4.32 due to fluctuations, the deviation Δ = 4.32 - 4.314 = 0.006. Deviation Δ = 4.32 - 4.314 = 0.006. Combined with the disturbance value disturbance1 = -0.00937, the adjustment G = 0.5 Δ + disturbance1 = 0.5 0.006 - 0.00937 = 0.003 - 0.00937 = -0.00637. Assuming the initial speed r2 = 6123 rpm, the ratio adjustment is converted to a speed change:

[0148] Assuming the initial speed r2 = 6123 rpm, the proportional adjustment is converted into speed change:

[0149] Δr2=-r2·GT1 / T2=≈6123·0.001477≈0.904rpm

[0150] r2′=6123+0.904≈6123.904rpm

[0151] The simulation results show that the sample perturbation sequence is [0.00937, 0.00874, 0.00812, ...], which ensures that the ratio fluctuations are small but non-periodic, avoiding potential cycles such as in non-repeating patterns.

[0152] Secondly, the grid structure adjustment amount: when the task requirements change, the processor 103 loads the pre-stored speed ratio configuration file according to the external instruction, and adjusts the ratio coefficient to change the scan line intersection angle α. Figure 11 The adjustment process is demonstrated: the grid structure adjustment amount controls the scanning line intersection angle by dynamically adjusting the period ratio coefficient O to adapt to different task requirements (such as global navigation, obstacle avoidance or cargo identification of unmanned forklifts). Among them, T1 is the rotation period of the first execution unit (spindle motor), T2 is the scanning period of each prism of the second execution unit (high-speed motor), the ratio T1 / T2=N+O, N is a positive integer, and O is a small irrational number. The processor 103 stores a variety of preset configuration files, each corresponding to a different grid type: acute-angle grid (high-density local encryption), right-angle grid (balanced coverage) or obtuse-angle grid (large-scale rough scanning). When triggered by an external instruction, the processor loads the configuration file, adjusts the high-speed motor speed r2, and changes the period ratio T1 / T2 to achieve the target intersection angle. The intersection angle α is determined by the geometry of the laser beam optical path, and the formula is:

[0153] is the period ratio, and n is the number of prisms. Bidirectional laser arrangement ( Figure 3) Two laser beams in opposite directions generate a cross grid, ensuring that the object boundary is scanned from two directions, overcoming the problem of missed detection in one-way parallel scanning. The cross angle changes with the scale as follows:

[0154] Small ratios (such as T1 / T2 < 4) generate sharp angle grids, which are suitable for high-density local scanning. The deflection frequency of the laser beam in the vertical direction increases, the scanning lines form denser intersections in space, and the angle α becomes smaller (acute angle). This is consistent with the bidirectional laser arrangement ( Figure 3 ), the two laser beams in opposite directions form a sharper crossing trajectory, which is suitable for local encryption.

[0155] A moderate ratio (such as T1 / T2≈4) generates a rectangular grid with balanced coverage. The scanning frequency of the laser beam in the vertical and horizontal directions is balanced, and the bidirectional laser beam ( Figure 3 ) to form orthogonal trajectories. Compared with unidirectional parallel scanning, this can evenly cover the object boundary from two directions, reducing missed detections (such as shelf edges).

[0156] Large ratio (such as T1 / T2>4) generates obtuse grid, which is suitable for fast large-scale scanning. The laser beam scans at a lower frequency in the vertical direction, and the laser beam forms a wider cross angle in space. Bidirectional laser arrangement ( Figure 3 ) makes the scan lines intersect at a larger angle, forming a coarse grid, which is suitable for quickly covering a large field of view (such as unmanned forklift navigation).

[0157] like Figure 12 The adjustment process of this embodiment is as follows:

[0158] Receive external instructions and determine the target intersection angle. For example, an acute angle of 45° is used for local encryption (such as tray jack alignment), a right angle of 90° is used for obstacle avoidance, and an obtuse angle of 120° is used for large-scale navigation.

[0159] Find the configuration file and select the corresponding cycle ratio.

[0160] For example, a sharp-angle grid corresponds to T1 / T2=2.5, a right-angle grid corresponds to T1 / T2≈4, and a blunt-angle grid corresponds to T1 / T2=5.5.

[0161] Calculate the new speed r2. The cycle formula is:

[0162]

[0163] Example: If r1 = 3000 rpm, n = 4, target T1 / T2 = 2.5 (acute angle)

[0164] but:

[0165] The processor 103 instructs the high-speed motor 52 to run at a new speed r2, supporting real-time switching between different grid types.

[0166] Based on a combination of the aforementioned methods, including non-integer scale motion models, histogram statistical uniformity quantification, gyroscopic effect compensation, aperiodic perturbation sequence generation, and bidirectional cross-grid control, processor 103 is able to efficiently control the 3D lidar to implement a variety of advanced scanning modes. These modes are seamlessly switched through pre-calculated mathematical models and open-loop control mechanisms, optimizing point cloud generation to adapt to dynamic environments. Figure 8 The scanning mode switching process is demonstrated, highlighting how the processor loads configuration files based on external instructions (such as task signals sent by an unmanned forklift control system), enabling flexible transitions from global coverage to local encryption. This mode switching not only improves the system's task adaptability but also reduces computational load and power consumption, supporting low-power devices such as battery-powered unmanned forklifts.

[0167] Based on a preset mathematical model and characteristic spectrum (a periodic waveform curve of δS changing with speed), the processor 103 controls the speed r2 of the high-speed motor 52 to generate three main point cloud shapes: uniform scanning, line-by-line scanning, and repeated scanning. Each mode corresponds to a different integral time curve analysis:

[0168] Global Uniform Mode: The processor loads a uniform scan profile and selects all speed points in the feature map where the integral-time curves coincide with valleys, generating a point cloud that gradually increases in density over time. This uniformly increases point cloud density, avoiding localized clustering or sparseness, making it suitable for wide-area environmental perception. For example, when an unmanned forklift performs path planning, uniform mode provides full field of view coverage, reducing δS to below 0.3.

[0169] Local Densification Mode: Load local weighted perturbation profiles to increase the point cloud density in the target area through non-periodic perturbation sequences (such as disturbance i =0.01·(2·frac i -1) Weighted rotational speed adjustment allows focus on specific areas (such as cargo pallets). This increases point cloud density by 20% in the target area, making it ideal for cargo identification or alignment tasks.

[0170] Line-by-line scanning mode: Select the speed points (such as r2≈6200rpm) of the short integral curve peak (sparse filling) and the long integral curve valley (gradually uniform), and fill the point cloud line by line along a specific direction.

[0171] Repeated scan mode: Select the speed point where the curve coincides with the peak (such as r2≈6500rpm), overlap the scan lines, and generate a stable grid, which is suitable for static calibration.

[0172] The configuration file stores specific control parameters, including speed values, integral time curves, and disturbance sequences. It is quickly loaded according to external instructions (such as task signals), shortening the task response time to milliseconds and significantly improving system efficiency.

[0173] In the application scenario of the present invention on an unmanned forklift, a 3D laser radar is installed on the vehicle body for environmental perception.

[0174] Global uniform mode: used for large-scale path planning and autonomous navigation, selecting the valley speed to generate a uniform point cloud;

[0175] Orthogonal Obstacle Avoidance Mode: For approaching shelves or avoiding dynamic obstacles, switches to a rectangular grid profile;

[0176] Local Encryption Mode: Used for cargo identification or fork alignment on pallet sockets, loads sharp-angle grid profiles to improve local resolution.

Claims

1. A 3D laser radar scanning method, applied to a 3D perception system comprising a first execution unit and a second execution unit capable of relative motion, wherein the method generates a non-repeatedly scanned 3D point cloud based on a non-integer ratio motion relationship between the two execution units, characterized in that: The method comprises: a) establishing a mathematical model describing the relationship between the rotational speed relationship between the first execution unit and the second execution unit and the final point cloud spatial angular distribution, wherein the first execution unit is a spindle execution unit with a fixed rotational speed r1, and the second execution unit is a high-speed execution unit with an adjustable rotational speed r2; based on the mathematical model, obtaining a distribution array S by performing histogram statistics on the point cloud spatial angular distribution, and calculating the standard deviation δ of the distribution array S s , to quantify the spatial distribution uniformity of the point cloud; b) Repeat the calculation of step a) for a series of different second execution unit motor speeds r2, thereby generating the uniformity index δ s A characteristic spectrum that changes with the speed r2 of the motor of the second execution unit; c) analyzing the feature maps based on an expected point cloud shape, and selecting a set of motor speeds corresponding to the expected point cloud shape, wherein the expected point cloud shape includes a uniformly scanned point cloud, a line-by-line scanned point cloud, or a repeatedly scanned point cloud; d) instructing the first execution unit and the second execution unit to operate at the motor speed corresponding to the optimal operating point, wherein the speed r1 of the first execution unit remains unchanged, and the expected point cloud shape is achieved by adjusting the speed r2 of the second execution unit, and supporting real-time adjustment of the speed during operation to switch between different point cloud shapes.

2. The method according to claim 1, characterized in that Calculate the rotation period of the first execution unit and the second execution unit; establish a mathematical model based on the rotation speeds r1 and r2, associate the output angle with the rotation speeds r1 and r2, and recursively generate a sequence f(n) representing the distribution of all scan line angles; perform histogram statistics on the sequence f(n) to obtain a distribution array S, and calculate the standard deviation δ of the distribution array S s , to quantify the spatial distribution uniformity of the point cloud.

3. The method according to claim 2, characterized in that The calculation of the rotation period of the first execution unit and the second execution unit specifically includes: calculating the rotation period T1=60 / r1 of the first execution unit, where r1 is the main shaft motor speed; calculating the scanning period T2=60 / (nr2) of each prism of the second execution unit, where n is the number of prism edges; in the mathematical model, the distribution angle of each prism in space is calculated as θ=mod(360(T2 / T1)*k, 360), where k is an integer index, and all angle distribution sequences f(n) are generated by recursion with a given initial phase; performing histogram statistics on the sequence f(n) to obtain a distribution array S, and calculating the standard deviation δ of the distribution array S s , to quantify the spatial distribution uniformity of the point cloud.

4. The method according to claim 1, wherein According to the rotation speed r2 value and the uniformity index δ s Generate the characteristic map, wherein the characteristic map includes multiple curves corresponding to different integration times, and the curves are periodic waveforms, which are used to identify the rotational speed points at which all curves coincide with valley values ​​to achieve a point cloud that is uniformly encrypted over time, or to identify the rotational speed points at which the short integration curve is at a peak value and the long integration curve is at a valley value to achieve line-by-line scanning of the point cloud.

5. The method according to claim 1, wherein By adjusting the rotation speed r2 of the second execution unit, three scanning states are switched: a repeated scanning point cloud state, in which all scan lines coincide; a uniform non-repetitive scanning point cloud state, in which the point cloud is evenly distributed and gradually encrypted; and a line-by-line scanning point cloud state, in which the point cloud is filled line by line along a specific direction.

6. The method according to claim 5, characterized in that The method for generating the gyroscopic effect by the rotation of the second execution unit includes: calculating the ratio of the moment of inertia J1 of the driving part of the second execution unit to the moment of inertia J2 of the compensation part; adjusting the rotation speed rb of the compensation part in real time according to the ratio so that it satisfies ra / rb=J2 / J1 with the rotation speed ra of the driving part of the second execution unit; and correcting the rotation speed rb in real time by controlling the driving signal to offset the gyroscopic effect.

7. The method according to claim 1, characterized in that The method includes controlling a scanning trajectory to generate a bidirectional cross-slash grid by adjusting a rotational speed ratio between a first execution unit and a second execution unit to replace a unidirectional parallel diagonal scanning, thereby scanning the object boundary from two directions; gradually refining the cross-slash grid as the integration time increases, from a coarse grid state to a dense state, wherein the scan lines cross to fill blank areas; scanning a square object, wherein the cross-slash scanning hits the side boundary of the object from two directions, and gradually encrypting the point cloud coverage under non-repetitive scanning conditions to achieve complete capture of the object boundary.

8. The method according to claim 7, characterized in that The method includes controlling the grid intersection angle by adjusting the ratio of the rotation period T1 of the first execution unit to the effective scanning period T2 of the second execution unit, wherein when the T2 / T1 ratio is small, an acute-angle intersection grid is generated, when the T2 / T1 ratio is moderate, a right-angle intersection grid is generated, and when the T2 / T1 ratio is large, an obtuse-angle intersection grid is generated.

9. The method according to claim 1, characterized in that The step a) further includes: setting the scanning period T1 of the first execution unit and the rotation period T2 of the second execution unit, controlling T1 / T2 to be a non-integer ratio to ensure non-repetitive scanning; setting a target non-integer ratio T1 / T2 = N±0+ε, where N is a positive integer, 0<ε<1, and O is a small irrational number.

10. The method according to claim 1, characterized in that The step c) further comprises: analyzing the uniformity index δ of the characteristic spectrum s The curve is selected to determine the rotation speed r2 corresponding to the expected point cloud shape, wherein when the rotation speed r2 at which all curves coincide with the valley value is selected, a uniform scanning point cloud is generated; when the rotation speed r2 at which the short integration time curve is at the peak value and the long integration time curve is at the valley value is selected, a line-by-line scanning point cloud is generated; and when the rotation speed r2 at which all curves coincide with the peak value is selected, a repeated scanning point cloud is generated.

11. A 3D laser radar comprising: a main scanning unit, configured to perform a first scanning motion with a first motion parameter; an auxiliary scanning unit, configured to perform a second scanning motion with a second motion parameter; a processor electrically connected to the main scanning unit and the auxiliary scanning unit; The processor is configured to: establish a mathematical model describing the relationship between the rotational speed relationship of the main scanning unit and the auxiliary scanning unit and the spatial angle distribution of the point cloud, control the first rotational speed r1 to be fixed, and adjust the second rotational speed r2 to form a non-integer ratio motion relationship; based on the mathematical model, obtain a distribution array S by performing histogram statistics on the spatial angle distribution of the point cloud, and calculate the standard deviation δ of the distribution array S. s For a series of different second speeds r2, the uniformity index δ is repeatedly calculated s , generating δ s A characteristic map that changes with r2; analyzing the characteristic map according to an expected point cloud shape to determine a rotational speed r1 and r2 combination corresponding to a target working point, wherein the expected point cloud shape includes a uniformly scanned point cloud, a line-by-line scanned point cloud, or a repeatedly scanned point cloud; instructing the main scanning unit and the auxiliary scanning unit to operate at a rotational speed corresponding to the target working point, and supporting real-time adjustment of the second rotational speed r2 to switch the point cloud shape.

12. The 3D laser radar according to claim 11, characterized in that The main scanning unit includes a first motor and a main reflection mirror, and the auxiliary scanning unit includes a second motor and a polygonal reflection prism.

13. The 3D laser radar according to claim 11, characterized in that It also includes an inertia compensation unit, which is coaxially arranged with the rotation axis of the auxiliary scanning unit. The processor controls the inertia compensation unit and the auxiliary scanning unit to move at the same angular velocity and in opposite rotation directions.

14. The 3D laser radar according to claim 13, characterized in that The inertia compensation unit includes a third motor and a counterweight.

15. The 3D laser radar according to claim 11, characterized in that The auxiliary scanning unit includes at least two laser transceiver modules, which are arranged in a manner to emit laser beams in opposite directions within the same scanning plane to generate a bidirectional cross-angle grid; the processor controls the grid cross angle to generate an acute angle, right angle or obtuse angle cross grid by adjusting the ratio of the first rotation speed r1 to the second rotation speed r2.

16. The 3D laser radar according to claim 15, characterized in that The laser transceiver module includes a laser emitting part and a laser echo receiving part. At least one internal reflector is provided inside the laser emitting part and / or the laser echo receiving part for folding the optical path so that the laser emitting optical path or the echo receiving optical path is L-shaped or U-shaped.

17. The 3D laser radar according to claim 12, characterized in that The polygonal reflecting prism is a hollow structure, the second motor is an outer rotor motor, and its rotor is arranged inside the hollow polygonal reflecting prism and coaxially connected to it; the main reflecting mirror is configured to support a spatial field of view angle range of -20° to 90° or -45° to 60° through position adjustment.

18. The 3D laser radar according to claim 11, characterized in that The processor stores a plurality of preset scanning mode configuration files, each configuration file corresponds to a point cloud distribution characteristic, and the processor can load different configuration files according to external instructions to switch the scanning mode.

19. An unmanned forklift, comprising a vehicle body, a control system, and a 3D laser radar for environmental perception, characterized in that: The 3D laser radar is the 3D laser radar according to any one of claims 11 to 18, and the control system is configured as follows: When performing large-scale path planning or autonomous navigation tasks, control the 3D lidar to operate in a globally uniform mode; When approaching shelves or avoiding dynamic obstacles, the 3D LiDAR is controlled to switch to orthogonal obstacle avoidance mode. When performing cargo identification or fork alignment with pallet holes, the 3D lidar is controlled to switch to local encryption mode.

Images