Safety protection method and system for visual guidance of robot loading and unloading

Abstract

The application relates to a safety protection method and system for visual guidance of robot feeding and discharging, which comprises the following steps: in the process of robot operation, sensor detection data is collected within a preset time period, the behavior state of personnel in the working area is judged based on the sensor detection data, a judgment result is obtained, and the operation parameters of the robot are adjusted according to the judgment result; when an operation request of personnel attempting to open the fence lock is detected, it is judged whether the robot is in a non-operation state, and the authority of the operator is authenticated, and the lock opening control is executed under the condition that the preset condition is met; the trigger signal of the emergency stop button is monitored, and when any emergency stop signal is detected, a global interruption control instruction is immediately issued to terminate the current task of the robot; and various sensor trigger events and corresponding control responses are recorded to generate a safety event log. The application has the effect of improving the safety of man-machine cooperation in the robot feeding and discharging scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial automation and robot control, and in particular to a safety protection method and system for vision-guided robot feeding and unloading. BACKGROUND

[0002] Currently, industrial robots are widely used in manufacturing enterprises, especially in automated feeding and unloading scenarios. With the increasing demand for flexible production and human-robot collaboration, robots are increasingly deployed in work environments that share space with human operators. In such co-domain operation mode, personnel safety protection becomes a key issue in system design and operation management.

[0003] Existing safety protection solutions usually use independent devices such as gratings, radars, physical fences, and emergency stop buttons to monitor and control the work area. For example, gratings are used to form a peripheral warning boundary, which triggers the robot to stop running when the light beam is blocked; radar sensors are used to detect personnel approaching and provide dynamic position monitoring; emergency stop buttons are used as an emergency means triggered by human to terminate robot operation; the fence entrance may be equipped with a lock to prevent personnel from mistakenly entering the dangerous area.

[0004] The existing technical solutions in the above have the following defects: the existing safety control mechanism is mostly single-point response, there is a lack of linkage strategy between sensors, the judgment logic is relatively static, and it often relies on fixed thresholds or manual operation, which is difficult to adapt to complex personnel behavior changes in real time, and therefore there is room for improvement. SUMMARY

[0005] In order to improve the safety of human-robot collaboration in robot feeding and unloading scenarios, the present application provides a safety protection method and system for vision-guided robot feeding and unloading.

[0006] The above invention objectives of the present application are achieved by the following technical solutions:

[0007] A safety protection method for vision-guided robot feeding and unloading, the safety protection method for vision-guided robot feeding and unloading comprising:

[0008] constructing a digital space model of a robot work area, registering fence boundary, sensor deployment position information and safety area constraint conditions;

[0009] acquiring sensor detection data within a preset time period during the robot operation, and judging the behavior state of personnel in the work area based on the sensor detection data to obtain a judgment result, and adjusting the robot operation parameters according to the judgment result, the sensor detection data including grating sensor detection data and radar sensor detection data;

[0010] When detecting an operation request of a person trying to open the fence door lock, it is judged whether the robot is in a non-running state, and the operator is authenticated for authority, and the door lock opening control is executed under the condition that the preset condition is met;

[0011] The triggering signals of the emergency stop buttons are monitored, and when any emergency stop signal is detected, a global interruption control instruction is immediately issued to terminate the current task of the robot;

[0012] Record various sensor trigger events and corresponding control responses, generate a safety event log, and dynamically adjust the risk assessment strategy based on the safety event log.

[0013] By adopting the above technical solutions, by constructing a digital space model of the robot operation area, registering the fence boundary, sensor deployment position information and safety area constraint conditions, the spatial closed management and multi-source perception layout of the robot operation environment can be realized, thereby providing accurate and structured space basis for subsequent risk judgment and safety control strategy. By collecting sensor detection data during robot operation and judging the behavior state of the person, the robot operation parameters are dynamically adjusted according to the judgment result, which can realize the timely identification and response of the risk of personnel intrusion, thereby effectively reducing the collision risk caused by the approach or entry of personnel into the dangerous area. By performing state judgment and authority authentication when detecting an operation request of a person trying to open the fence door lock, it can be ensured that the robot is in a non-running state and the operator has legal authority, thereby preventing security risks caused by accidental operation or unauthorized access. By monitoring the triggering signals of the emergency stop buttons and issuing a global interruption instruction, the robot task can be quickly terminated in an emergency, thereby realizing timely control and safety protection of sudden risks. By recording various sensor trigger events and corresponding control responses, generating a safety event log, and dynamically adjusting the risk assessment strategy based on the safety event log, the adaptive optimization of the safety strategy can be realized, thereby enhancing the dynamic protection ability and continuous improvement ability of the system for different risk situations.

[0014] The present application can be further configured in an example as follows: the construction of the digital space model of the robot operation area, the registration of the fence boundary, the sensor deployment position information and the safety area constraint conditions include:

[0015] Generate the spatial topology structure of the operation area based on the operation site layout diagram, and record the three-dimensional coordinate range of the fence boundary;

[0016] Map the installation position information of the sensor group to the spatial topology structure to form a spatial distribution diagram of the perception device, the sensor group including a grating sensor and a radar sensor;

[0017] Set a multi-level safety control area boundary condition corresponding to the robot motion path, the safety control area includes a dangerous area and a buffer area, for grading personnel close to risk and dynamically adjusting the robot operation strategy.

[0018] By adopting the technical scheme, the spatial model accurately expressing the work scene can be constructed by generating the spatial topology structure based on the work site layout and recording the three-dimensional coordinate range of the fence boundary, thereby improving the accuracy of subsequent perception, path planning and safety area constraint determination; the unified association of sensor data and spatial coordinates can be realized by mapping the deployment information of the grating sensor and the radar sensor to the spatial topology to form a perception device distribution diagram, thereby improving the integration and real-time analysis capability of multi-source perception information; the grading judgment of personnel close to risk and the hierarchical execution of response strategy can be supported by setting the multi-level safety control area boundary condition and dividing the dangerous area and the buffer area, thereby guaranteeing the safety and collaboration of the robot work.

[0019] In an example, the application can be further configured to: the judgment of the personnel behavior state in the work area based on the sensor detection data, obtaining a judgment result, and adjusting the robot operation parameter according to the judgment result include:

[0020] Based on the grating sensor detection data, when it is detected that the blocking signal generated by the grating sensor continuously satisfies a preset time length, it is determined that the personnel enter the dangerous area, and an emergency stop instruction is generated;

[0021] When the radar sensor detects that the personnel are located in the buffer area, the motion trend features of the personnel are extracted, the risk level is calculated based on the motion trend features, and the motion trend features include the relative distance between the personnel and the robot, the motion direction and the approaching speed;

[0022] Switch the robot operation parameter according to the risk level, the robot operation parameter includes the running speed, the joint activity range and the obstacle avoidance strategy.

[0023] By adopting the technical scheme, the emergency stop instruction is generated when the grating sensor blocking signal continuously satisfies the preset time length, the state of the personnel entering the dangerous area can be accurately identified and immediately responded, thereby avoiding the physical collision risk caused by the robot continuing to run in the personnel intrusion state; by extracting the relative distance, the motion direction and the approaching speed of the personnel located in the buffer area, the motion trend features can be realized, thereby providing data support for subsequent risk level judgment and robot strategy adjustment; by switching the robot running speed, the joint activity range and the obstacle avoidance strategy according to the risk level, the potential risk can be actively avoided, thereby realizing the dynamic balance between safety and running efficiency.

[0024] The application can be further configured in an example as follows: the calculating a risk level based on the motion trend feature comprises:

[0025] extracting a historical position sequence and a current speed vector of the person from the motion trend feature as model input data;

[0026] inputting the model input data into a pre-trained trajectory prediction model, predicting a future motion trajectory of the person, and calculating a minimum spatial distance between the future motion trajectory and a current position of the robot and a preset motion path, a predicted intersection time, and a trajectory overlap range;

[0027] generating a corresponding risk level value by using a hierarchical decision rule according to the minimum spatial distance, the predicted intersection time, and the trajectory overlap range, in combination with a current speed and acceleration of the person.

[0028] By using the above technical solution, the historical position sequence and the current speed vector of the person are extracted from the motion trend feature as model input, which can provide a time sequence basis for trajectory prediction, thereby improving the continuity and reliability of the prediction result. By inputting into the pre-trained trajectory prediction model to calculate the future motion trajectory and the minimum spatial distance between the future motion trajectory and the current position of the robot and the path, the predicted intersection time, and the trajectory overlap range, the potential contact risk between the person and the robot can be comprehensively evaluated, thereby realizing the early judgment and intervention of the future conflict scenario. By combining the speed and acceleration of the person with the hierarchical decision rule to generate the risk level value, fine risk stratification can be realized, thereby improving the agility and pertinence of the robot operation parameter adjustment.

[0029] The application can be further configured in an example as follows: the safety protection method for visual guidance of the robot loading and unloading further comprises:

[0030] constructing a training data set based on historical collected personnel position information and behavior sample data, the behavior sample data including a moving path, speed change, and obstacle avoidance behavior of the person in different work scenarios;

[0031] constructing a trajectory prediction model based on a long short-term memory network;

[0032] performing normalization processing on the training data set and inputting into the trajectory prediction model for training, using a loss function minimizing trajectory prediction error for model weight optimization, generating the pre-trained trajectory prediction model, and dynamically adjusting weight parameters and risk level decision thresholds of the pre-trained trajectory prediction model based on trigger frequency and response delay data recorded in the safety event log during running.

[0033] By adopting the technical scheme, the training data set is constructed based on the historical collected position information and behavior samples, the dynamic behavior characteristics of the personnel in the real work scene can be reflected, and the generalization ability of the prediction model is enhanced; the trajectory prediction model is constructed based on the long short-term memory network, the time correlation of the personnel behavior can be effectively captured, and the trajectory prediction accuracy and the complex scene adaptability are improved; the training data is normalized, and the model is optimized by using the loss function of minimizing the prediction error, the generated model has stronger prediction stability, and the accuracy and real-time performance of the risk level evaluation are improved; the model weight and the threshold are adjusted online by combining the safety event log, the continuous adaptive evolution of the prediction mechanism can be realized, and the response ability of the robot system to the complex personnel behavior is improved.

[0034] In an example, the application can be further configured to: the permission authentication of the operator, the door lock opening control includes:

[0035] Obtain the identity information of the operator, and the identity information includes the ID of the work card, fingerprint information and face image data;

[0036] Compare the identity information with the preset permission database, judge whether the operator has the authorized level of opening the fence door lock, and obtain the judgment result;

[0037] When the judgment result is that the identity authentication is passed and the robot is in a non-running state, a door lock opening control instruction is generated to drive the execution mechanism to complete the fence door opening operation.

[0038] By adopting the technical scheme, the ID of the work card, the fingerprint information and the face image data of the operator are obtained, multi-factor identity recognition can be realized, and the security of the operator identity authentication is improved; by comparing with the permission database to judge whether the operator has the opening permission, the personnel without permission can be effectively prevented from operating the fence door lock, and the security risk of illegal intrusion of the robot work area is reduced; by generating the door lock control instruction only when the authentication is passed and the robot is in a non-running state, the operation condition can be strictly limited, and the safety accidents caused by state judgment error or permission bypassing can be prevented.

[0039] In an example, the application can be further configured to: the record of various sensor trigger events and corresponding control responses to generate a safety event log includes:

[0040] Configure the corresponding rules of event trigger type and response action for each type of sensor, and record the event type, occurrence time, trigger sensor identifier, control response content and execution result when the trigger event is detected;

[0041] The trigger event and the control response are stored in a time sequence to obtain storage data, and a multi-dimensional security event index table is constructed based on the storage data, which is used to support security analysis and backtracking query of specific sensors, specific time periods or specific response results.

[0042] By adopting the above technical scheme, the safety response events of various types in the operation can be systematically tracked by configuring the event trigger type and the control response rule for each type of sensor and recording the event type, time, sensor identifier and execution result when triggered, thereby enhancing the traceability of the device behavior; the trigger and response data are stored in a time sequence, and a multi-dimensional index table is constructed, which can support efficient screening and analysis according to the sensor category, time period or response type, thereby providing detailed basis for subsequent safety policy optimization and fault backtracking.

[0043] The above-mentioned second invention object of the application is achieved by the following technical scheme:

[0044] A safety protection system for visual guidance robot feeding and discharging, the safety protection system for visual guidance robot feeding and discharging comprises:

[0045] A space modeling module is configured to construct a digital space model of a robot operation area, register a fence boundary, a sensor deployment position information and a safety area constraint condition;

[0046] A personnel behavior judgment module is configured to collect sensor detection data in a preset time period during robot operation, judge a personnel behavior state in the operation area based on the sensor detection data to obtain a judgment result, and adjust a robot operation parameter according to the judgment result, wherein the sensor detection data includes grating sensor detection data and radar sensor detection data;

[0047] A door lock control module is configured to judge whether the robot is in a non-operation state when detecting an operation request of a person trying to open a fence door lock, and perform authority authentication on the operator, and execute a door lock unlocking control if a preset condition is met;

[0048] An interrupt control module is configured to monitor a trigger signal of an emergency stop button, and immediately issue a global interrupt control instruction to terminate a current task of the robot when detecting any emergency stop signal;

[0049] An event recording and adjustment module is configured to record various sensor trigger events and corresponding control responses, generate a security event log, and dynamically adjust a risk assessment strategy based on the security event log.

[0050] By adopting the technical scheme, through constructing a digital space model of a robot operation area, registering a fence boundary, a sensor deployment position information and a safety area constraint condition, spatial closed management and multi-source perception layout of a robot operation environment can be realized, thereby providing an accurate and structured space basis for subsequent risk judgment and safety control strategy; through collecting sensor detection data in a robot operation process and judging a personnel behavior state, a robot operation parameter is dynamically adjusted according to a judgment result, thereby realizing timely identification and response to personnel intrusion risk, and effectively reducing collision risk caused by personnel approaching or entering a dangerous area; through executing state judgment and permission authentication when an operation request of a personnel attempting to open a fence door lock is detected, it can be ensured that the robot is in a non-operation state and the operator has legal permission, thereby preventing security risks caused by accidental operation or unauthorized entry; through monitoring a trigger signal of an emergency stop button and issuing a global interrupt instruction, the robot task can be quickly terminated in an emergency, thereby realizing timely control and safety guarantee of sudden risk; through recording various sensor trigger events and corresponding control responses, a safety event log is generated, and a risk assessment strategy is dynamically adjusted accordingly, thereby realizing adaptive optimization of safety strategy, and enhancing the dynamic protection capability and continuous improvement capability of the system for different risk situations.

[0051] In summary, the present application includes the following beneficial technical effects:

[0052] 1. By constructing a digital space model of a robot operation area, registering a fence boundary, a sensor deployment position information and a safety area constraint condition, spatial closed management and multi-source perception layout of a robot operation environment can be realized, thereby providing an accurate and structured space basis for subsequent risk judgment and safety control strategy; through collecting sensor detection data in a robot operation process and judging a personnel behavior state, a robot operation parameter is dynamically adjusted according to a judgment result, thereby realizing timely identification and response to personnel intrusion risk, and effectively reducing collision risk caused by personnel approaching or entering a dangerous area;

[0053] 2. By executing state judgment and permission authentication when an operation request of a personnel attempting to open a fence door lock is detected, it can be ensured that the robot is in a non-operation state and the operator has legal permission, thereby preventing security risks caused by accidental operation or unauthorized entry; through monitoring a trigger signal of an emergency stop button and issuing a global interrupt instruction, the robot task can be quickly terminated in an emergency, thereby realizing timely control and safety guarantee of sudden risk;

[0054] 3. By recording various sensor trigger events and corresponding control responses, a safety event log is generated, and a risk assessment strategy is dynamically adjusted accordingly, thereby realizing adaptive optimization of safety strategy, and enhancing the dynamic protection capability and continuous improvement capability of the system for different risk situations. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 is a flowchart of a safety protection method for visual guidance robot feeding and discharging in an embodiment of the present application;

[0056] Figure 2 is an implementation flowchart of step S10 in the safety protection method for visual guidance robot feeding and discharging in an embodiment of the present application;

[0057] Figure 3 is an implementation flowchart of step S20 in the safety protection method for visual guidance robot feeding and discharging in an embodiment of the present application;

[0058] Figure 4 is an implementation flowchart of step S22 in the safety protection method for visual guidance robot feeding and discharging in an embodiment of the present application;

[0059] Figure 5 is an implementation flowchart of step S222 in the safety protection method for visual guidance robot feeding and discharging in an embodiment of the present application;

[0060] Figure 6 is an implementation flowchart of step S30 in the safety protection method for visual guidance robot feeding and discharging in an embodiment of the present application;

[0061] Figure 7 is an implementation flowchart of step S50 in the safety protection method for visual guidance robot feeding and discharging in an embodiment of the present application

[0062] Figure 8 is a principle block diagram of a safety protection system for visual guidance robot feeding and discharging in an embodiment of the present application. DETAILED DESCRIPTION

[0063] The present application will be further described in detail below in conjunction with the accompanying drawings.

[0064] In an embodiment, as shown in FIG. 1, the present application discloses a safety protection method for visual guidance robot feeding and discharging, which specifically comprises the following steps: Figure 1

[0065] S10: Construct a digital space model of the robot working area, register the fence boundary, sensor deployment position information and safety area constraint conditions.

[0066] ​Specifically, based on the input robot work area layout and field device configuration data, the boundary coordinates, material properties and fixed point positions of the fence structure in the work area are extracted, the deployment points and direction information of various sensor devices for perception in the work space are identified, and a three-dimensional point cloud space model is established to restore the real work environment. In the modeling process, the boundary line and buffer level area for representing the safe work range are set, such as setting the area close to the robot arm length range as a first-level high-risk area, and expanding to twice the arm length as a second-level buffer area. By configuring boundary lines and virtual warning zones of different colors, the initialization construction and safety element registration of the digital space model are completed.

[0067] S20: In the process of running the robot, sensor detection data is collected within a preset time period, and the behavior state of the personnel in the work area is judged based on the sensor detection data to obtain a judgment result, and the robot running parameters are adjusted according to the judgment result. The sensor detection data includes light barrier sensor detection data and radar sensor detection data.

[0068] Specifically, based on the set timing trigger mechanism, the data collection task is periodically started, the light barrier sensor data interface is called to obtain the blocking state and judge whether there is a continuous blocking signal, the millimeter wave radar device is called to obtain the two-dimensional or three-dimensional coordinates of the target point in the work space, the speed vector and orientation angle of the target body are analyzed, and the above data and the data of the last time period are compared to calculate the relative displacement and acceleration value of the personnel. According to whether the personnel enter the defined high-risk area or the moving trend of the personnel is towards the core work area of the robot, it is judged whether the behavior state of the personnel in the work area has high-risk behaviors such as approaching, lingering, and crossing, and accordingly the parameters of the current motion speed, path priority or operation mode of the robot are adjusted. For example, when the personnel quickly approach the core work area, the robot action mode is adjusted to low-speed obstacle avoidance or pause waiting mode.

[0069] S30: When detecting that the personnel attempts to open the fence door lock operation request, it is judged whether the robot is in a non-running state, and the operator is authenticated for authority. If the preset conditions are met, the door lock unlocking control is executed.

[0070] Specifically, after receiving the unlocking attempt event sent by the door lock detection module, the current state checking logic of the robot is triggered to determine whether it is in an idle, standby or emergency stop state. In the non-running state, the identification record corresponding to the operator is called, such as card swiping record, fingerprint scanning result or face recognition image. An authentication request is sent to the identity database through the interface and a verification status code is returned. If the verification is passed and there is no robot action execution, an unlocking authorization command is generated to drive the lock controller to execute the electromagnetic release action. The action log and personnel operation time point are recorded together for audit and traceability during the entire operation process.

[0071] S40: Monitor the trigger signal of the emergency stop button, and immediately issue a global interrupt control instruction to terminate the current task of the robot when any emergency stop signal is detected.

[0072] Specifically, by accessing the multi-channel emergency stop button signal bus, all emergency stop trigger ports are monitored in real time. When any channel receives a level mutation or open circuit signal, the interrupt process logic is immediately executed. First, the execution state of the current instruction queue is forcibly terminated, the robot drive instruction is emptied and switched to zero power state, then the interrupt event is broadcast to all associated control nodes such as the above and below clamp controller, conveyor linkage unit, etc. to ensure that all actuators enter a static or locked state simultaneously, and an interrupt response record is generated and written into the central control event log file for subsequent tracking.

[0073] S50: Record various sensor trigger events and corresponding control responses, generate a safety event log, and dynamically adjust the risk assessment strategy based on the safety event log.

[0074] Specifically, during operation, set the corresponding trigger conditions and response action logic for each type of sensor, for example, when the optical grating sensor detects an obstruction, automatically mark the event as "intrusion alert"; when the radar senses a dynamic obstacle approaching, mark it as "approach alert", and define response actions such as deceleration, obstacle avoidance, emergency stop, etc. Through the event-driven model, automatically record the event occurrence time, sensor type, identification number, and executed control action after each trigger event, and record the action response status code and execution result such as "execution success" "no response" etc. All events are appended to the safety log file in chronological order, and a multi-dimensional index table is generated based on the log content for subsequent screening and analysis by time period, sensor type or response category. For example, if the emergency stop button is frequently triggered within a certain period of time, it can be used to identify abnormal operation trends or personnel misoperation risks, thereby dynamically adjusting the relevant safety strategy threshold.

[0075] By adopting the technical scheme, the digital space model of the robot operation area is constructed, the fence boundary, sensor deployment position information and safety area constraint condition are registered, the space closed management and multi-source perception layout of the robot operation environment can be realized, so as to provide accurate and structured space basis for subsequent risk judgment and safety control strategy; by collecting sensor detection data in the robot operation process and judging the behavior state of the personnel, the robot operation parameters are dynamically adjusted according to the judgment result, the personnel intrusion risk can be identified and responded in time, so as to effectively reduce the collision risk caused by the personnel approaching or entering the dangerous area; by performing state judgment and permission authentication when detecting the operation request of the personnel trying to open the fence door lock, it can be ensured that the robot is in a non-operation state and the operator has legal permission, so as to prevent the safety hidden danger caused by misoperation or unauthorized access; by monitoring the trigger signal of the emergency stop button and issuing a global interrupt instruction, the robot task can be quickly terminated in an emergency, so as to realize timely control and safety guarantee of the sudden risk; by recording various sensor trigger events and corresponding control responses, generating a safety event log, and dynamically adjusting the robot operation parameters and risk assessment strategy according to the safety event log, the adaptive optimization of the safety strategy can be realized, so as to enhance the dynamic protection ability and continuous improvement ability of the system to different risk situations.

[0076] In an embodiment, as shown in FIG. 10, in step S10, a digital space model of the robot operation area is constructed, the fence boundary, sensor deployment position information and safety area constraint condition are registered, specifically including: Figure 2

[0077] S11: generating a space topology structure of the operation area based on the operation site layout diagram, recording the three-dimensional coordinate range of the fence boundary.

[0078] Specifically, by loading the two-dimensional CAD diagram or three-dimensional point cloud diagram for representing the operation site, the boundary extraction and space positioning of the fence structure, operation table, equipment body and other key elements in the image are performed, the affine transformation and scale calibration method is adopted to convert the drawing coordinates into the actual physical space coordinate system, the four surrounding support points and vertex information of the fence boundary are extracted to generate a polygon boundary curve, the boundary height is labeled to form a three-dimensional coordinate range, the structure closed property and gap area attribute are recorded, which are used as the boundary judgment reference for subsequent robot path planning and personnel approaching detection.

[0079] S12: mapping the installation position information of the sensor group to the space topology structure to form a space distribution diagram of the perception device, the sensor group including a grating sensor and a radar sensor.

[0080] ​Specifically, after completing the spatial topology construction, the sensor deployment configuration file is called to obtain the specific installation point and orientation angle of each grating sensor and radar sensor. The physical coordinates of the sensors are matched with the scene model coordinate system through a spatial mapping algorithm to generate a spatial distribution map of sensing devices that corresponds one-to-one with the spatial topology. The coverage area, effective area number and device ID of each sensor are marked on the distribution map to support the rapid location of trigger sources and identification of the intersection relationship between the sensed target and the path in actual operation. For example, the radar numbered S1 is mapped to the top of the robot forearm, covering a 90° fan-shaped area on the front of the workbench for personnel approach detection.

[0081] S13: Set the boundary conditions of the multi-level safety control area corresponding to the robot's movement path. The safety control area includes the danger zone and the buffer zone, which are used to classify the risk of personnel approaching and dynamically adjust the robot's operation strategy.

[0082] Specifically, based on the robot's normal operating path planning results, the spatial areas that the robot may traverse during task execution are calculated, and layered safety control zones are set outward from its action radius. The first layer is defined as the danger zone, covering all possible locations that the robot may reach, and is used to trigger high-priority control responses such as emergency stop. The second layer is a buffer zone, set at a certain distance outside the danger zone, used to identify the approach trend of personnel in advance and initiate preventive measures such as deceleration and warnings. By configuring different judgment rules and response levels for each layer, the robot can dynamically switch action strategies according to the current position of the personnel and the zone level. For example, when a person enters the buffer zone but does not enter the danger zone, the robot speed is reduced and a voice warning is issued.

[0083] In one embodiment, such as Figure 3 As shown, in step S20, the behavior status of personnel within the work area is determined based on sensor detection data, a judgment result is obtained, and the robot's operating parameters are adjusted according to the judgment result. Specifically, this includes:

[0084] S21: Based on the detection data of the grating sensor, when the interruption signal generated by the grating sensor is continuously satisfied for a preset time length, it is determined that personnel have entered the danger zone and an emergency stop command is generated.

[0085] Specifically, by periodically sampling the output signal of the grating sensor, it is analyzed whether the blocking state continuously occurs, when the blocking signal is detected in a plurality of consecutive sampling periods and the cumulative duration exceeds a set safety threshold time value, for example, reaches 1.2 seconds, it is determined that a substantial object has entered the grating monitoring area based on the stability of the light beam shielding, and on this basis, the spatial position of the grating corresponding area and the current motion trajectory of the robot are combined to determine that the shielding behavior may cause direct contact risk between the personnel and the robot, thereby immediately generating an emergency stop command to trigger the robot to stop all motion axes and shut down the actuator power supply, for example, when the operator accidentally stretches his hand into the working range causing continuous shielding, the system can trigger the stop after the shielding duration exceeds 1 second to ensure personal safety.

[0086] S22: When the radar sensor detects that the personnel is located in the buffer area, the motion trend characteristics of the personnel are extracted, and the risk level is calculated based on the motion trend characteristics. The motion trend characteristics include the relative distance between the personnel and the robot, the motion direction, and the approaching speed.

[0087] Specifically, by performing spatial reconstruction and target identification operations on the echo data collected by the radar sensor, moving objects in the working area are extracted in real time and it is determined whether they are personnel targets. When the identification result points to personnel and their position is within the pre-set buffer area range, the relative distance between the personnel target and the current position of the robot is calculated from the radar data, and the motion direction and approaching speed of the personnel are further extracted from the detection results of a plurality of consecutive frames to form a complete motion trend characteristic vector. On this basis, the trend analysis logic is executed to determine whether the personnel has a trend of approaching the robot, for example, in a certain detection period, the personnel is identified to move from the periphery to the workbench direction, the direction is towards the robot, and the speed is 1.5 meters per second. Then, in combination with the buffer zone boundary, it is determined that the personnel has a medium risk level and needs early warning processing.

[0088] S23: Switching robot running parameters according to the risk level, the robot running parameters including running speed, joint activity range and obstacle avoidance strategy.

[0089] Specifically, according to the current calculated risk level value, it is mapped to the corresponding safety strategy level, and the running parameter adjustment operation is performed according to the level. If the current risk level is low, the default running speed and standard action range are maintained, if the risk level is medium, the robot running speed is reduced to 70% of the rated value and the joint activity range is reduced, and if the risk level is high, the obstacle avoidance strategy is immediately activated. By real-time path correction, the action direction is deviated from the personnel direction to avoid collision risk, for example, when the personnel is identified to approach the left side of the robot at a relatively fast speed, the control logic automatically folds the left arm action range and turns to the right side to complete the grabbing task, thereby reducing the possibility of potential interference and ensuring task continuity.

[0090] In one embodiment, such as Figure 4 As shown, in step S22, which calculates the risk level based on movement trend characteristics, the specific steps include:

[0091] S221: Extract the historical location sequence and current velocity vector of personnel from the motion trend features, and use them as input data for the model.

[0092] Specifically, the radar or visual detection results over multiple consecutive time periods are processed into a time series, from which the spatial coordinates of personnel within the work area are extracted to construct a historical position sequence of personnel. For example, data is collected every 100ms to form a data sequence of 50 position points within 5 seconds. At the same time, the velocity vector of the personnel relative to the robot coordinate system in the current frame is obtained. This velocity vector includes horizontal and vertical velocity components as well as the resultant velocity value. By combining the historical trajectory with the current velocity, a complete input vector is formed, which is used to provide subsequent prediction models to characterize the dynamic movement characteristics and trend changes of personnel.

[0093] S222: Input the model input data into the pre-trained trajectory prediction model to predict the future movement trajectory of the person, and calculate the minimum spatial distance, predicted intersection time, and trajectory overlap range between the future movement trajectory and the robot's current position and preset movement path.

[0094] Specifically, the extracted historical position sequences and velocity vectors are uniformly transformed into a standardized coordinate system and then used as input variables into a trajectory prediction model that has been trained in advance based on a typical working environment. This model can be a recurrent neural network or a spatiotemporal attention mechanism network. The model outputs a set of predicted trajectory points for personnel positions in several future time periods. Combined with the current precise position of the robot and the planned path, the minimum spatial distance between the pair of points closest to the robot's path in the predicted trajectory is calculated based on the Euclidean distance algorithm. The future intersection time of the pair of points is calculated by combining the time axis. Then, the range of overlapping segments between the predicted trajectory and the robot's path in space is statistically analyzed. For example, if the predicted trajectory enters the working range of the robot's right arm in 2 seconds and there is a 40cm path overlap distance, this situation is recorded as a potential intersection event.

[0095] S223: Based on the minimum spatial distance, predicted intersection time, and trajectory overlap range, combined with the current speed and acceleration of personnel, a corresponding risk level value is generated using a tiered judgment rule.

[0096] Specifically, the minimum spatial distance, the predicted intersection time and the trajectory overlap range are comprehensively judged, the speed and the direction change trend of the personnel in the current frame are combined to judge the acceleration index, when the personnel is in an accelerating approach state, the minimum spatial distance is less than a set safety threshold such as 1 meter, the intersection time is less than 1.5 seconds, and the trajectory overlap range exceeds 30 cm, it is determined as a high-risk level event, and the risk level is assigned a value of "3" in the grading judgment rule, which is used to force trigger the emergency avoidance action in the control logic. On the contrary, if the trajectories intersect but the speed is in the opposite direction and the acceleration is negative, the low risk level "1" is comprehensively assigned, which is used to maintain the current running state and continue to monitor the trend change.

[0097] In an embodiment, as shown in Figure 5 The safety protection method for visual guidance robot feeding and discharging further includes:

[0098] S22201: Based on the historical personnel position information and behavior sample data collected, a training data set is constructed, and the behavior sample data includes the moving path, speed change and obstacle avoidance behavior of the personnel in different work scenes.

[0099] Specifically, the personnel trajectory data recorded in the robot work area for a long period is discretely sampled, the personnel position information collected is arranged in time sequence as a continuous trajectory sequence, and the moving speed change, direction deviation, whether to actively avoid the robot path and other behavior characteristics at the corresponding time point are combined to construct structured sample data. Different work scenes include standard material handling area, closed fence operation area and temporary access passage area. By collecting diversified personnel behavior paths in each type of scene, sample diversity coverage is achieved. For example, typical behaviors collected in the passage area include looking back at the robot while walking, bypassing the robot travel direction and sudden acceleration crossing, thereby forming complete sample pairs in the training data set including moving path, speed curve and obstacle avoidance action label.

[0100] S22202: A trajectory prediction model is constructed based on a long short-term memory network.

[0101] Specifically, the long short-term memory network (LSTM) is used as the main structure of the trajectory prediction model, the input layer accepts fixed-length historical trajectory and speed sequence vectors, the hidden layer contains multiple gate units to capture long-term dependencies in time series, and the output layer generates future multi-frame position coordinate prediction values. In order to improve the prediction accuracy, an attention mechanism module can be introduced into the model structure to assign dynamic weights to input features at different time periods, for example, recent acceleration behavior is assigned a higher weight to reflect sudden trends. This network structure has good time series memory ability and can effectively learn the space-time evolution pattern of personnel behavior.

[0102] S22203: Normalizing the training data set and inputting it into the trajectory prediction model for training, using a loss function that minimizes trajectory prediction error to optimize model weights, generating a pre-trained trajectory prediction model, and dynamically adjusting the weight parameters and risk level determination threshold of the pre-trained trajectory prediction model based on the trigger frequency and response delay data recorded in the safety event log during operation.

[0103] Specifically, the normalized training data set is input into the constructed trajectory prediction model for supervised training, and the average Euclidean distance between the predicted trajectory and the actual trajectory is used as the loss function during training. The network weight parameters are continuously iteratively optimized until the prediction error of the model on the validation set converges within the preset threshold range. Meanwhile, during the running phase after the model is deployed, the recorded data in the safety event log is continuously monitored, and the trigger frequency of various sensors and the response delay of the robot are extracted as feedback indicators. When a high-risk trigger frequency rises or a response time delay exceeds the standard, the model layer weight is dynamically adjusted or the risk level determination threshold is reset, for example, the intersection time determined as "high risk" is advanced from 1.5 seconds to 1.2 seconds, to improve the prediction ability and response timeliness of the sudden event.

[0104] In an embodiment, as shown in FIG. 30, in step S30, the operator is authenticated for permission, and the door lock opening control is performed under the condition that the preset condition is met, specifically including: Figure 6

[0105] S31: Obtain the identity recognition information of the operator, including the ID card ID, fingerprint information and face image data.

[0106] Specifically, the identity recognition information actively submitted or passively collected by the operator when attempting to open the fence door lock is obtained, wherein the ID card ID can be read by the ID card recognition device through the card swiping mode, the fingerprint information is collected in real time by the capacitive fingerprint sensor embedded on the door lock control terminal, and the face image data is captured by the high-definition camera installed above the door lock to capture the front image of the current operator and perform image clarity and angle verification. For example, when the operator's face angle deviation exceeds 30 degrees or the insufficient light causes the image to be blurred, it will prompt to re-collect. All collected data is packaged in a structured format to form an identity recognition request as the input basis for subsequent authentication judgment.

[0107] S32: Compare the identity recognition information with the preset permission database to determine whether the operator has the authorized level to open the fence door lock, and obtain the judgment result.

[0108] ​Specifically, the collected employee ID, fingerprint feature value, and facial image vector are matched and compared with the authorized personnel data stored in the permission database. The employee ID is directly compared using a unique code. The fingerprint information is similar to the hash value generated by matching fingerprint feature points. The facial image data is compared by calling the embedded facial recognition engine to generate a confidence score. For example, if the employee ID is successfully matched but the fingerprint similarity is lower than a set threshold or the facial recognition confidence score is lower than 85%, the authentication is deemed to have failed. The accuracy and security of authentication are improved through a multi-factor joint comparison mechanism, and the final result is the judgment of whether the permission is granted.

[0109] S33: When the judgment result is that the identity authentication is successful and the robot is in a non-operating state, generate a door lock unlocking control command to drive the actuator to complete the unlocking operation of the fence door.

[0110] Specifically, when the identity authentication result is successful and the current robot task status is in a non-running state, a door lock unlocking control command is immediately generated. This control command includes the target door lock number, unlocking action type, and execution time limit parameters. The control command is then transmitted to the execution control unit connected to the door lock, and the door lock is unlocked by a low-voltage electric drive electromagnetic mechanism. For example, the effective time for a single unlock is set to 5 seconds. If the door is not opened within the time limit, it will automatically reset to the locked state. In addition, the control command will be recorded in the operation log for subsequent security audits and personnel behavior tracing.

[0111] In one embodiment, such as Figure 7 As shown, in step S50, various sensor trigger events and corresponding control responses are recorded to generate a safety event log, which specifically includes:

[0112] S51: Configure the corresponding rules for event triggering type and response action for each type of sensor, and record the event type, occurrence time, triggering sensor identifier, control response content and execution result when a triggering event is detected.

[0113] Specifically, each type of sensor such as a light barrier sensor, a radar sensor, and an emergency stop button is configured with a corresponding event trigger type and a response action rule, for example, the light barrier sensor is configured with "blockage signal lasting more than 2 seconds" as the trigger type, and the corresponding response action is "issue a stop command and record the state", the radar sensor is configured with "detecting a moving object entering the buffer zone" as the trigger type, and the corresponding response is "reduce the speed of the robot and enter the observation mode", and the emergency stop button is configured with "manual pressing" as the trigger type, and the corresponding response is "immediately interrupt all actions and maintain the original posture". After detecting any sensor trigger event, an event record is automatically generated and written into a log, and the record content includes the event type (such as personnel approaching or button pressing), the event occurrence time (in milliseconds), the unique identification number of the triggering sensor, the control response instruction content (such as speed limiting or stopping) executed, and the actual execution result (such as successful execution or abnormal interruption), and an execution time delay index is attached for evaluating the response efficiency of the control chain.

[0114] S52: structurally store the trigger events and control responses in time sequence to obtain storage data, and construct a multi-dimensional safety event index table based on the storage data, for supporting safety analysis and backtracking query of specific sensors, specific time periods, or specific response results.

[0115] Specifically, the complete record of the above-mentioned sensor trigger events and corresponding control responses is organized in time sequence as a structured data format, for example, using a timestamp as the main index to form a two-dimensional table data containing fields such as event type, sensor ID, response action, and execution result, and stored in the database in daily units, and a multi-dimensional safety event index table is constructed based on the recorded data, the index dimensions including sensor type, event trigger time range, control response type, and execution success rate, etc., so as to facilitate the operation and maintenance personnel to quickly retrieve event data under specific conditions when needed, for example, all unresponsive events triggered by a specified radar sensor in the past 72 hours or all emergency stop response records with failed execution results can be retrieved, thereby realizing safety analysis, cause tracing, and response strategy optimization of key events.

[0116] It should be understood that the size of the serial number of each step in the above-mentioned embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0117] In an embodiment, a safety protection system for visual guidance robot loading and unloading is provided, which corresponds to the safety protection method for visual guidance robot loading and unloading in the above-mentioned embodiments. As shown in Figure 8As shown, the safety protection system for the vision-guided robot loading and unloading includes a space modeling module, a personnel behavior judgment module, a door lock control module, an interruption control module, and an event record and adjustment module. The detailed description of each functional module is as follows:

[0118] The space modeling module is used to construct a digital space model of the robot operation area, register the fence boundary, sensor deployment position information, and safety area constraint conditions;

[0119] The personnel behavior judgment module is used to collect sensor detection data within a preset time period during the robot operation, judge the personnel behavior state in the operation area based on the sensor detection data, obtain a judgment result, and adjust the robot operation parameters according to the judgment result. The sensor detection data includes grating sensor detection data and radar sensor detection data;

[0120] The door lock control module is used to judge whether the robot is in a non-operation state when detecting an operation request of the operator trying to open the fence door lock, and perform authority authentication on the operator. The door lock unlocking control is executed under the condition that the preset condition is met;

[0121] The interruption control module is used to monitor the trigger signal of the emergency stop button. When any emergency stop signal is detected, a global interruption control instruction is immediately issued to terminate the current task of the robot;

[0122] The event record and adjustment module is used to record various sensor trigger events and corresponding control responses, generate a safety event log, and dynamically adjust the risk assessment strategy based on the safety event log.

[0123] Optionally, the space modeling module includes:

[0124] The space topology modeling submodule is used to generate a space topology structure of the operation area based on the operation site layout diagram, and record the three-dimensional coordinate range of the fence boundary;

[0125] The sensor mapping submodule is used to map the installation position information of the sensor group to the space topology structure to form a space distribution diagram of the perception device. The sensor group includes grating sensors and radar sensors;

[0126] The safety area configuration submodule is used to set the boundary conditions of the multi-level safety control area corresponding to the robot motion path. The safety control area includes a danger area and a buffer area, which are used to judge the personnel approach to the risk and dynamically adjust the robot operation strategy.

[0127] Optionally, the personnel behavior judgment module includes:

[0128] The interruption detection submodule is configured to detect data of the grating sensor, and determine that a person enters a dangerous area and generate an emergency stop command when it is detected that the interruption signal generated by the grating sensor lasts for a preset length of time.

[0129] The motion feature extraction submodule is configured to extract a motion trend feature of the person when the radar sensor detects that the person is located in the buffer area, and calculate a risk level based on the motion trend feature, the motion trend feature including a relative distance between the person and the robot, a motion direction, and an approaching speed.

[0130] The parameter adjustment submodule is configured to switch robot operation parameters including an operation speed, a joint activity range, and an obstacle avoidance strategy according to the risk level.

[0131] Optionally, the motion feature extraction submodule includes:

[0132] The trajectory modeling input unit is configured to extract a historical position sequence and a current speed vector of the person from the motion trend feature as model input data.

[0133] The future trajectory prediction unit is configured to input the model input data into a pre-trained trajectory prediction model, predict a future motion trajectory of the person, and calculate a minimum spatial distance between the future motion trajectory and a current position of the robot and a preset motion path, a predicted intersection time, and a trajectory overlap range.

[0134] The risk level evaluation unit is configured to generate a corresponding risk level value according to the minimum spatial distance, the predicted intersection time, and the trajectory overlap range, in combination with a current speed and an acceleration of the person, and using a hierarchical decision rule.

[0135] Optionally, the safety protection system for the vision-guided robot loading and unloading further includes:

[0136] The sample data construction module is configured to construct a training data set based on historical collected person position information and behavior sample data, the behavior sample data including a moving path, a speed change, and an obstacle avoidance behavior of the person in different work scenarios.

[0137] The trajectory model construction module is configured to construct a trajectory prediction model based on a long short-term memory network.

[0138] The model training and optimization module is configured to perform normalization processing on the training data set, input the training data set into the trajectory prediction model for training, perform model weight optimization using a loss function that minimizes trajectory prediction error, generate the pre-trained trajectory prediction model, and adjust weight parameters and a decision threshold of the pre-trained trajectory prediction model in real time based on a safety event log during operation.

[0139] Optionally, the door lock control module includes:

[0140] An identity information collection sub-module is configured to acquire identity recognition information of an operator, the identity recognition information including a badge ID, fingerprint information, and face image data.

[0141] A permission comparison sub-module is configured to compare the identity recognition information with a preset permission database to determine whether the operator has an authorized level to open the fence door lock and obtain a determination result.

[0142] A lock opening execution sub-module is configured to generate a door lock opening control instruction to drive an execution mechanism to complete the fence door opening operation when the determination result is that the identity authentication is passed and the robot is in a non-running state.

[0143] Optionally, the event recording and adjustment module includes:

[0144] A trigger rule configuration module is configured to configure a corresponding rule of an event trigger type and a response action for each type of sensor, and record the event type, occurrence time, trigger sensor identifier, control response content, and execution result when a trigger event is detected.

[0145] A log structuring module is configured to structurally store the trigger event and the control response in chronological order to obtain storage data, and construct a multi-dimensional security event index table based on the storage data to support security analysis and backtracking query of a specific sensor, a specific time period, or a specific response result.

[0146] For specific limitations of the safety protection system for the vision-guided robot loading and unloading, refer to the limitations of the safety protection method for the vision-guided robot loading and unloading described above, which will not be repeated here. Each module in the safety protection system for the vision-guided robot loading and unloading can be realized by software, hardware, and combinations thereof, in whole or in part. Each module described above can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0147] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual applications, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the functions described above.

[0148] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A safety protection method for loading and unloading materials using a vision-guided robot, characterized in that, include: Construct a digital spatial model of the robot's operating area, and register fence boundaries, sensor deployment location information, and safety zone constraints; The construction of the digital spatial model of the robot's operating area, including the registration of fence boundaries, sensor deployment location information, and safety zone constraints, includes: Generate the spatial topology of the work area based on the work site layout diagram, and record the three-dimensional coordinate range of the fence boundary; The installation location information of the sensor group is mapped to the spatial topology to form a spatial distribution map of the sensing devices. The sensor group includes grating sensors and radar sensors. Set multi-level safety control zone boundary conditions corresponding to the robot's movement path. The safety control zone includes a danger zone and a buffer zone, which are used to classify the risk of personnel approaching and dynamically adjust the robot's operation strategy. During the operation of the robot, sensor detection data is collected within a preset time period, and the behavior status of personnel in the work area is judged based on the sensor detection data to obtain a judgment result. The robot's operating parameters are adjusted according to the judgment result. The sensor detection data includes grating sensor detection data and radar sensor detection data. The step of determining the behavior status of personnel within the work area based on the sensor detection data, obtaining a determination result, and adjusting the robot's operating parameters according to the determination result includes: Based on the detection data of the grating sensor, when the blocking signal generated by the grating sensor is continuously satisfied for a preset time length, it is determined that personnel have entered the danger zone and an emergency stop command is generated. When the radar sensor detects that a person is in the buffer zone, the movement trend characteristics of the person are extracted, and the risk level is calculated based on the movement trend characteristics. The movement trend characteristics include the relative distance between the person and the robot, the direction of movement, and the approach speed. The robot's operating parameters are switched according to the risk level. The robot's operating parameters include operating speed, joint range of motion, and obstacle avoidance strategy. The calculation of risk level based on the movement trend characteristics includes: Extract the historical location sequence and current velocity vector of the person from the motion trend features, and use them as model input data; The model input data is input into a pre-trained trajectory prediction model to predict the future movement trajectory of the person, and to calculate the minimum spatial distance, predicted intersection time, and trajectory overlap range between the future movement trajectory and the robot's current position and preset movement path. Based on the minimum spatial distance, predicted intersection time, and trajectory overlap range, combined with the personnel's current speed and acceleration, a corresponding risk level value is generated using a tiered judgment rule. When a request to open the fence gate is detected, it is determined whether the robot is in a non-operating state, and the operator's permissions are authenticated. If the preset conditions are met, the door unlocking control is executed. Monitor the trigger signal of the emergency stop button, and when any emergency stop signal is detected, immediately issue a global interrupt control command to terminate the robot's current task; Record various sensor-triggered events and corresponding control responses, generate a safety event log, and dynamically adjust the risk assessment strategy based on the safety event log; The safety protection method for loading and unloading materials using a vision-guided robot also includes: A training dataset is constructed based on historically collected personnel location information and behavioral sample data. The behavioral sample data includes personnel movement paths, speed changes, and obstacle avoidance behaviors in different work scenarios. A trajectory prediction model is constructed based on a long short-term memory network; The training dataset is normalized and input into the trajectory prediction model for training. The model weights are optimized using a loss function that minimizes the trajectory prediction error to generate the pre-trained trajectory prediction model. During operation, the weight parameters and risk level judgment threshold of the pre-trained trajectory prediction model are dynamically adjusted based on the trigger frequency and response delay data recorded in the security event log.

2. The safety protection method for loading and unloading materials using a vision-guided robot according to claim 1, characterized in that, The step of authenticating the operator's permissions and executing the door lock unlocking control under preset conditions includes: Obtain the operator's identity information, which includes employee ID, fingerprint information, and facial image data; The identity information is compared with a preset permission database to determine whether the operator has the authorization level to open the fence gate lock, and the determination result is obtained. When the judgment result indicates that the identity authentication is successful and the robot is in a non-operating state, a door lock unlocking control command is generated to drive the actuator to complete the unlocking operation of the fence gate.

3. The safety protection method for loading and unloading materials using a vision-guided robot according to claim 1, characterized in that, The process of recording various sensor-triggered events and corresponding control responses to generate a safety event log includes: Configure the corresponding rules for event triggering types and response actions for each type of sensor, and record the event type, occurrence time, triggering sensor identifier, control response content and execution result when a triggering event is detected; The triggering events and control responses are structured and stored in chronological order to obtain stored data. A multi-dimensional security event index table is then constructed based on the stored data to support security analysis and backtracking queries for specific sensors, specific time periods, or specific response results.

4. A safety protection system for loading and unloading materials using a vision-guided robot, characterized in that, include: The spatial modeling module is used to construct a digital spatial model of the robot's operating area, registering fence boundaries, sensor deployment location information, and safety zone constraints. The personnel behavior judgment module is used to collect sensor detection data within a preset time period during robot operation, and judge the behavior status of personnel in the work area based on the sensor detection data, obtain the judgment result, and adjust the robot operation parameters according to the judgment result. The sensor detection data includes grating sensor detection data and radar sensor detection data. The door lock control module is used to determine whether the robot is in a non-operating state when it detects an operation request from a person attempting to open the fence door lock, and to authenticate the operator's permissions. If preset conditions are met, the module will execute the door lock unlocking control. The interrupt control module is used to monitor the trigger signal of the emergency stop button. When any emergency stop signal is detected, it immediately issues a global interrupt control command to terminate the robot's current task. The event logging and adjustment module is used to record various sensor-triggered events and corresponding control responses, generate safety event logs, and dynamically adjust risk assessment strategies based on the safety event logs. The spatial modeling module includes: The spatial topology modeling submodule is used to generate the spatial topology of the work area based on the work site layout diagram and record the three-dimensional coordinate range of the fence boundary. The sensor mapping submodule is used to map the installation location information of the sensor group to the spatial topology to form a spatial distribution map of the sensing devices. The sensor group includes grating sensors and radar sensors. The safety zone configuration submodule is used to set the boundary conditions of multi-level safety control zones corresponding to the robot's movement path. The safety control zones include danger zones and buffer zones, which are used to classify the risk of personnel approaching and dynamically adjust the robot's operation strategy. The personnel behavior judgment module includes: The interruption detection submodule is used to determine that personnel have entered the danger zone and generate an emergency stop command when the interruption signal generated by the grating sensor continuously meets the preset time length based on the detection data of the grating sensor. The motion feature extraction submodule is used to extract the motion trend features of a person when the radar sensor detects that the person is in the buffer area, and calculate the risk level based on the motion trend features. The motion trend features include the relative distance between the person and the robot, the direction of movement, and the approach speed. The parameter adjustment submodule is used to switch robot operating parameters according to the risk level. The robot operating parameters include operating speed, joint range of motion, and obstacle avoidance strategy. The motion feature extraction submodule includes: The trajectory modeling input unit is used to extract the historical position sequence and current velocity vector of the person from the motion trend features, as model input data; The future trajectory prediction unit is used to input the model input data into a pre-trained trajectory prediction model, predict the future movement trajectory of the person, and calculate the minimum spatial distance, predicted intersection time, and trajectory overlap range between the future movement trajectory and the robot's current position and preset movement path. The risk level assessment unit is used to generate a corresponding risk level value based on the minimum spatial distance, predicted intersection time, trajectory overlap range, and the current speed and acceleration of personnel, using a grading judgment rule. The safety protection system for loading and unloading a vision-guided robot also includes: The sample data construction module is used to construct a training dataset based on historically collected personnel location information and behavioral sample data. The behavioral sample data includes personnel movement paths, speed changes, and obstacle avoidance behaviors in different work scenarios. The trajectory model building module is used to build trajectory prediction models based on long short-term memory networks. The model training and optimization module is used to normalize the training dataset and input it into the trajectory prediction model for training. It uses a loss function that minimizes the trajectory prediction error to optimize the model weights, generate the pre-trained trajectory prediction model, and dynamically adjust the weight parameters and risk level judgment threshold of the pre-trained trajectory prediction model based on the trigger frequency and response delay data recorded in the security event log during operation.

Images