Intelligent gripper replacement control method and system using sensor fusion and automatic calibration algorithm

Abstract

An intelligent gripper replacement system for a part aligner proposed in one aspect of the present invention comprises: a hardware unit that aligns and moves parts using a gripper and replaces the gripper itself when necessary; a sensor unit including an image sensor that detects the part and a force-torque sensor that detects force and torque applied to the detachable gripper; and a processor unit that receives data from the image sensor and the force-torque sensor to determine the gripper to be replaced and instructs the replacement.

Description

Technology Field

[0001] The present invention relates to a part aligner gripper replacement system, and more specifically, to an intelligent gripper replacement system that determines a gripper to be replaced and directs the replacement. Background Technology

[0003] Today, component alignment machines are used in industrial settings to align bulk parts in a unidirectional manner before they are fed into automated product assembly processes. Furthermore, conventional component alignment machines are custom-designed to align parts precisely according to the shape or center of gravity of the client's components. In other words, a problem arose where conventional component alignment machines, once designed, could only align the parts originally intended for alignment and failed to properly align other parts.

[0004] Therefore, there was a limitation in that once the alignment of a specific part was completed, a new custom design tailored to that new part was required to align it. Consequently, existing part alignment machines could not be utilized after their use with a specific part was finished, which increased costs in the product assembly process.

[0005] Meanwhile, with the proliferation of multi-product, small-batch production systems, technologies for changing grippers at the ends of robot arms were being applied to stably grip and transport various parts during the process. Conventional part alignment machines manually replaced the grippers at the ends of the robot arms whenever a part was changed, which had the disadvantage of delaying production and increasing the risk of safety accidents for workers. Furthermore, since the selection of the optimal gripper for each part relied on the operator's experience, there was a lack of consistency and a persistent risk of part damage.

[0006] Therefore, there is a need for a gripper replacement system that can resolve issues of reduced productivity and quality fluctuations by performing gripper replacements quickly and reliably. The problem to be solved

[0008] The objective of the present invention is to solve the aforementioned problem and to propose a gripper replacement system with improved production efficiency by enabling the system to determine and replace the optimal gripper, which was previously replaced manually by a worker.

[0009] However, the problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned problems will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0011] To solve the aforementioned problem, an intelligent gripper replacement system for a part aligner proposed in one aspect of the present invention comprises: a hardware unit that aligns and moves parts using a gripper and replaces the gripper itself when necessary; a sensor unit including an image sensor that detects the part and a force-torque sensor that detects force and torque applied to the detachable gripper; and a processor unit that receives data from the image sensor and the force-torque sensor to determine the gripper to be replaced and instructs the replacement.

[0012] According to one embodiment, the hardware part includes a multi-joint robot arm, and the gripper is attached to one end of the multi-joint robot arm via a quick changer, and the quick changer may include a dual fixing means having an electromagnet element and a mechanical element.

[0013] According to one embodiment, the sensor unit further includes a Hall sensor and a position sensing sensor that detect the mounting status of the gripper in real time, and the processor unit may receive data from the Hall sensor and the position sensing sensor and determine whether the gripper instructed to replace is mounted.

[0014] According to one embodiment, the processor unit may calculate a Tool Center Point (TCP) for a replaced gripper and perform calibration to update a robot coordinate system based on the calculated TCP.

[0015] According to one embodiment, the part aligner may place a plurality of parts on a part alignment module including a checkerboard pattern and distribute and align them.

[0016] According to one embodiment, the gripper may be one of an adsorption gripper or a 2-finger type gripper.

[0017] According to one embodiment, the processor unit may determine the replacement target gripper by combining image sensor data and force-torque sensor data through sensor fusion to determine the characteristics of the part.

[0018] According to one embodiment, the sensor fusion may be performed by applying a Kalman filter.

[0019] According to one embodiment, the object derivation algorithm may be one of the YOLO (You Only Look Once) algorithm, a color derivation algorithm, and a template matching algorithm.

[0020] According to one embodiment, the processor unit may determine the gripper to be replaced after performing reinforcement learning in connection with a database in which data on the success or failure of work by part, force and torque data measured during gripping, and gripper replacement history data are accumulated. Effects of the invention

[0022] According to one embodiment of the present invention, there is an advantage of improving productivity and minimizing downtime by shortening gripper replacement time. In addition, there is an effect of reducing the part damage rate by selecting the optimal gripper based on artificial intelligence. Along with this, there is an advantage of reducing the risk of safety accidents by minimizing manual intervention by operators.

[0023] However, the effects of the present invention are not limited to those described above, but include all effects naturally realized through the various configurations proposed in the present invention. Brief explanation of the drawing

[0025] FIG. 1 is a block diagram of an intelligent gripper replacement system for a part aligner according to one embodiment of the present invention. FIG. 2 is a drawing of a component alignment device according to one embodiment of the present invention. FIG. 3 is a flowchart illustrating a method in which an intelligent gripper replacement system of a part aligner according to an embodiment of the present invention replaces a gripper and performs a part alignment operation. Specific details for implementing the invention

[0026] The embodiments of the present invention are illustrative for the purpose of explaining the technical concept of the present invention. The scope of rights according to the present invention is not limited to the embodiments presented below or the specific description thereof.

[0027] All technical and scientific terms used in this invention, unless otherwise defined, have the meaning generally understood by those skilled in the art to which this invention pertains. All terms used in this invention are selected for the purpose of further explaining this invention and are not selected to limit the scope of rights according to this invention.

[0028] Expressions such as "comprising," "having," "having," etc. used in the present invention should be understood as open-ended terms implying the possibility of including other embodiments, unless otherwise stated in the phrase or sentence containing such expressions.

[0029] Unless otherwise stated, singular expressions described in the present invention may include the meaning of the plural form, and this applies likewise to singular expressions described in the claims.

[0030] Embodiments of the present invention will be described below with reference to the attached drawings. In this process, the thickness of lines or the size of components depicted in the drawings may be exaggerated for clarity and convenience of explanation. Furthermore, in the description of the embodiments below, the description of identical or corresponding components may be omitted. However, even if a description of a component is omitted, it is not intended that such component is not included in any embodiment.

[0031] In addition, the following embodiments are not intended to limit the scope of the present invention but are merely exemplary details of the components presented in the claims of the present invention, and embodiments including components that are included in the technical concept throughout the specification of the present invention and are substitutable as equivalents for the components of the claims may be included in the scope of the present invention.

[0033] FIG. 1 is a block diagram of an intelligent replacement system for a parts aligner according to one embodiment of the present invention.

[0034] Referring to FIG. 1, an intelligent gripper replacement system for a part aligner proposed in one aspect of the present invention may include: a hardware unit (140) that aligns and moves parts using a gripper and replaces the gripper itself when necessary; a sensor unit (110) including an image sensor that detects the part and a force-torque sensor that detects force and torque applied to the detachable gripper; and a processor unit (150) that receives data from the image sensor and the force-torque sensor to determine the gripper to be replaced and instruct the replacement.

[0035] According to one embodiment, the intelligent gripper replacement system may further include an interface unit (120). The interface unit may enable a user to issue commands to the intelligent gripper replacement system and transmit the results of the system's operation to the outside. By using the interface unit, the user can access the system via a network to monitor and control the operation of the intelligent gripper replacement system. The interface unit may provide visualization of work status and performance indicators through a web-based dashboard, through which the user can check the status of the system in real time.

[0036] According to one embodiment, the hardware part includes a multi-joint robot arm, and the gripper is attached to and detached from one end of the multi-joint robot arm via a quick changer, and the quick changer may include a dual fixing means equipped with an electromagnetic element and a mechanical element. Accordingly, compared to a conventional gripper changer with a physical screw fixing method, the replacement time can be shortened and the fastening stability can be improved. In addition, the dual fixing structure allows for precise position adjustment, enabling stable gripping of micro-parts. The hardware part may not include special mechanical fixing parts other than those mentioned above, such as a lever clamp.

[0037] According to one embodiment, the sensor unit may include a camera (200). In addition, it may include various sensors for recognizing parts. The sensors may be, but are not limited to, radar sensors, position sensors, acceleration sensors, gyroscope sensors, laser sensors, and LiDAR sensors.

[0038] In addition, the processor unit may filter noise from the image captured by the camera and image sensor and determine the quality of the component using the Zhang-Suen algorithm.

[0039] The noise filtering described above may additionally include a filtering element that performs low-pass filtering on high-frequency noise of the captured image and removes the noise of the low-pass filtered image by Gaussian filtering. The Gaussian filtering may utilize a principle in which a mask value is determined by Gaussian coefficient values ​​to remove noise. The mask may refer to a kernel based on a Gaussian distribution.

[0040] The image that has undergone the Gaussian filtering described above can be thinned by the Zhang-Shuen algorithm. Subsequently, edge detection can be performed on the thinned image to obtain outline information. In addition, the Hough transform algorithm can be applied to the image from which the edges have been extracted to extract dimensional information of the target part. The quality can be determined by comparing the dimensional information with the previously stored target dimensional information of the target part.

[0041] According to one embodiment, the sensor unit further includes a Hall sensor and a position sensing sensor that detect the mounting status of the gripper in real time, and the processor unit may receive data from the Hall sensor and the position sensing sensor to determine whether the gripper instructed for replacement is mounted. As an example, the position sensing sensor may be a proximity sensor. When using the proximity sensor, the mounting status of the gripper can be detected quickly in a non-contact manner, which has the advantage of excellent durability and ensuring stability during repeated measurements.

[0042] According to one embodiment, the processor unit can monitor and control the gripper replacement status in real time by utilizing the ROS communication protocol. Accordingly, the scalability of the system can be improved by making it easy to add and replace sensors or control modules.

[0043] According to one embodiment, the processor unit may calculate a Tool Center Point (TCP) for a replaced gripper and perform calibration to update a robot coordinate system based on the calculated TCP. Accordingly, since there is no need for the user to perform calibration separately when replacing a gripper, there is an advantage in improving work efficiency and productivity in a multi-product, small-batch production environment.

[0044] The above TCP calculation method may include a reference pose setting step, a robot parameter loading step, a forward kinematics calculation step, a TCP inversion step, and a multi-pose verification step. The reference pose setting step may involve moving the robot to a specific reference pose when a gripper replacement is detected. Additionally, it may include forming a reproducible standard pose by setting target angles for each joint.

[0045] The above robot parameters may be Denavit-Hartenberg (DH) parameters that include link length, link twist angle, link offset, and joint rotation angle information. Accordingly, the kinematic characteristics of the robot can be reflected when calculating TCP. The above DH parameters can be used to generate a transformation matrix Ti for joint i.

[0046] The above forward kinematics calculation step may include a step of calculating a transformation matrix (Ti) corresponding to each joint and multiplying them sequentially to derive the overall transformation matrix. Subsequently, an end-effector position vector (P_end) can be extracted from the final matrix. For example, the overall transformation matrix can be expressed as T = T1(θ1) × T2(θ2) × T3(θ3) Х T4(θ4) Х T5(θ5)× T6(θ6). The end-effector position vector may represent the (0,3), (1,3), and (2,3) elements of the overall transformation matrix. That is, the top three elements of the last column in a 4x4 transformation matrix may represent 3D position coordinates (X,Y,Z).

[0047] The above TCP inversion step may derive an offset between the robot flange and the gripper end point by calculating the difference between the preset reference point coordinates (P_target) and the end effector position vector (P_end). For example, TCP can be calculated as TCP = P_target - P_end.

[0048] The above multi-pose verification step may involve setting multiple different reference poses and performing a robot parameter loading step, a forward kinematics calculation step, and a TCP inversion step. Subsequently, the average of the TCP values ​​calculated from the multiple different reference poses can be calculated to minimize measurement noise and error. For example, the multiple different reference poses may refer to three or four reference poses.

[0050] FIG. 2 is a drawing of a component alignment device according to an embodiment of the present invention. The component alignment device may be a device that places a plurality of components on a component alignment module (500), applies vibration to disperse and align them, and then supplies them to the outside through a multi-joint robot arm (300). At this time, the component alignment module may include a plate structure.

[0051] The above plate structure may include an elastic surface, and the vibration may be caused by a vibrating part directly applying physical vibration to the plate structure. Additionally, the vibration may be caused by air pressure generated by injecting air onto the elastic surface using an air blower, but is not limited thereto.

[0052] According to one embodiment, the part aligner may place a plurality of parts on a part alignment module including a checkerboard pattern and distribute and align them. For example, the checkerboard pattern may be formed on the elastic face. The part aligner may be used in an object recognition algorithm that recognizes information about the parts by photographing the elastic face.

[0053] In this case, the information recognized through the sensor may require periodic vision calibration due to distortion inherent in vision characteristics. However, if the checkerboard pattern is formed, the vision system can recognize the pattern and automatically calculate and correct camera distortion, thereby eliminating the need for separate vision calibration. Furthermore, the checkerboard pattern can be utilized for Hand-Eye Calibration to align the camera coordinate system with the robot coordinate system, thereby automating the process of updating the robot's work coordinate system. This simplifies the system configuration and reduces the time required for repetitive calibration.

[0054] According to one embodiment, the gripper may be either a suction gripper or a two-finger type gripper. The suction gripper may be advantageous for rapidly suctioning and transporting a part with a certain portion of its surface flat. Additionally, it may be suitable for application to parts that are lighter than a two-finger type gripper. On the other hand, a two-finger type gripper may facilitate gripping parts that have irregular shapes or are difficult to suction on the surface.

[0055] In this way, by using two types of grippers—suction type and 2-finger type—it is possible to handle a wide range of parts, from flat to irregularly shaped parts, without introducing additional complex multi-grip grippers or custom-made grippers. Consequently, system configuration and maintenance are simple, and production efficiency can be increased by speeding up replacement via a quick changer.

[0057] FIG. 3 is a flowchart illustrating a method in which an intelligent gripper replacement system of a part aligner according to an embodiment of the present invention replaces a gripper and performs a part alignment operation.

[0058] Referring to FIG. 3, the method of an intelligent gripper replacement system of a part sorter replacing a gripper and performing part sorting operations may include the steps of: inputting image sensor data (S110); deriving information of a target part using an object derivation algorithm (S120); querying a database (S130); determining a gripper to be replaced using a gripper matching algorithm (S140); verifying reliability (S150); replacing a gripper (S160); performing part sorting operations (S170); and accumulating training data (data on the success or failure of operations per part, force and torque data measured during gripping, and gripper replacement history data) in a database (S180).

[0059] According to one embodiment, the object derivation algorithm may be one of the YOLO (You Only Look Once) algorithm, a color derivation algorithm, and a template matching algorithm. Additionally, the object derivation algorithm may be applicable differently depending on the type and shape of the part.

[0060] The above YOLO algorithm may be a single-stage object extraction algorithm, which may have the characteristic of extracting objects faster than a dual-stage object extraction algorithm. In addition, the above color extraction algorithm may be an algorithm that defines colors using the HSV color space. Furthermore, the above template matching algorithm may be an algorithm that determines whether the target part included in the input image matches a pre-registered template of the target part. In addition, various unlisted vision algorithms may be additionally used to recognize the part itself and its arrangement state and to select the part to be moved.

[0061] The above image sensor data may include part images acquired through a camera, and the object extraction algorithm may utilize a CNN model-based vision algorithm, not limited to the YOLO (You Only Look Once) algorithm, color extraction algorithm, and template matching algorithm. Additionally, the processor unit may derive the shape, size, and material characteristics of the part by analyzing the image sensor data with the CNN model.

[0062] In one embodiment, the YOLO algorithm and the template matching algorithm may include a process of converting the image into a grayscale image in advance. This may be implemented for the purpose of reducing data throughput and increasing work speed. This process can be particularly useful in a process where a large number of parts are placed on a part alignment unit and processed, requiring the simultaneous processing of a large amount of data. Additionally, the application of the process of converting to a grayscale image may be determined based on the type and shape of the parts.

[0063] The step of deriving information of a target part using the object derivation algorithm may include a process of proposing a recommended order of the object derivation algorithm to the user based on the type and shape of the part, and the user determining the priority order of application of the object derivation algorithm.

[0064] The above recommendation ranking may vary depending on the type and shape of the parts. For example, when the product shape is complex and there is no distinct color difference between parts, the YOLO algorithm may be recommended first. Additionally, when the shape of the part is simple, the template matching algorithm may be recommended first. Furthermore, when there is a distinct color difference between parts, the color derivation algorithm may be recommended first.

[0065] The above algorithm does not stop at merely recommending the optimal first algorithm, but can additionally recommend second and third algorithms. In this case, the user can select one or more algorithms. If the user selects multiple algorithms from among various recommended algorithms, more accurate information regarding parts and their alignment status can be obtained, even if data processing is slightly delayed. This process of simultaneously selecting multiple algorithms can be utilized in processes where the downstream process is more complex and the speed of the upstream process for part supply can be relatively slow.

[0066] The interface unit may include a display screen. The recommendation algorithm may be displayed on the screen of the interface unit, and the user may determine the priority application order of the object extraction algorithm by referring to the recommendation ranking. However, the priority application order is not limited to the recommendation ranking and may be determined based on the user's judgment.

[0067] According to one embodiment, the deriving step of the part recognition method of the part alignment device may utilize a plurality of object derivation algorithms, and when the difference in result values ​​between the plurality of object derivation algorithms is greater than or equal to a preset threshold value, the step of applying vibration to the plate structure to redistribute the part and then returning to the step of photographing the part may be performed.

[0068] The above-mentioned preset threshold value may be entered in advance by the user. The threshold value may vary depending on the type and shape of the part, and may be determined through multiple experiments before the part alignment device is introduced into the mass production process.

[0069] If the difference between the above result values ​​exceeds a preset threshold, it can be determined that the reliability of the object extraction algorithm using the above-described captured image as an input image is not high. To use an image with high reliability, vibration can be applied to the plate structure to redistribute the components. An alarm can be sent to the user if the redistribution step is repeated a preset number of times. The number of times can be set by the user.

[0070] In addition, if the difference in the above result values ​​exceeds a preset threshold, the object extraction algorithm can be executed again to extract objects with high reliability. Since objects can be extracted with different probabilities even from the same image and there may be differences in the result values, the object extraction algorithm can be executed again without a redistribution process.

[0071] The information regarding the target part may include not only information about the external shape or arrangement of the part itself, but also information regarding whether the part is in contact with the dividing guide. Additionally, it may include information regarding whether the part is in contact with the edge of the part alignment module. When the part is in contact with the dividing guide or the part alignment module in this manner, the lifting operation of the robot arm described above may be difficult. Therefore, by considering the information regarding contact, the part to be moved by the robot arm can be selected. Through this, the probability of failure in gripping the part by the robot arm can be reduced.

[0073] According to one embodiment, the processor unit may determine the gripper to be replaced by determining the characteristics of the part through sensor fusion by combining image sensor data and the force-torque sensor data. As another example, the gripper matching algorithm used in the step (S140) of determining the gripper to be replaced using the gripper matching algorithm may refer to an algorithm that automatically selects and replaces the optimal gripper according to the physical properties and shape of the part by combining the force-torque sensor data and an AI vision system. The AI ​​vision system may be information about the target part derived from the image sensor data. Using such a gripper matching algorithm has the advantage of improving the problem where the selection of the optimal gripper for each part in the past relied on the operator's experience, resulting in a lack of consistency and a risk of part damage.

[0074] The above gripper matching algorithm may be an algorithm that analyzes part characteristics and matches the optimal gripper using a deep learning model. For example, the above deep learning model may be a deep learning model based on a framework such as TensorFlow or PyTorch, but is not limited thereto.

[0075] For example, the sensor fusion described above may be performed by applying a Kalman filter. The force-torque sensor data may contain noise as it is sensitive to minute changes in gripping force, and the image sensor data may be susceptible to noise caused by environmental factors such as lighting, reflection, and image processing speed. The Kalman filter can compensate for these noises by predicting the state over time and correcting new measurements to produce an optimal estimate. In addition, the Kalman filter has the advantage of being more suitable for multi-product, small-batch production environments due to its higher computational efficiency compared to other sensor fusion methods.

[0076] By performing sensor fusion using the Kalman filter described above, the gripper can be automatically adjusted to the optimal gripping force for each part. In addition, by reflecting feedback from the pressure sensor, damage to the part can be prevented and gripping safety can be improved.

[0077] According to one embodiment, the processor unit may determine the gripper to be replaced after performing reinforcement learning in connection with a database in which data on the success or failure of work by part, force and torque data measured during gripping, and gripper replacement history data are accumulated.

[0078] According to one embodiment, the reinforcement learning may be performed based on a Markov Decision Process (MDP) defined by a State, an Action, and a Reward. The State may include the shape, size, and material characteristics of a part extracted from an image sensor, and gripping force data detected from a force / torque sensor. Additionally, it may include the type of currently mounted gripper and data on the success or failure of the operation for each part.

[0079] The action can be defined as selecting either a suction gripper or a 2-finger gripper, or adjusting the gripping force of the selected gripper. Compensation can be set based on the success of the operation and replacement efficiency. For example, a positive (+) compensation can be given for successful gripping, and a negative (-) compensation can be given for gripping failure or part damage. In addition, a penalty can be imposed if gripper replacement occurs more than a certain number of times during continuous operation of the same part.

[0080] Through such a reinforcement learning structure, the processor unit can increase the likelihood of selecting the optimal gripper for each part through iterative learning, and has the advantage of being able to autonomously adapt to various part shapes and environmental changes.

[0081] According to one embodiment, the processor unit may include a metadata management system linked to the database and an ETL pipeline. The ETL pipeline may collect sensor data from the sensor unit, transform the collected sensor data through a preprocessing process, and then accumulate it in the database. The preprocessing process may refer to operations such as coordinate system transformation, noise removal, and unit standardization. For example, the learning speed and performance can be improved by standardizing the units and coordinate systems of the sensor data through the ETL pipeline and inputting them into a reinforcement learning algorithm.

[0082] According to one embodiment, the processor unit may use a script-based robot control method. Additionally, it may process image sensor data in real time and recognize parts using an image processing library. The script may be Python and the image processing library may be OpenCV, but is not limited thereto.

[0084] The foregoing description is merely an illustrative explanation of the technical concept of the present invention, and those skilled in the art to which the present invention pertains may make various modifications and variations within the scope of the essential characteristics of the present invention. Accordingly, the embodiments disclosed in the present invention are intended to explain, not limit, the technical concept of the present invention, and the scope of the technical concept of the present invention is not limited by these embodiments. The scope of protection of the present invention shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present invention. Explanation of the symbols

[0086] 100: Intelligent gripper system for part sorters 200: Camera 300: Multi-jointed robotic arm 310: Gripper 400: Parts 500: Part Alignment Module

Claims

Claim 1 An intelligent gripper replacement system for a part aligner, comprising: a hardware unit that aligns and moves parts using a gripper and replaces the gripper itself when necessary; a sensor unit including an image sensor that detects the part and a force-torque sensor that detects force and torque applied to the detachable gripper; and a processor unit that receives data from the image sensor and the force-torque sensor, determines the characteristics of the part through sensor fusion, determines the gripper to be replaced, and instructs the replacement. Claim 2 An intelligent gripper replacement system for a part aligner according to claim 1, wherein the hardware part includes a multi-joint robot arm, the gripper is detachably attached to one end of the multi-joint robot arm via a quick changer, and the quick changer includes a dual fixing means having an electromagnet element and a mechanical element. Claim 3 An intelligent gripper replacement system for a parts aligner, wherein, in paragraph 2, the sensor unit further includes a Hall sensor and a position sensing sensor that detect the mounting status of the gripper in real time, and the processor unit receives data from the Hall sensor and the position sensing sensor and determines whether the gripper instructed to be replaced is mounted. Claim 4 An intelligent gripper replacement system for a part aligner, wherein, in claim 1, the processor unit calculates a Tool Center Point (TCP) for a replaced gripper and performs calibration to update a robot coordinate system based on the calculated TCP. Claim 5 An intelligent gripper replacement system for a parts sorter, wherein, in claim 1, the parts sorter places a plurality of parts on a parts sorting module including a checkerboard pattern and distributes and sorts them. Claim 6 An intelligent gripper replacement system for a part aligner, wherein, in claim 1, the gripper is one of a suction gripper or a 2-finger type gripper. Claim 7 delete Claim 8 An intelligent gripper replacement system for a part aligner, wherein, in claim 1, the sensor fusion is performed by applying a Kalman filter. Claim 9 An intelligent gripper replacement system for a part sorter, wherein, in claim 1, the processor unit derives information of a target part using an object derivation algorithm, and the object derivation algorithm is one of the YOLO (You Only Look Once) algorithm, a color derivation algorithm, and a template matching algorithm. Claim 10 An intelligent gripper replacement system for a part sorter, wherein, in claim 1, the processor unit determines the gripper to be replaced after performing reinforcement learning connected to a database in which data on the success or failure of work by part, force and torque data measured during gripping, and gripper replacement history data are accumulated.

Images