Multi-robot optical processing safety control system and method based on machine vision

Abstract

The present application relates to the technical field of robot control, and especially relates to a multi-robot optical processing safety control system and method based on machine vision. The method covers the global vision of multi-robot collaborative processing through multiple vision sensors, and tracks multiple feature targets in the workspace in parallel. The high-reflective marker points on the robots in the image are located at the sub-pixel level. After the pixel boundary box is used to determine and issue a primary early warning event or a collision warning event, the pose solution frequency of the involved robot is dynamically adjusted, and the three-dimensional space pose data of the robot is solved. Whether to execute motion intervention or emergency stop is determined according to the three-dimensional space shortest Euclidean distance and the predicted collision time between the involved robots. The present application uses a low-complexity two-dimensional multi-target tracking method for global continuous monitoring, and performs high-precision pose solution and collision detection in stages, so as to realize dynamic optimization allocation of computing resources, avoid monitoring false negatives, and effectively reduce system running load.

Description

Technical Field

[0001] This invention belongs to the field of robot control technology, and in particular relates to a multi-robot optical processing safety control system and method based on machine vision. Background Technology

[0002] With the rapid development of the precision optical manufacturing industry, multi-robotic arm collaborative operations, with their outstanding advantages such as strong processing adaptability, excellent complex surface forming capability, and high operating efficiency, have been widely applied in high-end manufacturing scenarios such as optical component polishing, precision device assembly, and irregularly shaped optical component processing. They also possess significant engineering application value in the fields of aerospace optics and high-end optoelectronic component production. Currently, machine vision perception, intelligent control of robotic arms, and obstacle avoidance technologies are gradually being applied to robotic arm processing control systems. However, existing multi-robotic arm motion monitoring solutions mostly adopt a global multi-view vision continuous high-precision pose calculation mode, which suffers from problems such as large amounts of computational data, high computational resource consumption, and significant inference latency, making it difficult to meet the high real-time monitoring and control requirements in multi-robotic arm collaborative processing. Furthermore, existing publicly available patents and related technical solutions still have significant technical shortcomings in multi-robotic arm collaborative processing scenarios.

[0003] Patent application 1 (Chinese Patent Publication No. CN121120742A, Publication Date December 12, 2025, Patent Title: "A 3D Reconstruction and Pose Estimation System and Method Based on Binocular Structured Light") belongs to the technical field of machine vision and 3D reconstruction, focusing on high-precision modeling of non-standard irregular parts. The system in this patent application acquires multi-view data through binocular structured light images, combines gyroscope attitude feedback, and performs 3D point cloud reconstruction and pose estimation, involving visual pose estimation and 3D reconstruction, which is similar to the "machine vision pose calibration and real-time calculation" in this invention. However, patent application 1 focuses on static object reconstruction (such as irregular parts) and lacks support for dynamic multi-target collaborative operations. Its system relies on fixed posture acquisition and cannot handle real-time motion interference from robotic arms in optical processing. Furthermore, the pose estimation in this patent application relies on gyroscope priors, which may fail in complex lighting or multi-occlusion environments (such as optical processing workshops). Its point cloud completion algorithm (PCN model) mainly handles static missing data and has poor real-time performance for dynamic target tracking.

[0004] Patent application 2 (Chinese Patent Publication No. CN120791727A, Publication Date October 17, 2025, Patent Title: "A Mobile Robotic Arm Grasping Control Method and System Based on Machine Vision") belongs to the technical field of mobile robotic arm grasping technology. It uses the RT-DETR algorithm to identify dynamic obstacles, predict trajectories, and perform obstacle avoidance control. This patent application involves visual obstacle avoidance control for robotic arms, which is similar to the "global obstacle avoidance warning" requirement in this invention. Furthermore, patent application 2 uses real-time image processing and target tracking, which is similar to the "machine vision feature recognition" technical solution in this invention. Although patent application 2 involves mobile robotic arms, it focuses on single-arm grasping tasks, and the obstacle avoidance strategy only targets local obstacles, without considering global conflict warning under multi-arm collaboration. For example, its obstacle avoidance model is based on monocular or binocular vision, with a limited field of view, easily generating blind spots in multi-arm interactions. At the same time, the obstacle avoidance control solution in patent application 2 is based on a simplified physical model (such as parabolic prediction), without integrating multi-sensor fusion or hierarchical decision-making. While vibration compensation improves stability, it increases computational complexity and may affect the response speed of multi-machine collaboration. Neither of these systems employs hierarchical visual monitoring. The system in patent application 1 is a fixed assembly, lacking modular expansion capabilities; the control logic in patent application 2 is concentrated on a single machine, making multi-machine data sharing and collaborative optimization impossible.

[0005] In multi-robot collaborative optical processing scenarios, multiple robotic arms operate within a shared workspace, posing a high risk of motion interference. Current technologies primarily mitigate collision risks through pre-set motion control logic, lacking a global, independent, real-time monitoring system based on actual spatial positions. This limitation leads to two prominent safety hazards during multi-robot collaborative processing: firstly, it cannot address unexpected path deviations caused by accumulated robot absolute positioning errors, workpiece clamping deviations, and external environmental disturbances; secondly, for flexible end-effectors such as polishing tools, it is difficult to provide early warning of abnormal contact between the tool and the workpiece. Therefore, a real-time visual monitoring solution that balances operational safety and computational efficiency is urgently needed. Summary of the Invention

[0006] In view of this, the present invention aims to provide a multi-robot optical processing safety control system and method based on machine vision. It achieves global field of view coverage through multiple vision sensors, uses a two-dimensional image multi-target tracking method to monitor multiple robots globally, and triggers high-precision pose calculation and collision detection through early warning events to achieve dynamic optimization allocation of computing resources.

[0007] To achieve the above objectives, the technical solution created by this invention is implemented as follows: A machine vision-based multi-robot optical processing safety control system includes: A rotary table, in which the optical element to be processed is fixed in the processing area of ​​the rotary table, and at least two calibration patterns are installed in the non-processing area of ​​the rotary table; At least two robots, each with a calibration plate and a tool head fixed to its end. The tool head is used to process the optical components to be processed; each tool head is equipped with a marking pattern. At least two vision sensors are used to simultaneously acquire robot posture images; each vision sensor is arranged at a preset height above the rotating table so that the robot is simultaneously within the effective field of view and depth of field of at least two vision sensors. The vision processing module is used to locate the static center pixel coordinates of the marking pattern based on the static robot posture image and to mark the pixel bounding box. The target tracking module is used to track all the marker patterns in parallel in dynamic robot pose images, calculate the dynamic center pixel coordinates of each marker pattern in real time, and mark the pixel bounding box. The safety decision processing module is used to generate a primary warning event or a collision warning event when the pixel bounding boxes of any two identification patterns overlap; reallocate the robot's pose calculation frequency and trigger three-dimensional pose calculation; and determine whether the preset threshold condition is reached based on the real-time calculated shortest Euclidean distance in three-dimensional space and the estimated collision time. The control module is used to control the robot and tool head to perform motion intervention or emergency braking.

[0008] Furthermore, both the marking pattern and the identification pattern consist of at least three non-collinear, highly reflective marking points.

[0009] Furthermore, the vision processing module includes an image receiving unit and a pattern calibration unit; The image receiving unit is used to receive static robot posture images acquired by the vision sensor; The pattern calibration unit is used to calculate the static center pixel coordinates of each marker pattern in the static robot pose image and to calibrate the pixel bounding box.

[0010] Furthermore, the target tracking module includes a target tracking unit and an image processing unit; The target tracking unit is used to track all the marking patterns in parallel in dynamic robot pose images; The image processing unit is used to calculate the dynamic center pixel coordinates of each marker pattern in real time and mark the pixel bounding box.

[0011] Furthermore, the safety decision processing module includes an early warning unit, a pose calculation unit, and a safety decision unit; The warning unit is used to generate a primary warning event or a collision warning event when the pixel bounding boxes of any two marker patterns overlap, and to designate the robot corresponding to the marker pattern that generates the primary warning event or the collision warning event as the robot involved; to reallocate the pose calculation frequency of the robot involved, and to start the pose calculation unit. The pose calculation unit is used to perform three-dimensional pose calculation on the robots involved in the incident, obtain the pose data of the robots involved in the incident, and calculate the shortest Euclidean distance and the estimated collision time in three-dimensional space between the robots involved in the incident in real time using the pose data. The safety decision unit is used to determine whether the shortest Euclidean distance and / or the expected collision time in three-dimensional space have reached the preset threshold conditions, including a deceleration warning threshold or an emergency stop threshold. When the shortest Euclidean distance and / or the expected collision time in three-dimensional space reach the deceleration warning threshold, the control module is activated to perform motion intervention. When the shortest Euclidean distance and / or the expected collision time in three-dimensional space reach the emergency stop threshold, the control module is activated to perform emergency braking.

[0012] Furthermore, the robot is a six-degree-of-freedom robotic arm.

[0013] A machine vision-based multi-robot optical processing safety control method, implemented based on the aforementioned safety control system, includes the following steps: S1: Before the system runs, use a vision sensor to synchronously collect static images of multiple robots driving the tool head for close processing and static images of approaching the limit safety. Use a feature point detection algorithm to locate the center pixel coordinates of all highly reflective marker points on all marking patterns in the static images and mark the pixel bounding boxes. S2: After the system is running, the vision sensor is controlled to synchronously acquire dynamic images of multiple robots driving the tool head for processing. All marking patterns are tracked in parallel. A tracking algorithm based on discriminant correlation filtering is used to calculate the pixel coordinates of each marking pattern in each frame of synchronously acquired dynamic image and mark the pixel bounding box. When the pixel bounding boxes of any two marking patterns touch or overlap in the dynamic image, a primary warning event is generated. The robot corresponding to the marking pattern that generates a primary warning event or a collision warning event is designated as the robot involved in the incident, and the robot corresponding to the marking pattern that does not generate a primary warning event or a collision warning event is designated as the robot not involved in the incident. S3: Dynamically adjust the pose calculation frequency of the robot involved in the incident, while the robot not involved in the incident maintains its basic pose at a preset basic frequency; use an efficient perspective n-point algorithm to calculate the pose data of the robot involved in the incident in the workpiece coordinate system. S4: Construct a dynamic envelope model of the robots involved in the incident by combining pose data, and use a collision detection algorithm to calculate the shortest Euclidean distance between the robots in three-dimensional space in real time. Combine the relative approach speed of the robots involved to solve the estimated collision time. When the shortest Euclidean distance in three-dimensional space and / or the estimated collision time reach the deceleration warning threshold, intervene in the motion of the robots involved; or, when the shortest Euclidean distance in three-dimensional space and / or the estimated collision time reach the emergency stop threshold, perform emergency braking on all robots.

[0014] Furthermore, in step S1, the center pixel coordinates of the highly reflective marker ( The formula for calculating ) is: ; in, These are the pixel coordinates within the pixel window. This corresponds to the grayscale value of the pixel.

[0015] Furthermore, in step S3, the pose calculation frequency... The calculation formula is: ; in,{ A , B} represents the serial number of the robot involved. To preset the base frequency, This represents the maximum computation frequency corresponding to the highest frame rate of the visual sensor. The gain coefficient is calibrated based on the delay of the safety control system and the dynamic characteristics of the robot. This represents the relative approach speed of the robot involved in the incident in three-dimensional space.

[0016] Furthermore, after step S4, step S5 is also included: the safety control system continuously monitors the shortest Euclidean distance in three-dimensional space between the robots involved. When the shortest Euclidean distance in three-dimensional space continues to exceed the preset safe distance threshold for a preset time, it is determined that the risk has been eliminated; the primary warning event or collision warning event is turned off, and the pose calculation frequency of the robots involved is reduced to the preset base frequency.

[0017] Compared with the prior art, the present invention can achieve the following beneficial effects: (1) This invention addresses the needs of multi-robot collaborative operation in optical processing scenarios. It achieves full-field coverage through a safety control system based on multiple vision sensors. Combining machine vision two-dimensional image feature recognition, machine vision pose calibration, and real-time three-dimensional pose calculation, it constructs a hierarchical and layered visual monitoring and collision warning method, which effectively avoids the monitoring blind spot problem of a single vision sensor. The multiple vision sensors are reasonably arranged to synchronously monitor the dynamic multi-robot workspace, ensuring the timeliness and accuracy of safety warnings for operation under large field of view and multi-target conditions.

[0018] (2) This invention uses a hierarchical visual monitoring and judgment early warning method to decompose the task into two levels of execution: one is global monitoring based on real-time two-dimensional image monitoring and event triggering based on multi-target tracking; the other is local execution of event-driven high-precision three-dimensional pose calculation and collision early warning. This invention can flexibly adapt to the monitoring needs of complex operation scenarios, realize the dynamic optimization allocation of computing resources, and effectively reduce the average load of the system while ensuring zero missed reports, so that the system can always maintain operational stability and monitoring reliability under the high dynamic conditions of optical processing. Attached Figure Description

[0019] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 A schematic diagram of the structure of the machine vision-based multi-robot optical processing safety control system described in the embodiment of the present invention; Figure 2 This is a flowchart illustrating the machine vision-based multi-robot optical processing safety control method described in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0022] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0023] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0024] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] like Figure 1 As shown, this embodiment of a multi-robot optical processing safety control system based on machine vision includes a rotary table 1, at least two robots 2, at least two vision sensors 3, a vision processing module (not shown), a target tracking module (not shown), a safety decision processing module (not shown), and a control module (not shown).

[0026] A rotary worktable 1 has an optical element to be processed fixed in its processing area and a calibration pattern installed in its non-processing area.

[0027] At least two robots 2, each robot 2 has a calibration plate and a tool head fixed at its end. The tool head is used to process the optical element to be processed. Each tool head is equipped with a marking pattern.

[0028] In some embodiments, both the calibration pattern and the marking pattern consist of at least three non-collinear highly reflective marker dots.

[0029] At least two vision sensors 3 are used to synchronously acquire posture images of the robot 2; each vision sensor 3 is arranged at a preset height above the rotating worktable 1 so that the robot 2 is simultaneously within the effective field of view and depth of field of at least two vision sensors 3.

[0030] The vision processing module is used to locate the static center pixel coordinates of the identification pattern based on the static robot posture image and to mark the pixel bounding box. The vision processing module includes an image receiving unit and a pattern marking unit. The image receiving unit is used to receive the static robot posture image acquired by the vision sensor 3. The pattern marking unit is used to calculate the static center pixel coordinates of each identification pattern in the static robot posture image and to mark the pixel bounding box. The target tracking module is used to track all marker patterns in parallel in a dynamic robot pose image, calculate the dynamic center pixel coordinates of each marker pattern in real time, and mark the pixel bounding box. This target tracking module includes a target tracking unit and an image processing unit; the target tracking unit is used to track all marker patterns in parallel in a dynamic robot pose image; the image processing unit is used to calculate the dynamic center pixel coordinates of each marker pattern in real time and mark the pixel bounding box.

[0031] The safety decision processing module generates a primary warning event or a collision warning event when the pixel bounding boxes of any two marker patterns overlap; it reallocates the pose calculation frequency of robot 2 and triggers 3D pose calculation; and determines whether a preset threshold condition has been met based on the real-time calculated shortest Euclidean distance in 3D space and the estimated collision time. This safety decision processing module includes a primary warning unit, a pose calculation unit, and a safety decision unit. The primary warning unit generates a primary warning event or a collision warning event when the pixel bounding boxes of any two marker patterns overlap, designating robot 2 corresponding to the marker pattern that generates the primary warning event or collision warning event as the involved robot; it reallocates the pose calculation frequency of the involved robot and activates the pose calculation unit; the pose calculation unit performs 3D pose calculation on the involved robot to obtain its pose data, and calculates the shortest Euclidean distance in 3D space and the estimated collision time between the involved robots in real time using the pose data. The safety decision unit is used to determine whether the shortest Euclidean distance in three-dimensional space and / or the expected collision time have reached a preset threshold condition. The preset threshold conditions include a deceleration warning threshold or an emergency stop threshold. When the shortest Euclidean distance in three-dimensional space and / or the expected collision time reach the deceleration warning threshold, the control module is activated to perform motion intervention. When the shortest Euclidean distance in three-dimensional space and / or the expected collision time reach the emergency stop threshold, the control module is activated to perform emergency braking.

[0032] The control module is used to control robot 2 and the tool head to perform motion intervention or emergency braking.

[0033] In some embodiments, robot 2 can be a six-degree-of-freedom robotic arm.

[0034] The machine vision-based multi-robot optical processing safety control system of this invention is used to achieve blind-spot-free and highly redundant visual coverage of the collaborative workspace of multiple robots. It constructs a stable and unified measurement field under the tool coordinate system by arranging at least two vision sensors 3 through strict geometric constraints.

[0035] This embodiment demonstrates the layout of an optical processing scenario using a 1.5m diameter rotary worktable 1, three robots 2, and two vision sensors 3. First, the intersection of the rotation axis of the rotary worktable 1 and the table surface is defined as the origin of the tool coordinate system. , The plane is parallel to the workbench surface. The axis is vertically upward. Each vision sensor 3 is arranged at a preset height above the rotary table 1 so that the robot 2 is simultaneously within the effective field of view and depth of field of at least two vision sensors 3.

[0036] In some embodiments, the specific number of vision sensors 3 is determined based on the actual processing site size, achieving an optimal balance between cost and performance. The optical center of the vision sensor 3 ( l =1,2) Above the rotary table 1 at the same preset height The components are arranged at equal intervals on the same imaginary circle, which is concentric with the rotary table 1. The radius of the imaginary circle is... Larger than the radius of rotary table 1 The farthest distance that robot 2 moves to the rotary table 1 The sum of these ensures that any point within the workspace is simultaneously within the effective field of view and depth of field of at least one vision sensor 3. Preset height The range is Preset height H The height range is inversely proportional to the pixel resolution (mm / pixel) of the plane of the rotating worktable 1 (i.e., the main motion plane where the tool head end of robot 2 is located); this height range ensures both sufficient field of view and pixel resolution sufficient to meet the sub-pixel positioning requirements of highly reflective markers in the marking pattern, while avoiding physical interference between robot 2 and vision sensor 3 or its support when moving at its highest position. The angle between the optical axis of vision sensor 3 and the vertical direction... θ Setting the angle between 30° and 45° allows for an image closer to orthographic projection, reducing perspective distortion and facilitating subsequent measurements.

[0037] Secondly, a target feature benchmark based on two-dimensional image feature constraints is constructed, that is, a unique and robust visual feature benchmark is set for each monitored object.

[0038] To establish the association between the vision sensor 3 and the spatial pose, it is necessary to calibrate the pose relationship between the vision sensor 3 coordinate system and the workpiece coordinate system. The workpiece coordinate system is a reference coordinate system established on the plane of the rotary table 1, used to describe the feature points and the execution of the robot 2's movement position. The robot 2's movement position execution is based on the workpiece coordinate system reference. Therefore, by integrating a calibration plate on the end axis of the robot 2, the pose relationship between the vision sensor 3 and the workpiece coordinate system is calibrated through hand-eye calibration, obtaining the homogeneous transformation pose relationship matrix between the vision sensor 3 coordinate system and the workpiece coordinate system. Specifically, a calibration plate is fixed on the end flange of the fifth axis of each robot 2. The calibration plate can be square with a side length of 50mm. The surface of the calibration plate is set with a high-contrast checkerboard pattern. The pose transformation matrix is ​​obtained through the checkerboard calibration method. This allows the feature point information collected and identified by vision sensor 3 to be transformed into a workpiece coordinate system for description. This serves as a basis for subsequent calculation of the spatial pose relationship of feature target information in the event of a primary warning event or collision warning event.

[0039] In multi-robot collaborative processing, the part with the most frequent motion changes and the highest risk of interference is the tool head mounted on the end effector of robot 2. Each tool head is equipped with a marking pattern. This marking pattern consists of at least three non-collinear highly reflective markers, which can be made of reflective material and are fixed in space to form a predetermined geometric pattern, such as an isosceles triangle. This marking pattern appears as a cluster of bright spots with a fixed relative position in a two-dimensional image, constituting a unique two-dimensional image feature of the tool head, which is used by vision sensor 3 for real-time tracking and positioning. Meanwhile, to eliminate interference from the rotation of the rotary table 1 itself on the global measurement, at least two calibration patterns are installed in the non-machining area of ​​the rotary table 1, such as its outer edge or on a dedicated fixture. These calibration patterns consist of at least three non-collinear, highly reflective markers made of reflective material. The calibration patterns define the origin and direction of the coordinate system of the rotary table 1 and provide a stable static reference for the entire system. This provides the system with a real-time basis for determining whether the rotary table 1 is currently stationary, thus deciding whether to perform pose calculation on the tool head. The safety control system continuously tracks the calibration patterns fixed on the rotary table 1 to determine in real time whether the rotary table 1 is rotating. When the calibration pattern is identified as being in a rotating motion state in the real-time image acquired by the vision sensor 3, it indicates that the collaborative machining system is currently in operation, and the safety control system needs to continuously monitor the current working environment. Furthermore, by recognizing the calibration pattern, the safety control system can calculate the current rotation angle of the rotary table 1 based on the previously established homogeneous transformation pose relationship matrix.

[0040] like Figure 2 As shown, this embodiment of the invention also provides a multi-robot optical processing safety control method based on machine vision, implemented based on the above-mentioned safety control system, including the following steps: S1: Before the system runs, use vision sensors to synchronously collect static images of multiple robots driving tool heads in close processing and static images of extreme safety approach. Use feature point detection algorithm to locate the center pixel coordinates of all highly reflective marker points on all marking patterns in the static images and mark the pixel bounding boxes.

[0041] Before the system is officially put into operation, a preliminary early warning criterion based on image pixel distance is established through a one-time offline calibration process. This criterion includes a three-dimensional safety distance threshold. Mapped to pixel tolerance on the image plane. Step S1 includes the following steps: S11: Acquire static postures of multiple robots driving tool heads in close machining operations. Simulate multi-robot collaborative machining operations by moving three robots to a preset machining posture where the tool heads are relatively close together. Simultaneously acquire and record static images of the multiple robots driving tool heads in close machining operations using vision sensors. Ensure that the still image The logo pattern is covered in the middle.

[0042] S12: Collect the static posture of multiple robots when their tool heads approach the limit of safe proximity. Under the premise of absolute safety, manually control any two robots (e.g., robot A and robot B) to slowly move their end effector tool heads in three-dimensional space to a preset, permissible nearest safe distance threshold. ,For example Static images of multiple robots safely approaching the limit of their tool heads are simultaneously acquired and recorded using vision sensors. Ensure that the still image The logo pattern is covered in the middle.

[0043] S13: Based on static images and The feature point detection algorithm is used to locate the center pixel coordinates of all highly reflective marker points within each marker pattern in each image. The core task of this step is to perform stable identification and high-precision center positioning of each highly reflective marker point in each image. Step S13 includes the following steps: In the captured still images and In the process, high-quality image patches of the geometric structure pattern composed of highly reflective marker dots on each marker pattern are manually or automatically extracted as a reference template. ,in, r This is the robot's identification number. r It can be A or B. i This is the identification number for each highly reflective marker. Each reference template... It should include a complete, uniformly bright, highly reflective marker area and a small amount of surrounding background. For each reference template... This invention extracts a feature descriptor that is robust to changes in illumination, rotation, and scale. Preferably, the embodiment of this invention uses a histogram of oriented gradients (HARGs) descriptor based on grayscale information. The descriptor Used for calculating benchmark templates The gradient orientation histogram of a local region of an image is used to extract the overall geometric structure features formed by the spatial arrangement of all highly reflective markers on an identification pattern. For example, by calculating the directional distribution of the gradient in a local region, the orientation, side length ratio, and overall outline of the triangle formed by highly reflective markers on an identification pattern can be effectively characterized, thus using this geometric structure as the unique identifier of the robot.

[0044] Base template for all logo patterns and its corresponding descriptor The images are stored in a database to build a template library. Then, a global search and feature extraction are performed on the static images to be processed. (Right now and Using a sliding window or a region proposal method based on a brightness threshold, all possible candidate bright spot regions are initially identified. ,in, j This indicates the target sequence number identified during the actual monitoring process of the acquired images. For each candidate bright spot region... Extract the descriptor that is the same as the baseline template. Calculate sequentially. With all descriptors in the template library The similarity is determined using the nearest neighbor ratio method for matching. This identifies regions that are similar to candidate bright spots. If the ratio of the similarity distance between the first template with the highest similarity and the second template with the second highest similarity is lower than a preset empirical threshold (e.g., 0.8), then the candidate bright spot region is considered. Successfully matched the identity of the highly reflective marker corresponding to the first template. This step ensures that each detected bright spot area can be uniquely and correctly identified as a preset high-reflectivity marker. After coarse matching of the high-reflectivity marker identities, sub-pixel-level positioning is performed on the center pixel coordinates of the high-reflectivity markers to achieve a positioning accuracy better than 0.1 pixels. This embodiment of the invention uses a gray-scale weighted centroid method, extracting a local window of a preset pixel region centered on the coarse positioning result. For example, the preset pixel region can be 9×9 pixels. The gray-scale values ​​of all pixels within this preset pixel region are calculated as the centroid of the weights, and its center pixel coordinates... The calculation formula is: ; in, a , b These are the pixel coordinates within the local window. This is the grayscale value of that point.

[0045] Based on the above steps, the mapping relationship between the vision sensor coordinate system and the robot coordinate system is determined, and the reference templates for all marking patterns are extracted and stored through feature point detection algorithms and histogram of orientation gradients. and its descriptors To ensure that the system can stably recognize the identification patterns when it is put into operation; S14: Before the system is officially put into operation, a warning criterion based on image pixel distance is established through a one-time offline calibration process. The warning judgment is divided into two stages: close posture relationship calibration under close processing conditions and collision warning posture relationship calibration under extreme safety proximity conditions. Close posture relationship calibration is performed when multiple robots are operating in a simulated collaborative processing operation state, and when they reach a preset close processing posture, the vision sensor is triggered to acquire a static image. The pixel distance relationship between the marker patterns is then defined; the collision warning attitude relationship is defined by continuing to control multiple robots to approach to the limit of safe approach attitude (the corresponding three-dimensional spatial distance is set to the nearest safe distance threshold). (when the pixel bounding box touches or overlaps within 100mm), the vision sensor is triggered to acquire a static image. And mark the pixel distance relationship between the marker patterns at this time.

[0046] For different robots equipped with identification patterns, calculate still images separately. and In the diagram, the pixel distance between any pair of highly reflective marker points on two different marker patterns is defined. After iterating through all pairs of points, the maximum pixel distance value is found for each of the two states. and .

[0047] As the maximum pixel distance value under tight processing posture conditions, a square pixel bounding box is assigned to each marker pattern, denoted as . The physical meaning of this bounding box is: when multiple robots move to a preset close processing posture, the projection areas of the marking patterns of any two robots on the image plane are exactly adjacent or in critical contact. At this time, the safety control system should increase its attention to the relevant feature targets.

[0048] As the maximum pixel distance value under extreme safe approach posture conditions, another square pixel bounding box is defined for each marker pattern, denoted as . The physical meaning of this bounding box is: when the actual three-dimensional spatial distance between any two robots reaches a preset nearest safe distance threshold. If the projection areas of the markings of the two markings, which are 100mm or more, on the image plane will touch or overlap, the safety control system should immediately issue a collision warning.

[0049] During subsequent real-time online monitoring, the safety control system will determine the robot's pose relationship level based on the pixel bounding box status of the marker pattern in the current frame and execute the corresponding control strategy.

[0050] S2: When the robot drives the tool head to process, it controls the vision sensor to collect dynamic images and performs parallel tracking of all the marking patterns. It adopts a tracking algorithm based on discriminant correlation filtering to calculate the pixel coordinates of each marking pattern in each frame of synchronously acquired dynamic images and mark the pixel bounding box. When the pixel bounding boxes of any two marking patterns touch or overlap in the dynamic image, a primary warning event or a collision warning event is generated. The robot corresponding to the marking pattern that generates the primary warning event or collision warning event is designated as the robot involved in the incident, and the robot corresponding to the marking pattern that does not generate the primary warning event or collision warning event is designated as the robot not involved in the incident.

[0051] An independent visual tracking unit is set up for each feature target (i.e., the marker pattern and the calibration pattern) within the field of view of each visual sensor. A tracking algorithm based on discriminant correlation filtering is employed. In each frame of synchronously acquired dynamic image, the visual tracking units work in parallel, calculating the pixel coordinates of the marker pattern they are tracking in the current image. And mark the pixel bounding box. Center position coordinates: ; ; in, r The serial number representing the robot's identity. o This represents the current recognition tracking box record number. n The frame number representing the dynamic image. This represents the bounding box of the tracked marker pattern pixels. It represents the shortest pixel distance between two pixel borders and also outputs the tracking confidence.

[0052] Primary warning status judgment: When the shortest pixel distance between the real-time pixel bounding boxes of any two marker patterns... Less than or equal to When the system determines that the two robots are in a close processing posture, it generates a primary warning event. ,in, This indicates the warning event number at the current primary warning level, where A and B represent the robot's identity sequence number. This event is a sequence number containing an event trigger timestamp and information about the identified target. The system sends a data packet containing information such as the center pixel coordinates of the pattern and the current frame image. After a primary warning event is triggered, the vision sensor automatically increases its image acquisition frame rate to enhance its focus on the feature information of the robot involved. At the same time, it outputs the primary warning event to the control terminal, and the operator adjusts the robot's movement speed and movement strategy according to the actual situation until the robot returns to a safe posture, at which point the image acquisition frame rate of the vision sensor returns to normal.

[0053] Collision warning status determination: When the pixel bounding boxes of the identification patterns of any two different robots (robot A and robot B) are tracked in real time. and Contact or overlap occurs on the image, i.e., satisfying Under certain conditions, or The system determines that the two robots have reached their maximum safe approach posture and generates a collision warning event. ,in, This indicates the warning event number at the current safety collision warning level. The event also includes an event trigger timestamp and the identity of the involved target. The system displays information such as the center pixel coordinates of the pattern and the current frame image when triggered. The collision warning event will act as a wake-up signal, directly triggering and transmitting it to the second-level 3D geometric space precision adjudication layer, initiating a high-precision pose calculation and collision prediction process for the robot involved.

[0054] S3: The pose calculation frequency of the robot involved in the incident is dynamically adjusted, while the robot not involved in the incident maintains its basic pose at a preset basic frequency; the safety decision processing module uses an efficient perspective n-point algorithm to calculate the pose data of the robot involved in the incident in the workpiece coordinate system.

[0055] This step is the second level of the optical processing safety control method in this embodiment—the three-dimensional geometric space precise adjudication layer. This level is activated only after a primary warning event or collision warning event is triggered, and performs high-precision, high-frequency six-degree-of-freedom pose calculations on the robot involved. Step S3 includes the following steps: S31: When a primary warning event or a collision warning event is generated. Upon incident, the safety control system immediately switches from a low-load global monitoring state to a targeted high-precision calculation mode. The control logic shifts from a "low-power monitoring mode" for all feature targets to a "high-precision calculation mode" specific to the robot involved. For robots not involved, their pose calculation threads operate at a low frequency or are paused to optimize overall computing resources. Two key frequency parameters are preset: a preset base frequency... Used for maintaining the basic pose of a robot in a risk-free state; maximum solution frequency , is millihertz ( mHzThe frequency of pose calculation for the robots involved (Robot A and Robot B) is on the order of [number], synchronized with the highest frame rate of the vision sensor, for the highest risk response. This is used when generating a primary warning event or a collision warning event. Instead of being fixed, it is based on the relative approach speed of the two robots involved in the incident in three-dimensional space. Dynamic adjustments are made. The adjustment model is as follows: ; in, α This is the gain coefficient calibrated based on the delay of the safety control system and the dynamic characteristics of the robot. The current spatial coordinates of the robot involved in the incident are estimated by real-time calculation using visual tracking and an efficient perspective-n-point (EPnP) algorithm. The safety control system records the pose data of the robot involved in the incident from the previous frame image (denoted as...). ), and obtain the pose data calculated from the current frame image (denoted as ), ) ; This refers to the time interval between changes in the visual sensor frame rate. This adjustment model ensures tracking at a moderate pose calculation frequency (e.g., 10Hz~15Hz) when the two robots slowly approach each other (e.g., during precise alignment); and automatically increases to the maximum calculation frequency when the two robots rapidly approach each other (e.g., during unexpected rapid movement). Tracking enables dynamic matching of computing resources with real-time risk levels.

[0056] S32: After dynamically adjusting the pose calculation frequency, the feature point pose of the robot in question is calculated. First, high-precision matching of local image features is performed. Then, a feature point detection algorithm based on the image moment method is used to calculate the center pixel coordinates of each highly reflective marker in the robot's marking pattern. Sub-pixel-level localization is performed with an accuracy better than 0.1 pixels. Then, perspective-n-point (PnP) pose calculation is performed using a pre-measured point set of a 3D rigid body model. The EPnP algorithm, which is invoked in parallel with the pixel set containing the center pixel coordinates of all highly reflective marker points acquired synchronously from at least two different visual sensor perspectives, first calculates the pose transformation matrix of the marker pattern coordinate system relative to the coordinate systems of each visual sensor. Then, it combines the visual sensor coordinate system and the workpiece coordinate system obtained during the offline hand-eye calibration stage. Transformation matrix between Finally, the pose transformation matrices of the tool heads of robots A and B in the workpiece coordinate system are obtained. The pose data calculated by this process includes the three-dimensional positions of multiple tool heads in the workpiece coordinate system during collaborative machining. Location and Attitude angles and the coordinate transformation matrix between them .

[0057] In the visual sensor coordinate system Coordinate system of the marking pattern installed at the robot end effector In solving the pose transformation relationship between them, it is known that... n A reference point in the coordinate system of the logo pattern coordinates below for: ; in, , The highly reflective markings on the logo indicate its identity. These are fixed points integrated into the signage pattern, and their coordinates in the signage pattern coordinate system can be obtained directly through the pattern design.

[0058] Two-dimensional image projection coordinates of the reference point for: ; Intrinsic parameter matrix of visual sensor for: ; in, This represents the equivalent focal length of the vision sensor along the u-axis (horizontal direction), expressed in pixels. This represents the equivalent focal length of the vision sensor along the v-axis (vertical direction), expressed in pixels. Let be the pixel coordinates of the principal point of the image plane (the intersection of the optical axis and the imaging plane) along the u-axis. The pixel coordinates of the principal point of the image plane (the intersection of the optical axis and the imaging plane) in the v-axis direction. It is usually located approximately at the geometric center of the image resolution.

[0059] Based on the known 3D coordinates of reference points and the 2D image projection coordinates, four non-coplanar virtual control points are defined to linearly represent all reference points, thereby constructing a system of linear equations for solving the pose data. In the marker pattern coordinate system { S Four non-coplanar virtual control points are selected below, with coordinates as follows: ,in, , This is the serial number of the virtual control point. The virtual control point can be positioned by the robot's preset coordinate points.

[0060] Three-dimensional coordinates of each reference point This can be represented as a linear combination of these four virtual control points: ; Among them, the weighting coefficient It is a constant and satisfies the normalization constraint: ; Weighting coefficient The geometric relationship between the reference point and the virtual control point is predetermined.

[0061] Construct a perspective projection model and a system of linear equations for the visual sensor, and define the visual sensor coordinate system { C} Relative to the coordinate system of the logo pattern { S The rotation matrix of} is ， The translation vector is The reference point is in the visual sensor coordinate system. C Coordinates under} for: ; in, , This indicates the identity marker on the logo pattern, which features highly reflective markings.

[0062] Based on the perspective projection model of the visual sensor, the projection equation is: ; in, This is the depth scaling factor. Substitute the linear expression into it, and let We can obtain: ; Expanding the projection equations and eliminating the depth scaling factor, we obtain a system of linear equations: ; in, M 2 n A 12-dimensional coefficient matrix, which is composed of... n The projection constraint equations for each reference point are stacked vertically. Specifically, for each reference point, its projection constraint equations in the matrix are... M The middle occupies two consecutive rows, totaling 2 n Rows; the total number of columns in the matrix is ​​2. The specific form of the matrix is ​​determined by the intrinsic parameters of the visual sensor ( (and the normalized weighting coefficients of the reference point in the pattern coordinate system) Joint decision; x A multidimensional unknown vector containing four virtual control points in the visual sensor coordinate system { C The coordinates under}, specifically expanded as follows: ; This represents the set of four virtual control points in the visual sensor coordinate system. This represents specific coordinate data, where .

[0063] Using singular value decomposition to transform linear equations to obtain , The unknown vector is a right singular vector matrix. x Proportional to The last column of the matrix. The scale factor is recovered by constraining the known distances between the virtual control points, resulting in the coordinates of the four virtual control points in the visual sensor coordinate system { C The actual coordinates under} Finally, absolute orientation is solved based on the correspondence of virtual control points. Solve for the optimal rigid body transformation: .

[0064] Solving the matrix yields orthogonal matrix and column vectors The above two parts, combined into a 4×4 homogeneous transformation pose relation matrix, constitute the coordinate system of the marker pattern. S} relative to the visual sensor coordinate system { C pose transformation matrix of} : ; in, , .

[0065] pass = This allows us to obtain the tool head's position in the workpiece coordinate system. V The pose under}

[0066] Since the pose calculation algorithm is existing technology, it will not be described in detail in this invention.

[0067] S4: Construct a dynamic envelope model of the robots involved by combining pose data, and use a collision detection algorithm to calculate the shortest Euclidean distance between the robots in three-dimensional space in real time.

[0068] Let the markings of the robots involved (Robot A and Robot B) be relative to the workpiece coordinate system { V The transformation matrices of} are respectively and The corresponding rotation matrix and translation vector are respectively and .

[0069] The real-time coordinates of robot A are: ; The real-time coordinates of robot B are: ; Shortest Euclidean distance in three-dimensional space The calculation formula is: ; in, e This indicates the sequence number of the triggered event under the current working state (the state of calculating the pose of feature points for the robot involved); the estimated collision time is calculated by combining the relative approach speed of the robot involved; when the shortest Euclidean distance in three-dimensional space and / or the estimated collision time reach the deceleration warning threshold, motion intervention is performed on the robot involved; or, when the shortest Euclidean distance in three-dimensional space and / or the estimated collision time reach the emergency stop threshold, emergency braking is performed on the robot involved. Step S4 includes the following steps: S41: Combining the pose data from step S3 and the robot's kinematics model, construct a dynamic envelope model of the robots involved. Based on the dynamic envelope model, employ an efficient collision detection algorithm, specifically a fast detection algorithm based on axis-aligned bounding boxes (AABB), to calculate in real-time the shortest Euclidean distance in three-dimensional space between robots A and B. Based on the current relative approach speeds of robot A and robot B... Further calculation of the estimated collision time This provides key parameters for predictive security decisions.

[0070] Finally, based on the precisely calculated shortest Euclidean distance in three-dimensional space... Compared to the expected collision time The system executes hierarchical security decisions and outputs corresponding control commands.

[0071] Since the fast detection algorithm based on axis-aligned bounding box (AABB) is existing technology, it will not be described in detail in the embodiments of this invention.

[0072] S42: Set multi-level decision thresholds and output corresponding control commands.

[0073] The preset deceleration warning threshold is set to: distance threshold and / or time threshold When the shortest Euclidean distance in three-dimensional space and / or expected collision time If a potential interference risk is detected between robot A and robot B, a speed over-control command is generated and sent to robot A and robot B to intervene in their motion, instructing them to reduce their current motion speed by a preset ratio (e.g., to 50%) until the risk is eliminated.

[0074] The preset emergency stop threshold is set to a more stringent distance threshold. and / or time threshold When the shortest Euclidean distance in three-dimensional space and / or expected collision time When it is determined that robot A and robot B are about to collide, a global emergency stop signal is immediately generated to stop all robots. This signal interrupts the motion planning and servo drive of all robots in the safety control system with the highest priority, causing them to enter the braking state.

[0075] S5: The safety control system continuously monitors the shortest Euclidean distance in three-dimensional space between the robots involved. When the shortest Euclidean distance in three-dimensional space continues to exceed the preset safe distance threshold for a preset time, the risk is determined to be eliminated; the primary warning event or collision warning event is turned off, and the pose calculation frequency of the robots involved is reduced to the preset base frequency.

[0076] After issuing a motion intervention or emergency stop command, the safety control system continuously monitors the shortest Euclidean distance in three-dimensional space. When the shortest Euclidean distance in three-dimensional space The distance remains greater than the safe distance threshold in the initial warning. After a preset time (e.g., 60 seconds), the risk is determined to be eliminated. Subsequently, the primary warning event or collision warning event is disabled, and the calculation frequency of robot A and robot B is reduced to the preset base frequency. It also releases or suspends its computational threads, allowing the security control system to return to a low-power global monitoring mode.

[0077] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0078] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A multi-robot optical processing safety control system based on machine vision, characterized in that, include: A rotary worktable, wherein an optical element to be processed is fixed in the processing area of ​​the rotary worktable, and at least two calibration patterns are installed in the non-processing area of ​​the rotary worktable; At least two robots, each with a calibration plate and a tool head fixed to its end, the tool head being used to process the optical element to be processed; each tool head is equipped with an identification pattern; At least two vision sensors are used to simultaneously acquire robot posture images; each vision sensor is arranged at a preset height above the rotary table so that the robot is simultaneously within the effective field of view and depth of field of at least two vision sensors. The vision processing module is used to locate the static center pixel coordinates of the marking pattern based on the static robot posture image, and to mark the pixel bounding box. The target tracking module is used to perform parallel tracking of all the marked patterns in a dynamic robot pose image, calculate the dynamic center pixel coordinates of each marked pattern in real time, and mark the pixel bounding box. The safety decision processing module is used to generate a primary warning event or a collision warning event when the pixel bounding boxes of any two of the marked patterns overlap; reallocate the pose calculation frequency of the robot and trigger three-dimensional pose calculation; and determine whether the preset threshold condition is reached based on the real-time calculated shortest Euclidean distance in three-dimensional space and the estimated collision time. A control module is used to control the robot and the tool head to perform motion intervention or emergency braking.

2. The multi-robot optical processing safety control system based on machine vision according to claim 1, characterized in that, Both the calibration pattern and the marking pattern consist of at least three non-collinear highly reflective marker points.

3. The multi-robot optical processing safety control system based on machine vision according to claim 1, characterized in that, The vision processing module includes an image receiving unit and a pattern calibration unit; The image receiving unit is used to receive static robot posture images acquired by the vision sensor; The pattern calibration unit is used to calculate the static center pixel coordinates of each of the marked patterns in the static robot posture image and to calibrate the pixel bounding box.

4. The multi-robot optical processing safety control system based on machine vision according to claim 1, characterized in that, The target tracking module includes a target tracking unit and an image processing unit; The target tracking unit is used to track all the marked patterns in parallel in the dynamic robot posture image; The image processing unit is used to calculate the dynamic center pixel coordinates of each of the marked patterns in real time and to mark the pixel bounding box.

5. The multi-robot optical processing safety control system based on machine vision according to claim 1, characterized in that, The safety decision processing module includes an early warning unit, a pose calculation unit, and a safety decision unit. The warning unit is used to generate a primary warning event or a collision warning event when the pixel bounding boxes of any two of the marked patterns overlap, and the robot corresponding to the marked pattern that generates the primary warning event or the collision warning event is designated as the robot involved. The pose calculation frequency of the robot in question is reallocated, and the pose calculation unit is activated. The pose calculation unit is used to perform three-dimensional pose calculation on the robot involved in the incident, obtain the pose data of the robot involved in the incident, and calculate the shortest Euclidean distance in three-dimensional space and the estimated collision time between the robots involved in the incident in real time using the pose data. The safety decision unit is used to determine whether the shortest Euclidean distance in three-dimensional space and / or the expected collision time have reached a preset threshold condition, the preset threshold condition including a deceleration warning threshold or an emergency stop threshold; when the shortest Euclidean distance in three-dimensional space and / or the expected collision time reach the deceleration warning threshold, the control module is activated to perform motion intervention; when the shortest Euclidean distance in three-dimensional space and / or the expected collision time reach the emergency stop threshold, the control module is activated to perform emergency braking.

6. The multi-robot optical processing safety control system based on machine vision according to claim 1, characterized in that, The robot is a six-degree-of-freedom robotic arm.

7. A multi-robot optical processing safety control method based on machine vision, implemented based on the safety control system described in any one of claims 1 to 6, characterized in that, Includes the following steps: S1: Before the system runs, use the vision sensor to synchronously collect static images of multiple robots driving the tool head for close processing and static images of extreme safety approach. Use the feature point detection algorithm to locate the center pixel coordinates of all highly reflective marker points on all the marking patterns in the static images and mark the pixel bounding box. S2: After the system is running, control the vision sensor to synchronously acquire dynamic images of multiple robots driving tool heads for processing, perform parallel tracking of all the marking patterns, use a tracking algorithm based on discriminative correlation filtering to calculate the pixel coordinates of each marking pattern in each frame of synchronously acquired dynamic image, and mark the pixel bounding box. When the pixel bounding boxes of any two identifier patterns touch or overlap in the dynamic image, a primary warning event or a collision warning event is generated, and the robot corresponding to the identifier pattern that generates the primary warning event or the collision warning event is designated as the robot involved, while the robot corresponding to the identifier pattern that does not generate the primary warning event or the collision warning event is designated as the robot not involved. S3: The pose calculation frequency of the robot involved in the incident is dynamically adjusted, and the robot not involved in the incident maintains its basic pose at a preset basic frequency; the pose data of the robot involved in the incident in the workpiece coordinate system is calculated using an efficient perspective n-point algorithm. S4: Construct a dynamic envelope model of the robot involved in the incident based on the pose data, and use a collision detection algorithm to calculate the shortest Euclidean distance between the robots involved in the incident in three-dimensional space in real time, and solve the estimated collision time based on the relative approach speed of the robots involved in the incident. When the shortest Euclidean distance in the three-dimensional space and / or the expected collision time reach the deceleration warning threshold, motion intervention is performed on the robot involved; or, when the shortest Euclidean distance in the three-dimensional space and / or the expected collision time reach the emergency stop threshold, emergency braking is performed on all the robots.

8. The multi-robot optical processing safety control method based on machine vision according to claim 7, characterized in that, In step S1, the center pixel coordinates of the highly reflective marker point ( The formula for calculating ) is: ； in, These are the pixel coordinates within the pixel window. This corresponds to the grayscale value of the pixel.

9. The multi-robot optical processing safety control method based on machine vision according to claim 7, characterized in that, In step S3, the pose calculation frequency The calculation formula is: ； in,{ A , B } represents the identification number of the robot involved. To preset the base frequency, This is the maximum calculation frequency corresponding to the highest frame rate of the visual sensor; The gain coefficient is calibrated based on the delay and robot dynamics characteristics of the safety control system. The relative approach speed of the robot in the three-dimensional space is given by the value of 'V'.

10. The multi-robot optical processing safety control method based on machine vision according to claim 7, characterized in that, After step S4, step S5 is also included: the safety control system continuously monitors the shortest Euclidean distance in three-dimensional space between the robots involved. When the shortest Euclidean distance in three-dimensional space is continuously greater than the preset safe distance threshold for a preset time, it is determined that the risk has been eliminated; the primary warning event or collision warning event is turned off, and the pose calculation frequency of the robots involved is reduced to the preset base frequency.

Images