Special-shaped material high-precision stacking operation control method and system based on PPO algorithm

Abstract

The invention discloses a special-shaped material high-precision stacking operation control method and system based on a PPO algorithm, and the method comprises the following steps: S1, multi-mode sensing data collection and preprocessing: starting an RGB-D camera and a six-axis force control sensor which are mounted at the tail end of a mechanical arm, and a 360-degree laser radar mounted in the middle of a robot chassis, multi-source data timestamp hard synchronization is realized through an industrial-grade synchronization pulse device. The invention relates to the technical field of storage robot control. According to the special-shaped material high-precision stacking operation control method and system based on the PPO algorithm, the target function is replaced by multi-mode sensing fusion and PPO algorithm truncation, the special-shaped material stacking precision is achieved, meanwhile, through dynamic grabbing force adaptation, the material damage rate and the situation of unstable stacking are reduced, and the stacking efficiency is improved. The problem that sensing of the contact force of the special-shaped materials is not enough is thoroughly solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of warehouse robot control, in particular to a high-precision stacking operation control method and system for irregular-shaped materials based on a PPO algorithm. BACKGROUND

[0002] Intelligent warehousing is a link in the logistics process. The application of intelligent warehousing ensures the speed and accuracy of data input in each link of warehouse management, ensures that enterprises can accurately grasp the real data of inventory in a timely manner, reasonably maintains and controls enterprise inventory, and through scientific coding, can also facilitate the management of the batch and shelf life of inventory goods. With the warehouse location management function of the SNHGES system, all inventory goods can be tracked in real time, which is beneficial to improving the work efficiency of warehouse management.

[0003] In the field of intelligent warehousing, the stacking operation of irregular-shaped materials (such as irregular geometric bodies, complex surface texture, and materials with off-center gravity) has always been a technical difficulty. The traditional warehouse robot has the following problems for such materials: 1. Low action exploration efficiency: The traditional mechanical arm relies on a preset path or manual teaching. In the face of the asymmetric structure of irregular-shaped materials, the action parameters need to be repeatedly adjusted, resulting in low operation efficiency and inability to adapt to multi-category irregular-shaped material switching scenarios. 2. Inadequate grasping and stacking precision: Most devices rely only on a single visual sensor for positioning, lack force feedback, and lack contact force perception for irregular-shaped materials. When stacking, the materials cannot be accurately matched in shape, which has limitations. 3. Weak environmental adaptability: In complex warehouse environments such as backlight and dust, the single visual sensor has insufficient recognition accuracy, and cannot accurately obtain the surface texture and three-dimensional structure of irregular-shaped materials, further exacerbating stacking errors.

[0004] Therefore, the present application provides a high-precision stacking operation control method and system for irregular-shaped materials based on a PPO algorithm. SUMMARY

[0005] To overcome the deficiencies of the prior art, the present application provides a high-precision stacking operation control method and system for irregular-shaped materials based on a PPO algorithm, which solves the problems of low action exploration efficiency, inadequate grasping and stacking precision, strategy collapse, and poor adaptability to complex environments for traditional warehouse robots in irregular-shaped material stacking.

[0006] To achieve the above purpose, the present application is implemented by the following technical solution: a high-precision stacking operation control method for irregular-shaped materials based on a PPO algorithm, comprising the following steps: S1, multi-modal perception data acquisition and preprocessing: start the RGB-D camera installed at the end of the mechanical arm, the 6-axis force control sensor, and the 360° laser radar installed in the middle of the robot chassis, realize multi-source data timestamp hard synchronization through an industrial-grade synchronous pulse device, unify all data to the mechanical arm base coordinate system through a self-developed point cloud pixel feature matching algorithm to complete coordinate soft calibration, and then perform Gaussian filtering denoising on the calibrated data; S2, modeling of special-shaped materials and working environment: based on the denoised depth data and texture data, a three-dimensional grid model of the special-shaped material is constructed, the center of gravity coordinates of the material are located, and the vulnerable areas are labeled, and the environment data collected by the laser radar are combined with the tray size to generate an operable area heat map and a physical stable placement candidate area of the working space; S3, action space optimization driven by PPO algorithm: import the special-shaped material stacking expert rules to preliminarily limit the action space of the PPO algorithm, compress the invalid action dimension through action masking technology, retain the action exploration probability, and simultaneously limit the strategy update amplitude through the truncation replacement objective function; S4, power take-off dynamic matching and stacking execution: the mechanical arm moves to the target grabbing point according to the initial grabbing instruction output by the PPO algorithm, the 6-axis force control sensor feedbacks the contact force data in real time, the grabbing force is dynamically matched, the special-shaped material is moved to the stable placement candidate area and the posture is fine-tuned to complete the stacking; S5, stacking feedback and strategy iteration optimization: judge the stability of the stacking, if stable, update the PPO algorithm strategy parameters, if unstable, trigger the punishment mechanism and re-execute S3-S4, and based on batch samples, the algorithm is updated in parallel after a preset number of stacking operations.

[0007] Preferably, the error of the industrial-grade synchronous pulse device in S1 for realizing multi-source data timestamp hard synchronization is ≤1ms.

[0008] Preferably, the self-developed point cloud pixel feature matching algorithm in S1 realizes coordinate soft calibration by extracting the edge and corner features of the RGB image and the topological features of the laser radar point cloud.

[0009] Preferably, the action masking technology in S3 compresses 60% of the invalid action dimension and retains 10% of the action exploration probability; the truncation replacement objective function limits the PPO algorithm strategy update step length to ≤0.2.

[0010] Preferably, the dynamic matching of the grabbing force in S4 is as follows: S4-1, for flexible special-shaped materials, apply a 5-10N wrapping grabbing force, and for rigid special-shaped materials, apply a 15-20N precise clamping force; S4-2, based on the material posture data collected in real time by the RGB-D camera during stacking, fine-tune the joint angle of the mechanical arm, so that the stacking accuracy reaches ±0.1mm.

[0011] The application also discloses a high-precision palletizing operation control system for special-shaped materials based on a PPO algorithm, comprising a mechanical arm execution unit, a multi-modal perception unit, a body control unit and a communication unit. The mechanical arm execution unit adopts a multi-joint mechanical arm with a load of 20-100 kg, and an end integrated clamping mechanism, which is used for performing the actions of grabbing, translating and stacking the special-shaped materials. The multi-modal perception unit comprises two 5 million pixel RGB-D cameras, one 6-axis force control sensor and one 360° laser radar, the two RGB-D cameras are installed at the end of the mechanical arm in an anti-shock manner and are distributed at a preset included angle, the 6-axis force control sensor is integrated between the end of the mechanical arm and the clamping mechanism, and the 360° laser radar is horizontally installed in the middle of the robot chassis through a special bracket. The body control unit comprises an edge computing module and built-in data preprocessing submodules, material modeling submodules, PPO algorithm submodules, prior constraint submodules, geometric constraint submodules and force control adaptation submodules, and each submodule cooperates to realize data processing, modeling, algorithm optimization and force control adaptation. The communication unit realizes the data interaction between the multi-modal perception unit, the body control unit and the mechanical arm execution unit based on an industrial Ethernet at a preset rate.

[0012] Preferably, the special-shaped material palletizing expert rules built in the prior constraint submodule include that the joint angle of the mechanical arm is prohibited from exceeding a limit value and the fragile area of the special-shaped material is prohibited from being grabbed, and the physical stable placement candidate area generated by the geometric constraint submodule is labeled with optimal stacking points, suboptimal stacking points and prohibited placement areas.

[0013] Preferably, the preset rate of the communication unit is 100 Mbps, and the data interaction content includes the perception data collected by the multi-modal perception unit, the control instructions output by the body control unit and the action execution data fed back by the mechanical arm execution unit. Advantages

[0014] The application provides a high-precision palletizing operation control method and system for special-shaped materials based on a PPO algorithm. 1. The high-precision palletizing operation control method and system for special-shaped materials based on a PPO algorithm realize the palletizing precision of special-shaped materials by relying on the truncated replacement target function of multi-modal perception fusion and the PPO algorithm, reduce the material damage rate and the unstable stacking condition through dynamic grabbing force adaptation, and completely solve the problem of lack of contact force perception for special-shaped materials.

[0015] 2、The high-precision palletizing operation control method and system for irregular materials based on the PPO algorithm generate a candidate map of a physically stable placement area according to the shape, size, and operation space limitations of the materials through a geometric constraint algorithm. With the help of action masking technology, the high-dimensional action space of the mechanical arm is compressed by 60%, reducing invalid action exploration and improving operation efficiency. At the same time, 10% of the exploration probability is reserved to allow the system to try better strategies outside the candidate area to adapt to complex operation scenarios such as asymmetric stacking of irregular materials.

[0016] 3、The high-precision palletizing operation control method and system for irregular materials based on the PPO algorithm uses a chassis 360° laser radar to compensate for the performance shortcoming of visual sensors in backlight and dusty environments, maintaining material recognition accuracy above 95%. It can meet the high-concurrency order demand of e-commerce warehouses and adapt to flexible production of irregular parts in the automotive manufacturing field without the need for additional equipment modification due to environmental or scenario changes, reducing warehouse operation costs. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The step flowchart of the high-precision palletizing operation control method for irregular materials of the present application; Figure 2 The multi-modal perception data processing flowchart of the present application; Figure 3 The modeling flowchart of irregular materials and operating environment of the present application; Figure 4 The PPO optimization and palletizing execution flowchart of the present application; Figure 5 The strategy feedback iteration flowchart of the present application.

[0018] In the figure: 1-mechanical arm execution unit, 2-multi-modal perception unit, 3-embodied control unit, 4-communication unit. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0020] Please refer to Figures 1-5 The present application provides a technical solution: A high-precision palletizing operation control method for irregular materials based on the PPO algorithm, comprising the following steps: S1, multimodal perception data acquisition and preprocessing: start the RGB-D camera installed at the end of the mechanical arm, the 6-axis force control sensor and the 360-degree laser radar installed in the middle of the robot chassis, realize the hard synchronization of the time stamp of multi-source data through the industrial level synchronous pulse device, unify all data to the mechanical arm base coordinate system through the self-developed point cloud pixel feature matching algorithm to complete the coordinate soft calibration, and then perform Gaussian filtering denoising on the calibrated data; S2, modeling of special-shaped materials and working environment: based on the denoised depth data and texture data, a three-dimensional grid model of the special-shaped material is constructed, the center of gravity coordinates of the material are located and the vulnerable area is labeled, and the environment data collected by the laser radar is combined with the tray size to generate a heat map of the operable area of the working space and a physical stable placement candidate area; S3, action space optimization driven by PPO algorithm: import the special-shaped material stacking expert rules to preliminarily limit the action space of the PPO algorithm, compress the invalid action dimension through action masking technology, retain the action exploration probability, and simultaneously limit the strategy update amplitude through the truncation replacement objective function; S4, power take-off dynamic matching and stacking execution: the mechanical arm moves to the target grabbing point according to the initial grabbing instruction output by the PPO algorithm, the 6-axis force control sensor feedbacks the contact force data in real time, the grabbing force is dynamically matched, the special-shaped material is moved to the stable placement candidate area and the posture is fine-tuned to complete the stacking; S5, stacking feedback and strategy iteration optimization: judge the stability of the stacking, if stable, update the strategy parameters of the PPO algorithm, if unstable, trigger the punishment mechanism and re-execute S3-S4, and based on batch samples, the algorithm is updated in parallel after a preset number of stacking operations.

[0021] 2 sets of 5 million pixel RGB-D cameras: installed at the end of the mechanical arm in a shockproof manner, distributed at an angle of 15°-30°, covering the key area of the grabbing operation, collecting the surface texture, two-dimensional image and depth data of the special-shaped material, and positioning accuracy up to ±0.05mm; 1 set of 6-axis force control sensor: integrated between the end of the mechanical arm and the clamping mechanism, real-time acquisition of contact force distribution data with a sampling frequency of ≥1kHz, sensing material deformation force or rigid contact force; 360-degree laser radar: installed in the middle of the robot chassis, scanning frequency 10Hz, point cloud density 200 points / °, collecting ground topological structure and environmental obstacle information, making up for the performance short board of the visual sensor in low light and high dust environment; Relying on the truncated replacement objective function of the multimodal perception fusion and PPO algorithm, the special-shaped material stacking accuracy of ±0.1mm level is realized.

[0022] In the embodiment of the application, the error of the industrial level synchronous pulse device in S1 for realizing the hard synchronization of the time stamp of multi-source data is ≤1ms.

[0023] In the embodiment of the application, the self-developed point cloud pixel feature matching algorithm in S1 realizes coordinate soft calibration by extracting the edges and corner features of the RGB image and the topological features of the laser radar point cloud.

[0024] In the embodiment of the application, the action masking technology in S3 compresses 60% of the invalid action dimensions and retains 10% of the action exploration probability; the truncated alternative target function limits the PPO algorithm strategy update step length to be less than or equal to 0.2.

[0025] In the embodiment of the application, the dynamic matching gripping force in S4 is specifically: S4-1, a 5-10N wrapping gripping force is applied to the flexible special-shaped material, and a 15-20N precise clamping force is applied to the rigid special-shaped material; S4-2, the joint angle of the mechanical arm is fine-tuned based on the material posture data collected in real time by the RGB-D camera during stacking, so that the stacking accuracy reaches ±0.1mm.

[0026] The application also discloses a special-shaped material high-precision stacking operation control system based on a PPO algorithm, which comprises a mechanical arm execution unit 1, a multi-modal perception unit 2, a body control unit 3 and a communication unit 4. The mechanical arm execution unit 1 adopts a multi-joint mechanical arm with a load of 20-100kg, and the end is integrated with a clamping mechanism, which is used for executing the actions of grabbing, translating and stacking the special-shaped materials; The multi-modal perception unit 2 comprises two 5 million pixel RGB-D cameras, one 6-axis force control sensor and one 360° laser radar, the two RGB-D cameras are installed at the end of the mechanical arm in an anti-shock manner and are distributed at a preset included angle, the 6-axis force control sensor is integrated between the end of the mechanical arm and the clamping mechanism, and the 360° laser radar is horizontally installed in the middle of the robot chassis through a special support; The body control unit 3 comprises an edge computing module and built-in data preprocessing submodules, material modeling submodules, PPO algorithm submodules, prior constraint submodules, geometric constraint submodules and force control adaptation submodules, and each submodule cooperates to realize data processing, modeling, algorithm optimization and force control adaptation. The communication unit 4 is based on an industrial Ethernet, and realizes data interaction between the multi-modal perception unit, the body control unit and the mechanical arm execution unit at a preset rate.

[0027] The data preprocessing submodule: the multi-modal perception data is subjected to time stamp hard synchronization error ≤1ms, coordinate conversion soft calibration through self-developed point cloud pixel feature matching algorithm unified to the mechanical arm base coordinate system and Gaussian filter denoising; The material modeling submodule: based on the preprocessed RGB image, depth data and laser point cloud, a three-dimensional grid model of the special-shaped material is constructed, and key parameters such as the center of gravity position, surface curvature and vulnerable area are labeled; The PPO algorithm submodule adopts a proximal policy optimization algorithm, supports batch sample parallel updating, and outputs "grabbing / placing" discrete decisions and continuous control instructions such as joint angles and motion speeds. The geometric constraint submodule generates a candidate graph of a physically stable placement area according to the three-dimensional model of the special-shaped material and the size of the tray.

[0028] The force control adaptation submodule dynamically matches the 5-10N package type grabbing force of the flexible special-shaped material and the 15-20N precise clamping force of the rigid special-shaped material according to the contact force data fed back by the 6-axis force control sensor. In the embodiment of the application, the special-shaped material stacking expert rules built in the prior constraint submodule include prohibiting the joint angle of the mechanical arm from exceeding the limit value and prohibiting grabbing the vulnerable area of the special-shaped material, and the physically stable placement candidate area generated by the geometric constraint submodule is marked with the optimal stacking point, the suboptimal stacking point and the prohibited placement area.

[0029] In the embodiment of the application, the preset rate of the communication unit 4 is 100Mbps, and the data interaction content includes the perception data collected by the multi-modal perception unit 2, the control instructions output by the embodied control unit 3 and the action execution data fed back by the mechanical arm execution unit 1.

[0030] Through the PPO algorithm combined with the action masking technology, 60% of the invalid action dimensions are compressed, and the expert prior constraint is matched, so that the action exploration time of the special-shaped material is shortened by 70%, the debugging time of the new type of special-shaped material is reduced from 24 hours to 2 hours or less, and the system is compatible with the 20-100kg load interval, can adapt to flexible and rigid materials simultaneously, and a single robot can complete 120-150 stable stacking times per hour, which is 50% higher than the efficiency of traditional equipment, and is suitable for multi-category special-shaped material switching scenes.

[0031] The chassis 360° laser radar makes up for the performance short board of the visual sensor in the backlight and dust environment, so that the material recognition accuracy is maintained at more than 95%, and if a multi-machine cooperative system is matched, the stacking times of a 4-machine cluster per hour can reach 480-600 times, which can meet the high-concurrent order demand of e-commerce warehousing and adapt to the flexible production of special-shaped parts in the automobile manufacturing field, without the need for additional modification of equipment due to environmental or scene changes, thereby reducing the warehousing operation cost.

[0032] Meanwhile, the contents not described in detail in the specification all belong to the prior art known to those skilled in the art.

[0033] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0034] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, and it is intended that the scope of the application be limited solely by the scope of the appended claims and the equivalents thereof.

Claims

1. A high-precision palletizing operation control method for special-shaped materials based on a PPO algorithm, characterized in that, The method comprises the following steps: S1, multi-modal perception data acquisition and preprocessing: start the RGB-D camera installed at the end of the mechanical arm, the 6-axis force control sensor, and the 360° laser radar installed in the middle of the robot chassis, realize the hard synchronization of the time stamps of multi-source data through an industrial-grade synchronous pulse device, unify all data to the mechanical arm base coordinate system through a self-developed point cloud pixel feature matching algorithm to complete coordinate soft calibration, and then perform Gaussian filtering denoising on the calibrated data; S2, modeling of special-shaped materials and working environment: based on the denoised depth data and texture data, a three-dimensional grid model of the special-shaped material is constructed, the center of gravity coordinates of the material are located, and the vulnerable areas are labeled, and the operable area heat map of the working space and the physical stable placement candidate area are generated in combination with the tray size and the environment data collected by the laser radar; S3, action space optimization driven by PPO algorithm: import the special-shaped material stacking expert rules to preliminarily limit the action space of the PPO algorithm, compress the invalid action dimension through action masking technology, retain the action exploration probability, and simultaneously limit the update amplitude through the truncation replacement target function; S4, power take-off dynamic matching and stacking execution: the mechanical arm moves to the target grabbing point according to the initial grabbing instruction output by the PPO algorithm, the 6-axis force control sensor feeds back the contact force data in real time, the grabbing force is dynamically matched, the special-shaped material is moved to the stable placement candidate area and the posture is fine-tuned to complete the stacking; S5, stacking feedback and strategy iteration optimization: judge the stability of the stacking, if stable, update the strategy parameters of the PPO algorithm, if unstable, trigger the punishment mechanism and re-execute S3-S4, and after a preset number of stacking operations, update the algorithm based on batch samples in parallel.

2. The PPO algorithm-based high-precision control method for the operation of stacking special-shaped materials, according to claim 1, characterized in that: The error of the industrial-grade synchronous pulse device in S1 for realizing the hard synchronization of the time stamps of multi-source data is ≤1ms.

3. The PPO algorithm-based high-precision control method for the operation of stacking special-shaped materials, according to claim 1, characterized in that: The self-developed point cloud pixel feature matching algorithm in S1 realizes the coordinate soft calibration by extracting the edge and corner features of the RGB image and the topological features of the laser radar point cloud.

4. The PPO algorithm-based high-precision control method for the operation of stacking special-shaped materials, according to claim 1, characterized in that: In S3, the action masking technology compresses 60% of the invalid action dimension and retains 10% of the action exploration probability; the truncation replacement target function limits the PPO algorithm strategy update step length to ≤0.

2.

5. The PPO algorithm-based high-precision abnormal-shaped material palletizing operation control method and system according to claim 1, characterized in that: In S4, the dynamic matching of the grabbing force is specifically: S4-1, for flexible special-shaped materials, apply a 5-10N wrapping grabbing force, and for rigid special-shaped materials, apply a 15-20N precise clamping force; S4-2, based on the material posture data collected in real time by the RGB-D camera during stacking, fine-tune the joint angle of the mechanical arm, so that the stacking accuracy reaches ±0.1mm.

6. A high-precision palletizing operation control system for special-shaped materials based on a PPO algorithm, characterized in that, The PPO algorithm-based special-shaped material high-precision stacking operation control method according to any one of claims 1-5 comprises a mechanical arm execution unit (1), a multi-modal perception unit (2), an embodied control unit (3), and a communication unit (4); The mechanical arm execution unit (1) adopts a multi-joint mechanical arm with a load of 20-100kg, and a clamping mechanism is integrated at the end, which is used to execute the grabbing, translation, and stacking actions of special-shaped materials; The multi-modal perception unit (2) comprises two 5 million pixel RGB-D cameras, a 6-axis force control sensor and a 360° laser radar, the two RGB-D cameras are installed at the end of the mechanical arm in an anti-shock manner and are distributed at a preset included angle, the 6-axis force control sensor is integrated between the end of the mechanical arm and the clamping mechanism, and the 360° laser radar is horizontally installed in the middle of the robot chassis through a special support; The body control unit (3) comprises an edge computing module and built-in data preprocessing sub-module, material modeling sub-module, PPO algorithm sub-module, prior constraint sub-module, geometric constraint sub-module and force control adaptation sub-module, and each sub-module cooperates to realize data processing, modeling, algorithm optimization and force control adaptation; The communication unit (4) is based on industrial Ethernet, and data interaction between the multi-modal perception unit, the body control unit and the mechanical arm execution unit is realized at a preset rate.

7. The PPO algorithm-based high-precision control system for the operation of the irregular-shaped material palletizing according to claim 6, characterized in that: The special-shaped material stacking expert rules built in the prior constraint sub-module include prohibiting the joint angle of the mechanical arm from exceeding the limit value and prohibiting grabbing the fragile area of the special-shaped material, and the physical stable placement candidate area generated by the geometric constraint sub-module is labeled with optimal stacking points, suboptimal stacking points and prohibited placement areas.

8. The PPO algorithm-based high-precision control system for the operation of the irregular-shaped material palletizing according to claim 6, characterized in that: The preset rate of the communication unit (4) is 100 Mbps, and the data interaction content includes the perception data collected by the multi-modal perception unit (2), the control instructions output by the body control unit (3) and the action execution data fed back by the mechanical arm execution unit (1).

Images