Stacking regulation and control method, system and equipment of stacking robot and medium

Abstract

The invention relates to the technical field of stacking, in particular to a stacking regulation and control method, system and equipment of a stacking robot and a medium, and the stacking regulation and control method comprises the steps that system errors are eliminated by collecting and fusing data of multiple sensors, layered dynamic path planning is conducted to generate the optimal track of a mechanical arm, the posture of an end effector is updated, and stacking is executed through double-closed-loop control. The method comprises the steps of performing joint self-calibration and clamp self-adaptive adjustment, and realizing digital twin monitoring and group intelligent scheduling, and further comprises a corresponding system, electronic equipment and a computer readable storage medium. The effects of improving the cargo box stacking precision, stability and safety, adapting to the cargo boxes of different sizes, achieving state monitoring, optimizing task allocation and improving the overall operation efficiency and reliability of the stacking robot are achieved.

Description

Technical Field

[0001] This application relates to the field of stacking technology, and in particular to a stacking control method, system, equipment and medium for a stacking robot. Background Technology

[0002] In the field of modern logistics and warehousing automation, stacking robots are increasingly widely used, with their core function being the automated stacking of goods. However, stacking robots currently face numerous technical challenges in actual operation. As logistics and warehousing scenarios become more complex and the types of goods more diverse, higher demands are being placed on the positioning accuracy, path planning flexibility, and load balancing capabilities of stacking robots.

[0003] In existing technologies, stacking robots mostly use a single sensor for positioning, path planning relies on preset programs, gripper sizes are fixed and require manual adjustment, control commands are centrally processed by a central controller, and multi-robot collaboration is mainly based on simple division of labor. Some robots only obtain environmental data through LiDAR, which is susceptible to positioning deviations caused by temperature changes; the path algorithm cannot be adjusted in real time according to the specifications of the cargo boxes, resulting in low stacking efficiency for multi-sized goods; grippers need to be manually changed to adapt to different sized cargo boxes, which takes several minutes; when the central controller processes all data, the communication latency is high, making it difficult to meet the requirements of high-speed operation; multi-robot collaboration lacks global optimization, the proportion of idle travel distance is high, and equipment status monitoring relies on manual inspection, resulting in delayed fault warnings.

[0004] However, existing technologies still have some shortcomings. In terms of positioning, they suffer from insufficient positioning accuracy, rigid path planning, and an inability to adapt to the high-precision stacking of multi-sized containers in complex environments. In terms of fixtures, mechanical errors accumulate significantly after long-term operation, fixture adjustments are time-consuming and labor-intensive, and system maintenance costs are high. In terms of equipment status, there is a lack of equipment status monitoring, weak fault prediction capabilities, low efficiency of multi-machine collaboration, and a high rate of unexpected downtime. Summary of the Invention

[0005] To address the issues of insufficient positioning accuracy, rigid path planning, and unstable contact force when stacking fragile containers in traditional stacking robots, this application provides a stacking control method, system, equipment, and medium for stacking robots.

[0006] Firstly, this application provides a stacking control method for a stacking robot, employing the following technical solution: A stacking control method for a stacking robot includes the following steps: S1: Collect and fuse data from multiple sensors. Acquire cargo and environmental data through lidar, vision system and inertial measurement unit. Input cargo and environmental data into Kalman filter for weighted fusion to eliminate system errors; S2: Layered dynamic path planning. The bottom layer generates the optimal trajectory of the robotic arm based on the DQN reinforcement learning neural network, and the upper layer updates the planned end effector posture according to the set posture update time threshold. S3: Dual closed-loop control executes stacking, with the position loop controlling the servo motor step angle and the force control loop adjusting the clamping force to stabilize the contact force of fragile cargo boxes; S4: Joint self-calibration and fixture adaptive adjustment, real-time detection of joint transmission error and automatic compensation, adaptive adjustment of variable configuration fixture to adapt to different sized boxes. S5: Digital twin monitoring and swarm intelligence scheduling. It builds a virtual image based on equipment operation data to achieve status monitoring and uses an improved ant colony algorithm to optimize the allocation of handling tasks for each stacking robot.

[0007] By adopting the above technical solutions, data from multiple sensors is collected and fused. Data from LiDAR, vision systems, and inertial measurement units is acquired and weighted by a Kalman filter to establish an error compensation model, eliminating systematic errors such as LiDAR temperature drift and vision lens distortion, thus improving positioning accuracy. Layered dynamic path planning uses a DQN reinforcement learning neural network at the bottom layer to generate the optimal trajectory for the robotic arm, while the upper layer periodically updates the end effector's posture, solving the problem of rigid path planning. Dual closed-loop control stacking stabilizes the robotic arm's position by controlling the servo motor's step angle, while the force control loop adjusts the clamping force to stabilize the contact force with vulnerable cargo boxes. Joint self-calibration and gripper adaptive adjustment detect and automatically compensate for joint transmission errors in real time, and adaptively adjusts variable-configuration grippers to fit cargo boxes of different sizes, solving the problems of accumulated mechanical errors and the need for frequent manual adjustments to grippers. Digital twin monitoring and swarm intelligent scheduling build a virtual mirror based on equipment operation data to achieve status monitoring, and an improved ant colony algorithm optimizes task allocation, solving the problems of missing equipment status monitoring, weak fault prediction capabilities, and low efficiency in multi-machine collaboration.

[0008] Optionally, the acquisition and fusion of multiple sensor data includes: The lidar uses time-of-flight ranging to generate millimeter-level point cloud data. This data includes information on the cargo box dimensions, spatial location, surface features, center of gravity, corner coordinates, and relative distances between the cargo box and surrounding obstacles. The vision system extracts key points on the cargo box edges using a feature matching algorithm. The inertial measurement unit outputs acceleration and angular velocity compensation signals at a preset frequency to obtain three-channel data. This three-channel data is then input into a Kalman filter, and a weighted fusion model is used to establish an error compensation model. This model eliminates systematic errors caused by lidar temperature drift and vision lens distortion, outputting high-precision relative position information between the cargo box and the robotic arm.

[0009] By adopting the above technical solution, the lidar uses the time-of-flight ranging principle to generate millimeter-level point cloud data containing multi-faceted information about the cargo box. The vision system extracts key points on the edge of the cargo box through a feature matching algorithm. The inertial measurement unit outputs acceleration and angular velocity compensation signals according to a preset frequency to obtain three-channel data. The three-channel data is input into a Kalman filter for weighted fusion to establish an error compensation model, which can eliminate systematic errors such as lidar temperature drift and vision lens distortion. Finally, it outputs high-precision relative position information between the cargo box and the robotic arm, solving the problem of insufficient positioning accuracy in traditional stacking robots. This provides a precise spatial positioning basis for subsequent layered dynamic path planning and dual closed-loop control, ensuring high precision in robotic arm operation.

[0010] Optionally, the hierarchical dynamic path planning includes: The bottom layer, based on a DQN reinforcement learning neural network, receives millimeter-level point cloud data of the cargo box and stacking layer parameters. The hidden layer generates an initial trajectory by calculating the collision probability and the torque load of the robotic arm. The output layer outputs the optimal trajectory containing the joint angles of the robotic arm with 6 degrees of freedom. The upper layer, a real-time trajectory optimization module, receives feedback on the current cargo box position and robotic arm posture at regular intervals according to a set posture update time threshold, and fine-tunes the posture of the end effector to achieve dynamic execution trajectory adaptation to cargo box specifications and stacking layer parameters.

[0011] By adopting the above technical solution, the bottom layer is based on DQN reinforcement learning neural network. It takes millimeter-level point cloud data of the cargo box and stacking layer parameters as input, calculates the collision probability and the torque load of the robot arm through the hidden layer, and outputs the optimal trajectory of the robot arm joint angle with 6 degrees of freedom, which can generate the basic motion framework of the robot arm. The upper layer real-time trajectory optimization module updates the posture according to the set posture update time threshold, receives the current position of the cargo box and the posture feedback of the robot arm at regular intervals, and fine-tunes the posture of the end effector. It can realize the dynamic execution trajectory and adapt to the cargo box specifications and stacking layer parameters, solve the problem of rigid path planning in traditional stacking robots, and adapt to the high-precision stacking of multi-specification cargo boxes in complex environments.

[0012] Optionally, the dual closed-loop control execution stack includes: The position loop uses a high-resolution encoder to detect the step angle of the servo motor in real time, and controls the step angle value of the servo motor to be within the set step angle threshold, so that the motion position accuracy of the robotic arm is stable. The force control loop integrates a six-axis torque sensor to collect clamping force data in real time and dynamically adjusts the drive current to control the contact force of stacking fragile boxes within a safe range.

[0013] By adopting the above technical solutions, the position ring uses a high-resolution encoder to detect the step angle of the servo motor in real time and control it within a set threshold, which can keep the position accuracy of the robotic arm stable; the force control ring integrates a six-axis torque sensor to collect clamping force data in real time and dynamically adjust the drive current, which can regulate the contact force when stacking fragile boxes within a safe range, solving the problem of unstable contact force when stacking fragile boxes in traditional stacking robots.

[0014] Optionally, the joint self-calibration and fixture adaptive adjustment include: Each rotary joint has a built-in reference grating ruler, which, together with a high frame rate industrial camera, forms a closed-loop detection system. When the detected joint transmission error exceeds the set action threshold, the micro-stepping motor of the harmonic reducer is triggered to perform compensation calibration, and the calibration time is controlled within the calibration threshold. The variable configuration fixture is based on the parallelogram linkage principle and the set width threshold, and the ball screw is driven by a servo motor to perform stepless width adjustment. According to the material of the cargo box, the pressure sensing array of the clamping surface adjusts the distribution of contact pressure.

[0015] By adopting the above technical solutions, the rotary joint incorporates a built-in reference grating ruler and a high frame rate industrial camera to form a closed-loop detection system. This system can detect joint transmission errors in real time. When the error exceeds the action threshold, it triggers the micro-stepping motor of the harmonic reducer for compensation and calibration, and controls the calibration time within the calibration threshold. This system can automatically compensate for joint transmission errors in a timely manner, reduce the accumulation of mechanical errors, and improve positioning accuracy and equipment stability. The variable configuration fixture is based on the parallelogram linkage principle and width threshold. It uses a servo motor to drive a ball screw for stepless width adjustment, which can adaptively adjust the fixture width to fit different sized boxes without the need for frequent manual adjustments, thus improving work efficiency. According to the material of the box, the pressure sensor array on the clamping surface adjusts the contact pressure distribution, which can adjust the contact pressure for different material boxes to avoid damaging the boxes.

[0016] Optionally, the digital twin monitoring includes: The system retrieves real-time operational data from 5G edge computing nodes, including joint torque of the robotic arm, current of the servo motor, vibration spectrum of the mechanical structure, joint rotation angle, and operating temperature. The 5G edge computing nodes encrypt the collected real-time operational data and upload it to the cloud platform. The cloud platform then constructs a digital twin virtual image that is identical to the physical stacking robot based on the real-time operational data. The cloud platform uses a physics engine to simulate the stress distribution under actual working conditions, and sends an early warning message to the operation and maintenance center when structural fatigue risk is detected.

[0017] By adopting the above technical solutions, 5G edge computing nodes can collect real-time operational data from multiple dimensions, including robotic arm joint torque and servo motor current, and upload it to the cloud to build a digital twin virtual image, enabling real-time monitoring of the physical stacking robot's status. The cloud platform calls a physics engine to simulate stress distribution, and when structural fatigue risk is detected, it can send early warning information to the operation and maintenance center to help operation and maintenance personnel intuitively locate problems, improve the response speed and accuracy of preventive maintenance, detect potential equipment failures in advance, reduce unexpected downtime, and lower equipment maintenance costs.

[0018] Optionally, the swarm intelligent scheduling includes: obtaining constraint parameters for each stacking robot based on an improved ant colony optimization algorithm, specifically including: remaining battery power, current position, and load capacity; generating a globally optimal handling sequence based on the constraint data, and then dynamically allocating handling tasks; deploying a pruned convolutional neural network on the embedded device for rapid fault diagnosis; automatically reducing speed when an abnormal vibration mode is detected, and uploading a diagnostic report.

[0019] By adopting the above technical solution, the remaining power, current position, and load capacity of each stacking robot are obtained as constraint parameters. Based on the improved ant colony optimization algorithm, a globally optimal handling sequence is generated and handling tasks are dynamically allocated. This optimizes the allocation of handling tasks for each stacking robot, reduces the idle travel distance, achieves load balance among robots, and avoids the impact of a single device's malfunction on the overall progress. By deploying a pruned convolutional neural network on the embedded end, faults can be quickly diagnosed. When an abnormal vibration mode is detected, the system automatically slows down and uploads a diagnostic report, which can reduce further wear on faulty components and promptly notify maintenance personnel for targeted repairs.

[0020] Secondly, this application provides a stacking control system for a stacking robot, which adopts the following technical solution: A stacking control system for a stacking robot includes: The data acquisition and fusion module is used to acquire cargo box and environmental data. The Kalman filter eliminates system errors based on the acquired cargo box and environmental data. The hierarchical dynamic path planning module is used to obtain the optimal trajectory of the robotic arm and periodically update the planned end effector posture. The dual closed-loop control execution stacking module is used to control the step angle of the servo motor and adjust the clamping force to stabilize the contact force of the fragile cargo box; The joint self-calibration and fixture adaptive adjustment module is used to detect joint transmission errors and automatically compensate for them, as well as to adaptively adjust the variable configuration fixture to fit different sized boxes. The digital twin monitoring module is used to build a virtual image to achieve status monitoring; The swarm intelligence scheduling module is used to optimize the allocation of handling tasks among the stacking robots.

[0021] Thirdly, this application provides a computer electronic device that features stable transmission of encrypted data.

[0022] The above-mentioned objective of this application is achieved through the following technical solution: An electronic device includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed using the aforementioned data encryption transmission method.

[0023] Fourthly, this application provides a computer-readable storage medium capable of storing corresponding programs, which facilitates stable transmission of encrypted data.

[0024] The fourth objective of this application is achieved through the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and executed by any of the above-described data encryption transmission methods.

[0025] In summary, this application has the following beneficial effects: 1. By collecting and fusing data from multiple sensors, hierarchical dynamic path planning, dual closed-loop control for stacking, joint self-calibration and gripper adaptive adjustment, as well as digital twin monitoring and swarm intelligent scheduling, this technology solves the problems of insufficient positioning accuracy, rigid path planning, unstable contact force when stacking fragile boxes, accumulation of mechanical errors, frequent manual adjustment of grippers for boxes of different sizes, lack of equipment status monitoring, weak fault prediction capability, and low efficiency of multi-machine collaboration in traditional stacking robots. 2. Data is collected by lidar, vision system and inertial measurement unit, and an error compensation model is established by weighted fusion through Kalman filter to eliminate systematic errors such as lidar temperature drift and vision lens distortion, and output high-precision relative position information of the cargo box and the robotic arm to improve the positioning accuracy of the stacking robot; 3. The bottom layer generates the optimal trajectory of the robotic arm based on the DQN reinforcement learning neural network, and the upper layer updates the planned end effector posture in a timely manner according to the set posture update time threshold, so as to realize the dynamic execution trajectory and the adaptation of the cargo box specifications and stacking layer parameters, making the path planning more flexible. 4. The position loop controls the step angle of the servo motor, and the force control loop adjusts the clamping force to stabilize the contact force of the fragile cargo box, ensuring the accuracy of the robotic arm's movement position and the stability of the contact force when the fragile cargo boxes are stacked; 5. Real-time detection and automatic compensation of joint transmission errors; adaptive adjustment of variable configuration fixtures to fit different sized boxes; reduction of mechanical error accumulation; improvement of fixture adjustment efficiency; and reduction of system maintenance costs. 6. A virtual image is built based on equipment operation data to realize status monitoring. An improved ant colony algorithm is used to optimize the allocation of handling tasks for each stacking robot, realize equipment status monitoring and fault prediction, improve multi-machine collaboration efficiency, and reduce unexpected downtime rate. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a stacking control method for a stacking robot disclosed in an embodiment of this application.

[0027] Figure 2 This is a schematic diagram of a stacking control system for a stacking robot disclosed in an embodiment of this application.

[0028] Figure 3 This is a schematic diagram of a device according to one embodiment of this application. Detailed Implementation

[0029] The present application will be further described in detail below with reference to the accompanying drawings.

[0030] This application discloses a stacking control method for a stacking robot, see [link to relevant documentation]. Figure 1 This includes the following steps: S1: Collect and fuse data from multiple sensors. Acquire cargo and environmental data through lidar, vision system and inertial measurement unit. Input cargo and environmental data into Kalman filter for weighted fusion to eliminate system errors; Specifically, the collection and fusion of data from multiple sensors includes: The lidar system, based on the time-of-flight ranging principle, emits laser signals into the cargo box and its surrounding environment. By calculating the round-trip time of the laser, it generates millimeter-level point cloud data, fully presenting the spatial shape of the cargo box and the distribution of environmental obstacles. The vision system captures images of the cargo box using an industrial camera and employs a feature matching algorithm to identify and extract key points on the cargo box's edges, accurately locating the cargo box's position and contour in the two-dimensional image. The inertial measurement unit outputs acceleration and angular velocity compensation signals in real time at a preset frequency ranging from 180 to 220 Hz, reflecting the dynamic motion state of the robotic arm. The three-channel data obtained from the above steps include millimeter-level point cloud data, key points on the cargo box edges, and acceleration and angular velocity compensation signals.

[0031] In this embodiment, the millimeter-level point cloud data includes information on the cargo box dimensions, spatial location, surface features, center of gravity, corner coordinates, and relative distances between the cargo box and surrounding obstacles. Key points on the cargo box edges include corner points and edges. The inertial measurement unit outputs acceleration and angular velocity compensation signals in real time at a high frequency of 200Hz, reflecting the dynamic motion state of the robotic arm, including its vibration and offset states. The feature matching algorithm uses the SIFT algorithm; in other embodiments, the ORB algorithm may be used.

[0032] S11: Perform preliminary processing on the three-channel data collected by the three types of sensors. One specific implementation method is as follows: Remove environmental noise points from the millimeter-level point cloud data of the LiDAR, such as those caused by dust and light interference, and align the coordinate system of the millimeter-level point cloud data with the base coordinate system of the robotic arm.

[0033] The distortion model preset by the lens intrinsic parameters is retrieved, and the distortion correction is performed on the key points of the cargo box edge extracted by the vision system to eliminate image distortion caused by the optical characteristics of the lens.

[0034] High-frequency vibration noise, such as jitter noise generated by the movement of robotic arm joints, is filtered out from the acceleration and angular velocity compensation signals output by the inertial measurement unit, thereby ensuring the stability of the compensation signals.

[0035] S12: Based on the three-channel data, perform weighted fusion of the three-channel data. One specific implementation method is: The preprocessed three-channel data are simultaneously input into a Kalman filter. The Kalman filter dynamically assigns weights based on the characteristics of the three types of sensors: LiDAR, vision system, and inertial measurement unit. For example, it increases the weight of vision data in stable lighting conditions, increases the weight of inertial measurement unit data in fast-moving conditions, and increases the weight of LiDAR point cloud data in complex environments. Then, through weighted calculation, it achieves preliminary fusion of multi-dimensional data to form a unified dataset of environment and cargo container status.

[0036] During the fusion process, an error compensation model is simultaneously established based on iterative calculations using the Kalman filter.

[0037] When the lidar experiences measurement drift due to temperature changes, the offset of the millimeter-level point cloud data is corrected by comparing the dynamic acceleration signal of the inertial measurement unit with the static characteristics of the millimeter-level point cloud data.

[0038] When optical distortion of the vision lens is addressed, the 3D coordinates of the millimeter-level point cloud data from the LiDAR are combined to reverse-calibrate the positions of key points on the cargo box edge extracted by the vision system, eliminating the deviation in the mapping from 2D image to 3D space. Simultaneously, the high-frequency compensation signal from the inertial measurement unit is used to correct the time synchronization error between the LiDAR and vision systems caused by differences in sampling frequencies. Through this error compensation model, the systematic errors of the three types of sensors are completely eliminated.

[0039] S13: After weighted fusion and error compensation of the three-channel data, the system finally outputs high-precision relative position information between the cargo box and the robotic arm. One specific implementation method is as follows: The high-precision relative position information between the cargo box and the robotic arm includes the cargo box's three-dimensional coordinates in the robotic arm's coordinate system, the cargo box's attitude angles, and the relative distance between them. This provides a precise spatial positioning basis for subsequent hierarchical dynamic path planning and dual closed-loop control, ensuring high precision in robotic arm operation. The attitude angles include pitch, yaw, and roll angles.

[0040] The above scheme collects and fuses data from multiple sensors. It uses LiDAR, vision system and inertial measurement unit to acquire data and then fuses it through Kalman filter weighting. This can establish an error compensation model, eliminate systematic errors such as LiDAR temperature drift and vision lens distortion, and improve positioning accuracy.

[0041] S2: Layered dynamic path planning. The bottom layer generates the optimal trajectory of the robotic arm based on the DQN reinforcement learning neural network, and the upper layer updates the planned end effector posture according to the set posture update time threshold. Specifically, hierarchical dynamic path planning includes: The bottom layer, based on a DQN reinforcement learning neural network, receives millimeter-level point cloud data of the cargo box and stacking layer parameters. The hidden layer generates an initial trajectory by calculating the collision probability and the robot arm torque load. The output layer outputs the optimal trajectory containing the robot arm joint angles of 6 degrees of freedom. The upper layer, a real-time trajectory optimization module, receives feedback on the current cargo box position and robot arm posture at set attitude update time thresholds, and fine-tunes the end effector posture to adapt the trajectory to the cargo box specifications and stacking layer parameters. One specific implementation method is as follows: S21: Preprocessing and integrating the underlying input data to obtain millimeter-level point cloud data and stacking layer parameters of the cargo box.

[0042] Preprocessing is performed on two types of data: millimeter-level point cloud data and stacking layer parameters. Key features are extracted from the millimeter-level point cloud data, including the cargo box's center of gravity, corner coordinates, and relative distances between the cargo box and surrounding obstacles. The stacking layer parameters are converted into quantization constraints, including the maximum permissible speed and minimum safe distance for high-level stacking. The two types of data are then integrated to obtain structured data adapted to the input layer of the underlying DQN reinforcement learning neural network. The stacking layer parameters include information on the target stacking height and the current number of stacked layers.

[0043] S22: The hidden layer generates the initial trajectory by calculating the collision probability and the torque load of the robotic arm, and the output layer outputs the optimal trajectory containing the joint angles of the robotic arm with 6 degrees of freedom.

[0044] Specifically, the preprocessed structured data is input into the DQN reinforcement learning neural network. The input layer receives the structured data and passes it to the hidden layer. The hidden layer completes two core calculations based on the stacking scenario experience learned during network training.

[0045] First, collision probability calculation is performed by retrieving a preset environmental obstacle model, which includes information on stacked boxes and shelf structures. By comparing the millimeter-level point cloud data of the boxes with the environmental obstacle model, the collision risk of the robotic arm with surrounding objects under different motion trajectories is predicted. Second, torque load calculation is performed. Combining the rated torque of each joint of the robotic arm, the arm length parameter, and the weight of the cargo box, the torque load under different joint angle combinations is evaluated to determine whether it is within the safe threshold and to avoid overload and wear of the mechanism.

[0046] Based on the collision probability and torque load results calculated by the hidden layer, the output layer selects the joint angle combination with the lowest collision probability and the most reasonable torque load from the preset candidate set of 6-DOF robotic arm joint angles, and obtains the initial optimal trajectory by combining the action and reward learning mechanism of the DQN reinforcement learning neural network.

[0047] The action and reward learning mechanism includes efficient and safe trajectory-corresponding high-reward rules learned through training. The candidate set includes multiple motion angle datasets such as robot arm base rotation, shoulder pitch, upper arm rotation, elbow pitch, forearm rotation, and wrist adjustment. The optimal trajectory specifically includes the complete joint motion sequence of the robot arm from its current position to the target stack position.

[0048] S23: The upper-level real-time trajectory optimization module receives feedback on the current cargo box position and robotic arm posture at regular intervals according to the set posture update time threshold, and fine-tunes the posture of the end effector to achieve the adaptation of the dynamic execution trajectory to the cargo box specifications and stacking layer parameters.

[0049] Specifically, the attitude update time threshold ranges from 40 to 50 ms. In this embodiment, the attitude update time is set to 50 ms.

[0050] The upper-level real-time trajectory optimization module continuously receives two types of real-time feedback data every 50ms: current cargo box position feedback and robotic arm posture feedback. The current cargo box position feedback specifically includes the offset of key points on the cargo box edge captured in real-time by the vision system and the positional changes in the millimeter-level point cloud data updated by the LiDAR, reflecting the minute displacements of the cargo box during handling. The robotic arm posture feedback specifically includes real-time joint angle readings from the high-resolution encoder and dynamic posture changes output by the inertial measurement unit, reflecting the deviation between the robotic arm's current actual motion state and the underlying planned trajectory.

[0051] Based on the feedback data collected every 50ms, the upper-level module analyzes the deviation between the current trajectory and the target requirements.

[0052] Specifically, if a slight displacement of the cargo box due to handling vibration is detected, such as a positional error exceeding ±0.5mm, or if the robotic arm's posture exhibits angular deviation due to joint clearances, such as a wrist pitch angle deviation exceeding 0.1°, the end effector's posture is fine-tuned according to the cargo box specifications and stacking layers. This adapts to the cargo box's dimensions and material rigidity, as well as the stable vertical posture required for high-level stacking, achieving stable handling and stacking of the cargo box. Specifically, by correcting the wrist and elbow joint parameters in the 6-DOF joint angles, it ensures that the contact position and clamping angle between the end effector and the cargo box always adapt to the cargo box's characteristics, and that the motion trajectory avoids collision risks caused by cargo box displacement or slight robotic arm wobbling. Elbow joint parameters include fine-tuning the wrist rotation angle to align with the cargo box edge and adjusting the elbow pitch angle to compensate for height deviations.

[0053] S24: The bottom and upper layers work together to output the final dynamic execution trajectory. The bottom layer, based on the DQN reinforcement learning neural network, generates the initial optimal trajectory to provide the basic motion framework for the robotic arm. The upper layer module fine-tunes every 50ms to correct local deviations in real time, adapting to the dynamically changing cargo box state and robotic arm posture. Together, they form the final dynamic execution trajectory, which is then sent to the dual closed-loop control module to drive the robotic arm to complete high-precision stacking operations. The dynamic execution trajectory includes a continuous sequence of 6-DOF joint angle changes, covering the entire motion process from grasping to stacking, and includes real-time responses to differences in cargo box specifications and stacking layer requirements.

[0054] S3: Dual closed-loop control executes stacking, with the position loop controlling the servo motor step angle and the force control loop adjusting the clamping force to stabilize the contact force of fragile cargo boxes; The dual closed-loop control execution stacking specifically includes: The system receives dynamic execution trajectory data. The position loop uses a high-resolution encoder to detect the servo motor step angle in real time, controlling the servo motor step angle value to remain within a set step angle threshold, thus ensuring stable positional accuracy of the robotic arm. The force control loop integrates a six-axis torque sensor to collect clamping force data in real time and dynamically adjusts the drive current to control the contact force of stacked fragile cargo boxes within a safe range. One specific implementation method is as follows: S31: Core parameters such as preset step angle threshold, contact force of vulnerable cargo boxes, sampling frequency of the six-axis torque sensor, and sensitivity parameters for drive current adjustment. One specific implementation method includes: In the position loop, the step angle threshold of the servo motor is set to 0.05°, and the detection accuracy parameters of the high-resolution encoder are entered.

[0055] In the force control loop, a safe range of contact force for vulnerable cargo boxes is set, with the contact force value ranging from 5 ± 0.2 N.

[0056] The sampling frequency and sensitivity parameters of the six-axis torque sensor are preset. The sampling frequency ranges from 0.5 to 1 kHz. In this embodiment, the sampling frequency is set to 1 kHz to ensure real-time performance.

[0057] The sensitivity parameter represents the correlation between changes in current and changes in clamping force. Simultaneously, the system calibrates the synchronization mechanism between the position loop and the force control loop to ensure that the timing of data acquisition and control commands is consistent.

[0058] S32: When the robotic arm performs stacking operations, the position ring continuously works to perform real-time detection and adjusts the step angle according to the detected structure.

[0059] Specifically, the position loop operates continuously when the robotic arm performs the stacking action. A high-resolution encoder detects the step angle value of each servo motor in real time and feeds the detected value back to the control unit in real time.

[0060] The control unit compares the current step angle with the preset step angle threshold. If the step angle deviation exceeds the step angle threshold, it immediately sends a fine-tuning command to the servo motor.

[0061] By adjusting the motor's drive pulse signal, the rotation of the motor rotor is controlled, and the step angle is corrected back to the step angle threshold. This process is continuously repeated to ensure that the movement angles of each joint of the robotic arm are precisely controllable, ultimately enabling the end effector to stably reach the target stacking position.

[0062] S33: The force control ring collects clamping force data in real time and transmits it to the control unit. The control unit adjusts the current of the gripper drive motor based on the clamping force data. One specific implementation method is as follows: During the process of the robotic arm gripping the cargo box and performing stacking, the force control loop is activated simultaneously. The six-axis torque sensor collects the gripping force data between the end effector and the cargo box in real time, focusing on extracting the contact force value perpendicular to the cargo box surface, and transmitting it to the control unit in real time.

[0063] The control unit compares the collected contact force with the preset safe range of contact force for vulnerable cargo boxes: If the contact force is less than 4.8N, the current of the drive motor is increased to enhance the gripping force of the grippers and improve the contact force, thus preventing the cargo box from slipping off.

[0064] If the contact force is higher than 5.2N, reduce the drive current and decrease the gripping force of the grippers to prevent damage to the fragile cargo box.

[0065] During the adjustment process, the system dynamically adjusts the rate of change of current according to the material of the cargo box. For example, a smoother current adjustment is used for a paper cargo box to avoid instantaneous force fluctuations and ensure that the contact force is stable within a safe range.

[0066] S34: One implementation method for the position loop and force control loop to coordinate the stacking action is as follows: When the robotic arm moves above the target stacking position, the position loop prioritizes the control of the end effector to precisely align with the cargo box placement point, and the step angle adjustment ensures the positional accuracy in both the horizontal and vertical directions.

[0067] When the end effector contacts the cargo box and begins to place it, the force control ring takes the lead in adjustment, controlling the clamping force through real-time force value feedback. This avoids sudden changes in force value caused by position fine-tuning. For example, during the descent of the robotic arm, when the position ring fine-tunes the height, the force control ring synchronously compensates for the clamping force to prevent excessive instantaneous pressure.

[0068] The position ring and force control ring interact internally to ensure accurate positioning and safe and stable contact force. For example, the height signal of the position ring will trigger the force control ring to increase the force detection frequency.

[0069] S35: After stacking is completed, the status is confirmed to prepare for the next stacking action. At the same time, the position accuracy and force fluctuation data of this stacking are recorded for subsequent optimization of system parameters.

[0070] Specifically, after the cargo box is placed in place, the dual closed-loop execution system performs final status checks. The position loop confirms via encoder that the robotic arm joint angle has stabilized at the target position, while the force control loop confirms via torque sensor that the clamping force has decreased to the safe release threshold, such as below 1N, thus preventing the cargo box from being pulled out during release. After confirmation, the system outputs a stacking completion signal, preparing for the next stacking action, and simultaneously records the positional accuracy and force fluctuation data of this stacking operation for subsequent system parameter optimization.

[0071] S4: Joint self-calibration and fixture adaptive adjustment, real-time detection of joint transmission error and automatic compensation, adaptive adjustment of variable configuration fixture to adapt to different sized boxes. Joint self-calibration and fixture adaptive adjustment specifically include: Each rotary joint incorporates a built-in reference grating ruler, forming a closed-loop detection system with a high-frame-rate industrial camera. When the detected joint transmission error exceeds a set action threshold, the micro-stepping motor of the harmonic reducer is triggered for compensation calibration, with the calibration time controlled within the calibration threshold. The variable configuration fixture, based on the parallelogram linkage principle and a set width threshold, uses a servo motor to drive a ball screw for stepless width adjustment. The pressure distribution on the clamping surface is adjusted according to the material of the cargo box by a pressure sensing array. One specific implementation method is as follows: S41: One implementation method for joint self-calibration is: A reference grating ruler for each rotary joint measures the actual rotation angle of the joint in real time, while a high-frame-rate industrial camera synchronously captures the displacement of specific marker points of the joint movement. The data from both are fused to form a closed-loop detection system.

[0072] The target joint angle of the path planning module is retrieved, the current actual rotation angle is compared with the target joint angle, the angle deviation caused by the wear of the harmonic reducer is calculated, and the joint transmission error is obtained.

[0073] When a joint transmission error exceeding the action threshold of 0.5mm is detected, the system immediately pauses the current task and triggers the calibration procedure. The microstepper motor of the harmonic reducer starts and performs reverse compensation based on the direction and magnitude of the error; for example, if the error is 0.3° clockwise, the microstepper motor rotates 0.3° counterclockwise. During the calibration process, the grating ruler and camera continuously provide real-time position feedback, forming a dynamic adjustment closed loop.

[0074] The motion threshold ranges from 0.3 to 0.5 mm. In this embodiment, the motion threshold is 0.5 mm.

[0075] The acceleration / deceleration curves and feedback frequency of the microstepper motor were optimized, with the feedback frequency increased to 1kHz. This allowed the entire calibration process to be completed within 3 seconds, achieving highly efficient calibration. After calibration, the system automatically verifies whether the error has dropped below the threshold. If it does, normal operation resumes; otherwise, calibration continues or an anomaly is reported to the operations and maintenance center.

[0076] S42: One implementation method for adaptive adjustment of the clamp is: The system retrieves key point data from the edge of the cargo box. Based on the cargo box width data from these key points, a servo motor drives a ball screw to rotate, causing the gripper jaws, designed based on the parallelogram linkage principle, to synchronously open and close within a width threshold for stepless width adjustment. In this embodiment, the width threshold ranges from 50 to 300 mm. During adjustment, a position encoder provides real-time feedback on the gripper position, ensuring an accuracy of ±0.1 mm.

[0077] The system acquires material sensing and pressure distribution data. When the grippers contact the cargo box, a pressure sensing array on the gripping surface collects contact pressure data in real time. The pressure sensing array includes thin-film pressure sensors distributed on the gripper surface. Based on preset material parameters, the system dynamically adjusts the drive current at each sensing point, thereby changing the contact pressure distribution. The material parameters include different pressure thresholds corresponding to cardboard boxes, wooden boxes, and plastic boxes.

[0078] Depending on the material parameters, for fragile cardboard boxes, increase the contact area and reduce single-point pressure, concentrating the pressure on the edges of the box; for rigid plastic boxes, allow for higher local pressure while maintaining overall pressure balance.

[0079] S43: To perform collaborative work, after receiving a new stacking task, the system first automatically adjusts the clamp width to the target value according to the cargo box information, and presets the corresponding pressure distribution mode.

[0080] During the movement of the robotic arm, if the joint self-calibration detects a joint transmission error, the system synchronously corrects the target position of the gripper to ensure that the relative position of the gripper and the cargo box is not affected by the joint error.

[0081] When the clamp contacts the cargo box, the force control ring and the position ring work together. The position ring ensures that the gripper reaches the target width, and the force control ring finely adjusts the gripper posture through the pressure sensor array to achieve precise control of the contact pressure.

[0082] If an abnormal situation of pressure change exceeding the width adjustment limit is detected during the adjustment process, the system will immediately trigger the safety mechanism, stop the current action, release the clamp, and issue an alarm.

[0083] The system records the width value, pressure distribution curve parameters, and execution time for each adjustment. By analyzing historical data through machine learning algorithms, it optimizes future adjustment strategies, such as predictive width adjustment and adaptive pressure threshold, thereby improving the smoothness and efficiency of the work.

[0084] S5: Digital twin monitoring and swarm intelligence scheduling. It builds a virtual image based on equipment operation data to achieve status monitoring and uses an improved ant colony algorithm to optimize the allocation of handling tasks for each stacking robot.

[0085] Digital twin monitoring includes: The system retrieves real-time operational data from 5G edge computing nodes, collecting data across multiple dimensions, including robotic arm joint torque, servo motor current, vibration spectrum of the mechanical structure, joint rotation angle, and operating temperature. The 5G edge computing nodes encrypt the collected real-time operational data and upload it to a cloud platform. The cloud platform then constructs a digital twin virtual image of the physical stacking robot based on this data. The cloud platform uses a physics engine to simulate stress distribution under actual working conditions, and sends an early warning to the operations and maintenance center when structural fatigue risk is detected. One specific implementation method is as follows: S51: Retrieve 5G edge computing nodes and collect real-time operational data from multiple dimensions.

[0086] 5G edge computing nodes are deployed at key locations on the stacking robot, and these nodes are directly connected to the robot arm's sensors. The 5G edge computing nodes collect over 30 dimensions of data in real time during the robot arm's operation, including information on joint torque, servo motor current, vibration spectrum of the mechanical structure, joint rotation angle, and operating temperature.

[0087] During the data collection process, the low latency of the 5G network ensures smooth data transmission from the sensor to the edge node. At the same time, the TSN (Time Sensitive Network) is used to synchronize the data of each node in time, ensuring the time consistency of multi-dimensional data.

[0088] Key components include the rotary joints, drive motors, and end effectors. Sensors for the nodes and robotic arm include torque sensors, current sensors, and vibration sensors. The vibration spectrum of the mechanical structure is acquired to reflect the wear condition of the components.

[0089] S52: Construct a digital twin virtual image consistent with the physical stacking robot.

[0090] The 5G edge computing node encrypts the real-time operational data collected from multiple dimensions and uploads it to the cloud platform. The cloud platform then constructs a 1:1 digital twin virtual image that is completely identical to the physical stacking robot based on the real-time operational data from multiple dimensions.

[0091] The virtual mirror accurately replicates the physical structure, motion characteristics, and real-time operating status of the robotic arm. Each virtual component in the mirror is bound to a corresponding component on the physical robot. Any slight fluctuation in joint torque or increase in motor current on the physical robot is mapped to the virtual mirror in real time, achieving a dynamic correlation between physical and virtual movements. The physical structure includes information on joint connection methods, harmonic reducer parameters, and gripper dimensions. Motion characteristics include information on joint rotation range and speed limits. Real-time operating status includes information on current joint angles and load weight.

[0092] S53: Performs physics engine simulation of working conditions and stress distribution analysis. The cloud platform calls the physics engine to simulate the stress distribution of the robotic arm under actual working conditions based on the real-time status of the virtual image.

[0093] Specifically, the simulation calculates the tooth surface stress of the harmonic reducer based on joint torque data, analyzes the resonance risk of the robotic arm frame using vibration spectrum analysis, and assesses the load strength of the drive components based on motor current changes. During the simulation, the physics engine continuously compares the current stress value with a preset safety threshold, tracking the stress accumulation trend under continuous high load operation. The physics engine includes simulation tools based on finite element analysis to simulate the stress distribution of the robotic arm under actual operating conditions. The safety threshold includes the fatigue limit of the joint transmission components and the rated load range of the motor.

[0094] S54: Structural fatigue risk detection and early warning triggering.

[0095] When the simulation results from the physics engine show that the stress value of a certain part is close to or exceeds the safety threshold, such as when the joint transmission error is about to exceed 0.5mm due to stress accumulation, or when abnormal peaks indicating component wear appear in the vibration spectrum, it is judged as a structural fatigue risk. At this time, the system automatically triggers the early warning mechanism, generates early warning information including the risk location, risk level, and recommended maintenance measures, and sends it to the monitoring platform of the operation and maintenance center in real time via the 5G network. At the same time, the risk location is highlighted in the virtual image to help operation and maintenance personnel intuitively locate the problem. The risk levels include low risk, medium risk, and high risk. Low risk reports predict that the accuracy may decrease after 10-15 days, medium risk reports predict that the accuracy may decrease within 24 hours, and high risk reports predict that the accuracy may decrease within 1 hour.

[0096] After receiving the early warning information, the operations and maintenance center can quickly formulate a maintenance plan based on the virtual image simulation maintenance scheme. After the maintenance is completed, the 5G edge computing node continuously collects real-time operational data from multiple dimensions after the robotic arm resumes operation, uploads it to the cloud to update the virtual image, and the physical engine re-evaluates the stress distribution after the repair, continuously optimizing the early warning threshold and maintenance strategy, thereby improving the response speed and accuracy of preventive maintenance.

[0097] Furthermore, swarm intelligence scheduling includes: Based on an improved ant colony optimization algorithm, constraint parameters for each stacking robot are obtained, including remaining battery power, current position, and load capacity. A globally optimal handling sequence is generated based on the constraint data, and handling tasks are then dynamically allocated. On the embedded system, a pruned convolutional neural network is deployed for rapid fault diagnosis. When an abnormal vibration pattern is detected, the robot automatically reduces its speed and uploads a diagnostic report.

[0098] S55: Real-time acquisition and integration of constraint parameters for generating the globally optimal handling sequence. One method the system uses a Time-Sensitive Network (TSN) to acquire the constraint parameters of all stacker robots in real time is as follows: 1) Remaining power: Provided by the battery management system of each robot, presented as a percentage, reflecting the robot's ability to continue operating; 2) Current position: The robot is located by fusing LiDAR and vision system to output three-dimensional coordinates and determine its real-time position in the warehouse space; 3) Load Capacity: Calculated by combining the robot arm's rated load with the weight of the currently held cargo box, the remaining load capacity is expressed as the remaining weight that can be carried, ensuring that task allocation does not exceed the robot's load-bearing limit. All parameters are preprocessed by 5G edge computing nodes to remove outliers and standardize the data format before being aggregated to the group scheduling center.

[0099] The remaining load capacity is calculated as follows: Rated load - Current load = Remaining load capacity.

[0100] S56: One way to generate the globally optimal transport sequence based on an improved ant colony optimization algorithm is as follows: The scheduling center invokes an improved ant colony optimization algorithm to calculate the integrated constraint parameters with the optimization objectives of minimizing total handling time, minimizing idle travel distance, and balancing the load of each robot.

[0101] The algorithm simulates the path selection mechanism of ants foraging, treating each transport task as a food source and the robot as an ant. The current position determines the basic distance from the robot to the task point, while the remaining battery power determines the feasibility of the path. Robots with low battery power are prioritized for tasks closer to the target location. The number of cargo boxes that can be transported in a single trip is limited by the robot's load capacity.

[0102] Then, the pheromone concentration is dynamically adjusted, and the pheromone weight is increased for the paths assigned to tasks with high completion efficiency, guiding the algorithm to converge quickly to the global optimum, similar to the path executed by a short-distance and high-power robot.

[0103] Then, the output is a globally optimal handling sequence that includes the task execution order and the matching relationship between the robot and the task. For example, robot A first moves the boxes on shelf 1 to the stacking area, and then assists robot B in moving the boxes on shelf 3, so as to maximize the overall operation efficiency.

[0104] S57: One implementation method for dynamically assigning handling tasks to each robot by the scheduling center based on the globally optimal handling sequence is as follows: Initial allocation is performed, and specific tasks are assigned to the corresponding robots according to the globally optimal handling sequence, clarifying the task start point, end point and time node, just like grabbing a 5-layer box from a shelf and stacking it to the 3rd layer of stacking area C.

[0105] Real-time adjustments are made. If a robot's constraint parameters suddenly change, such as a sudden drop in remaining battery power to 10% or a minor malfunction causing a decrease in load capacity, the ant colony optimization algorithm immediately recalculates the globally optimal transport sequence and assigns the unfinished tasks to other robots in better condition, such as those with sufficient battery power and redundant remaining load capacity, thus preventing the overall progress from being affected by a single robot's malfunction. Task distribution is synchronized at the microsecond level via the TSN network, ensuring that all robots receive new instructions simultaneously and avoiding task conflicts.

[0106] S58: On the embedded side, a pruned convolutional neural network is deployed for rapid fault diagnosis. One implementation method for rapid fault diagnosis is: Each robot's embedded terminal deploys a pruned convolutional neural network, occupying ≤2MB of memory. It enables real-time fault monitoring and data acquisition. Vibration signals are continuously collected by vibration sensors installed at the robot arm joints, generating a vibration spectrum data segment every 1ms. The convolutional neural network extracts features from the vibration spectrum and completes fault diagnosis in a short time. If an abnormal vibration pattern is detected, an early warning is immediately triggered, and the robot automatically reduces its speed to minimize further wear on faulty components. A detailed diagnostic report is then sent to the dispatch center and operations and maintenance center via the 5G network. The dispatch center adjusts task allocation accordingly and notifies maintenance personnel to perform targeted repairs.

[0107] After each task is completed, the system records the actual transport time, idle distance, and failure rate, compares this data with the optimal transport sequence predicted by the ant colony optimization algorithm, and calculates the deviation rate. Machine learning algorithms are used to analyze the causes of deviations, dynamically optimizing and improving the parameters of the ant colony algorithm and the fault diagnosis model of the convolutional neural network, thereby improving the efficiency and reliability of swarm scheduling.

[0108] By adopting the above scheme, data from multiple sensors are collected and fused. Data from LiDAR, vision system and inertial measurement unit are acquired and then weighted and fused by Kalman filter. An error compensation model can be established to eliminate systematic errors such as LiDAR temperature drift and vision lens distortion, thereby improving positioning accuracy.

[0109] The hierarchical dynamic path planning uses a DQN reinforcement learning neural network at the bottom layer to generate the optimal trajectory for the robotic arm, and the upper layer updates the end effector posture periodically, thus solving the problem of rigid path planning.

[0110] The dual closed-loop control system executes stacking, with the position loop controlling the servo motor step angle to stabilize the robotic arm's position accuracy and the force control loop adjusting the clamping force to stabilize the contact force of the fragile cargo box.

[0111] The joint self-calibration and fixture adaptive adjustment system detects and automatically compensates for joint transmission errors in real time. The adaptive adjustable variable configuration fixture adapts to different sized boxes, solving the problems of mechanical error accumulation and the need for frequent manual adjustment of the fixture.

[0112] Digital twin monitoring and swarm intelligence scheduling, based on equipment operation data, constructs a virtual image to achieve status monitoring. An improved ant colony algorithm is used to optimize task allocation, solving the problems of missing equipment status monitoring, weak fault prediction capabilities, and low efficiency of multi-machine collaboration.

[0113] Secondly, this application provides a stacking control system for a stacking robot, which adopts the following technical solution: A stacking control system for a stacking robot, see [link / reference] Figure 2 ,include: The data acquisition and fusion module is used to acquire cargo box and environmental data. The Kalman filter eliminates system errors based on the acquired cargo box and environmental data. The hierarchical dynamic path planning module is used to obtain the optimal trajectory of the robotic arm and periodically update the planned end effector posture. The dual closed-loop control stacking module is used to control the step angle of the servo motor and adjust the clamping force to stabilize the contact force of the fragile cargo box.

[0114] The joint self-calibration and fixture adaptive adjustment module is used to detect joint transmission errors and automatically compensate for them, as well as to adaptively adjust the variable configuration fixture to fit different sized boxes. The digital twin monitoring module is used to build a virtual image to achieve status monitoring; The swarm intelligence scheduling module is used to optimize the allocation of handling tasks among the stacking robots.

[0115] Thirdly, this embodiment also discloses an electronic device. An electronic device includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as in any of the methods described above.

[0116] The electronic device can be a desktop computer, a laptop computer, or a cloud server, and includes, but is not limited to, a processor and a memory. For example, the electronic device may also include input / output devices, network access devices, and buses.

[0117] The processor in this application may include one or more processing cores. The processor executes instructions, programs, code sets, or instruction sets stored in memory, and calls data stored in memory to perform various functions and process data. The processor may be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), controller, microcontroller, and microprocessor. It is understood that, for different devices, the electronic devices used to implement the above-described processor functions may also be other types, and the embodiments of this application do not specifically limit this.

[0118] The memory can be an internal storage unit of an electronic device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) equipped on the electronic device. Furthermore, the memory can be a combination of an internal storage unit and an external storage device. The memory is used to store computer programs and other programs and data required by the electronic device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0119] Fourthly, this embodiment also discloses a computer-readable storage medium. A computer-readable storage medium stores a computer program that can be loaded by a processor and executed as in any of the methods described above.

[0120] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0121] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete the above description.

Claims

1. A method for stack regulation of a stacking robot, characterized by: The method comprises the following steps: S1: collecting and fusing multiple sensor data, obtaining the box and environment data through laser radar, vision system and inertial measurement unit, inputting the box and environment data into Kalman filter for weighted fusion to eliminate system error; S2: hierarchical dynamic path planning, the bottom layer generates the optimal trajectory of the mechanical arm based on the DQN reinforcement learning neural network, and the upper layer updates the end effector pose at a set pose update time threshold; S3: double closed loop control execution stacking, the position loop controls the step angle of the servo motor, and the force control loop adjusts the clamping force to stabilize the contact force of the fragile box; S4: joint self-calibration and clamp adaptive adjustment, real-time detection of joint transmission error and automatic compensation, adaptive adjustment of variable configuration clamp to adapt to different size boxes; S5: digital twin monitoring and swarm intelligence scheduling, based on the device operation data to build a virtual mirror to realize state monitoring, and using improved ant colony algorithm to optimize the allocation of each stacking robot handling task.

2. The stacking robot's stacking regulation method according to claim 1, characterized in that: The collecting and fusing multiple sensor data comprises: The laser radar generates millimeter level point cloud data using time of flight ranging principle, which contains information of box size, box spatial position, box surface feature, box center of gravity, box corner coordinate and relative distance between box and surrounding obstacles; the vision system extracts box edge key points through feature matching algorithm, and the inertial measurement unit outputs acceleration and angular velocity compensation signals according to preset frequency to obtain three channel data; The three channel data is input into Kalman filter, an error compensation model is established through weighted fusion to eliminate systematic errors of laser radar temperature drift and vision lens distortion, and high precision relative position information of the box and the mechanical arm is output.

3. The stacking robot and stacking regulation method according to claim 1, characterized in that: The hierarchical dynamic path planning comprises: The input layer of the bottom layer based on DQN reinforcement learning neural network receives millimeter level point cloud data of the box and stacking layer number parameters, the hidden layer generates initial trajectory by calculating collision probability and mechanical arm torque load, and the output layer outputs optimal trajectory containing 6 degrees of freedom of mechanical arm joint angle; The upper layer real-time trajectory optimization module receives current box position and mechanical arm pose feedback at a set pose update time threshold, fine tunes the end effector pose, and realizes the adaptation of dynamic execution trajectory to box specification and stacking layer number parameters.

4. The stacking robot and stacking regulation method according to claim 1, characterized in that: The double closed loop control execution stacking comprises: The position loop detects the step angle of the servo motor in real time through high resolution encoder, controls the step angle value of the servo motor within the set step angle threshold, and makes the motion position accuracy of the mechanical arm stable; The force control loop integrates six-axis torque sensor to collect clamping force data in real time, and dynamically adjusts the driving current to control the contact force of the fragile box stacking within a safe range.

5. The stacking robot method of claim 1, wherein: The joint self-calibration and clamp adaptive adjustment comprises: A reference grating ruler is built in each rotary joint to form a closed loop detection system with a high frame rate industrial camera, when the joint transmission error is detected to exceed the set motion threshold, the micro stepping motor of the harmonic reducer is triggered to compensate and calibrate, and the calibration time is controlled within the calibration threshold; The variable configuration clamp adjusts the width steplessly through the ball screw driven by the servo motor based on the parallelogram linkage principle and the set width threshold; The pressure distribution of the clamping surface is adjusted according to the material of the cargo box.

6. The stacking robot method of claim 1, wherein: The digital twin monitoring includes: Access 5G edge computing nodes to collect real-time operational data from multiple dimensions, including robotic arm joint torque, servo motor current, vibration spectrum of mechanical structure, joint rotation angle, and operating temperature. The 5G edge computing node encrypts the real-time operating data it collects and uploads it to the cloud platform. The cloud platform then builds a digital twin virtual image that is identical to the physical stacking robot based on the real-time operating data. The cloud platform uses a physics engine to simulate stress distribution under actual working conditions. When structural fatigue risk is detected, it sends an early warning message to the operation and maintenance center.

7. The stacking robot method of claim 1, wherein: The swarm intelligence scheduling includes: Based on the improved ant colony optimization algorithm, the constraint parameters of each stacking robot are obtained. The constraint parameters specifically include: remaining power, current position and load capacity. Based on the constraint data, a globally optimal transport sequence is generated, and then transport tasks are dynamically allocated. In embedded systems, a pruned convolutional neural network is deployed for rapid fault diagnosis. When an abnormal vibration pattern is detected, the system automatically reduces its speed and uploads a diagnostic report.

8. A stacking robot system for regulating stacking, characterized by include: The data acquisition and fusion module is used to acquire cargo box and environmental data. The Kalman filter eliminates system errors based on the acquired cargo box and environmental data. The hierarchical dynamic path planning module is used to obtain the optimal trajectory of the robotic arm and periodically update the planned end effector posture. The dual closed-loop control execution stacking module is used to control the step angle of the servo motor and adjust the clamping force to stabilize the contact force of the fragile cargo box; The joint self-calibration and fixture adaptive adjustment module is used to detect joint transmission errors and automatically compensate for them, as well as to adaptively adjust the variable configuration fixture to fit different sized boxes. The digital twin monitoring module is used to build a virtual image to achieve status monitoring; The swarm intelligence scheduling module is used to optimize the allocation of handling tasks among the stacking robots.

9. An electronic device comprising a memory and a processor, said memory having stored thereon a computer program, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images