Automatic deviation correction mechanism for material grabbing and operation control method

Abstract

The present invention proposes an automatic deviation-correcting mechanism for material grabbing and an operation control method. The mechanism is provided with a frame, an input conveyor belt and an output conveyor belt are installed inside the frame, movable guide rails are arranged on both sides of the upper part of the frame, a terminal moving platform is arranged between the movable guide rails, symmetrical structures are arranged on both sides of the terminal moving platform and are installed and connected to the movable guide rails, and the structure is that a connecting rotating rod, a parallelogram mechanism long rod and a T-shaped connecting piece are sequentially installed at both ends of the terminal moving platform, a connecting long rod and a connecting side rod are respectively installed at both ends of the T-shaped connecting piece, the other end of the connecting long rod is installed on a front-end moving driving pair, and a connecting short rod and a rear-end moving driving pair are sequentially installed at the other end of the connecting side rod, and a suction cup and a video detection module are installed on the lower side of the terminal moving platform; the operation algorithm of the mechanism adopts a deep Q network improved by a TensorFlow Lite hardware acceleration library to realize the neat discharge of scattered food or medicine on the input conveyor belt onto the output conveyor belt.

Description

Technical Field

[0001] The invention relates to the technical field of automatic material grabbing and deviation-correcting equipment, and in particular to an automatic deviation-correcting mechanism for material grabbing and an operation control method. Background Art

[0002] In actual processing and production, automated mechanical equipment can save a lot of labor and effectively reduce production costs. However, for some small and medium-sized enterprises, the cost of replacing a full set of automated mechanical equipment is high, which makes it difficult for small factories or factories with a large number of semi-automatic mechanical equipment to achieve production automation in a short period of time. To solve this dilemma, some supporting auxiliary production equipment can be introduced to improve production efficiency by partially realizing automation. At present, the more common solution is to introduce some mechanical arms or robots to automate certain processes and processing steps. The coverage of this method is small, and the mechanical arms are limited by high-precision reducers and control motors, and the cost is still high. For this reason, the applicant has designed an automatic deviation correction mechanism for material grabbing according to the later sorting and packaging needs of industries such as food and medicine. The automatic deviation correction mechanism for material grabbing can quickly complete the messy state of food or medicine sent to the conveyor belt after the sorting mechanical equipment is completed. In order to facilitate the subsequent processing, packaging and sorting work, the automatic deviation correction mechanism for material grabbing adopts a video detection module with high recognition accuracy and high sorting efficiency. It replaces high-precision parts and reducers through clever structures and algorithms to achieve local production automation at a lower cost. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes an automatic deviation correction mechanism for material grabbing and an operation control method, by setting a frame, and installing an input conveyor belt and an output conveyor belt inside the frame, movable guide rails are arranged on both sides of the upper part of the frame, and an end moving platform is arranged between the movable guide rails, and symmetrical structures are arranged on both sides of the end moving platform to be installed and connected to the movable guide rails; one side of the symmetrical structure is described as: a connecting rotating rod, a parallelogram mechanism rod and a T-shaped connecting piece are installed at both ends of the end moving platform in sequence, and a connecting long rod and a connecting side rod are installed at both ends of the T-shaped connecting piece respectively; the other end of the connecting long rod is installed on the front end moving drive pair, and the other end of the connecting side rod is installed with a connecting short rod and a rear end moving drive pair in sequence, the front end moving drive pair and the rear end moving drive pair on the same side are installed on the same guide rail, and a suction cup and a video detection module are installed on the lower side of the end moving platform; the mechanism operation algorithm adopts the deep Q network improved by the TensorFlow Lite hardware acceleration library to realize the neat discharge of scattered food or medicine on the input conveyor belt to the output conveyor belt.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] An automatic deviation-correcting mechanism for grabbing materials comprises a frame, foot pads, an input conveyor belt, an output conveyor belt, a moving guide rail, a front-end moving drive pair, a rear-end moving drive pair, a connecting long rod, a connecting short rod, a connecting side rod, a T-shaped connector, a parallelogram mechanism long rod, a terminal moving platform, a suction cup, a video detection module, a connecting rotating rod and a linear motor, characterized in that: the automatic deviation-correcting mechanism for grabbing materials is provided with a frame, foot pads are provided at the bottom of four groups of supporting feet of the frame, an input conveyor belt and an output conveyor belt are installed inside the frame, moving guide rails are provided on both sides of the upper part of the frame, a terminal moving platform is provided between the two groups of moving guide rails, the terminal moving platform is located on the upper part of the input conveyor belt and the output conveyor belt, and symmetrical structures are provided on both sides of the terminal moving platform to connect to the moving platform On the moving guide rail, the symmetrical structure at both ends of the terminal moving platform is described as a connecting rotating rod installed on the upper part of both ends of the terminal moving platform, and two parallelogram mechanism long rods are installed on the connecting rotating rod, and the other ends of the parallelogram mechanism long rods are respectively connected to the upper part of the T-shaped connecting piece, and the two ends of the lower side of the T-shaped connecting piece are respectively installed with a connecting long rod and a connecting side rod, and the other end of the connecting side rod is installed with a connecting short rod, the other end of the connecting long rod is connected to the front end moving drive pair, and the other end of the connecting short rod is connected and installed on the rear end moving drive pair, the front end moving drive pair and the rear end moving drive pair are installed on the same moving guide rail on one side, and linear motors are installed on the front end moving drive pair and the rear end moving drive pair; a suction cup and a video detection module are installed on the lower side of the terminal moving platform.

[0006] Furthermore, the automatic deviation-correcting mechanism for material grabbing is provided with a terminal moving platform, a suction cup and a video detection module are installed on the lower side of the terminal moving platform, and the video detection module is provided with a high-definition camera.

[0007] Furthermore, the operation algorithm of the automatic deviation correction mechanism for material grabbing adopts a deep Q network improved by the TensorFlow Lite hardware acceleration library, which is used to neatly arrange scattered food or medicine on the input conveyor belt to the output conveyor belt; the algorithm automatically adjusts parameters such as the angle of grabbing and placing through continuous trial and error and learning to achieve the optimal grabbing and placing effect; specifically, the following steps are included:

[0008] S1. Deploy data collection and detection equipment;

[0009] By deploying a data acquisition and detection device on the material grabbing mechanism and establishing a cloud network management system to monitor the status of the monitoring device and data transmission;

[0010] S2, data preprocessing;

[0011] By establishing a cloud storage system, the data streams of various monitoring devices in the material grabbing mechanism are received and stored, and then the motion data, motor speed data, grabbing data and other sequences are generated; the sequence data is deeply learned by generating adversarial networks (GANs), and the batch normalization technology is used to optimize the data acquisition model of the automatic correction mechanism;

[0012] S3, model design and training;

[0013] After obtaining the pre-processed grabbing mechanism sequence data of each room, the deep Q network DQN algorithm is used to train the data model; the S3 model design and training includes the following sub-steps:

[0014] S31, determining the state space and the action space;

[0015] S32, define the neural network structure;

[0016] S33, define loss function and optimizer;

[0017] S34, start training;

[0018] S35, testing and optimization;

[0019] S36, model saving and loading;

[0020] S4, Improved deep Q network model construction;

[0021] Through the improved deep Q network model, we can better judge whether all subsequent running states meet the perfect conditions. If they meet the conditions, they can be released. If not, they can be adjusted immediately. In addition, this application proposes to use the TensorFlowLite hardware acceleration library to improve the performance of the DQN algorithm, and use the algorithm conversion tuning formula to improve the accuracy and efficiency of the calculation results. Finally, the improved deep Q network function is used to complete the capture feedback adaptive adjustment optimization work.

[0022] Furthermore, the GAN loss function formula in S2 of the operation algorithm of the automatic deviation correction mechanism for material grabbing is expressed as:

[0023] L(D,G)=E[D(x)]-E[G(z)]+max(E[D(G(z))])

[0024] Among them, G represents the generator and D represents the discriminator, which are used to generate and judge various data sets sent by the correction mechanism; z~N(0,σ 2 ) is a random noise vector, σ is the standard deviation of the Gaussian distribution, ε is a very small positive number used to prevent division by zero; E[] represents the expected value; max() represents the maximum value.

[0025] Further, the algorithm conversion and tuning formula in S4 of the operation algorithm of the automatic deviation correction mechanism for material grabbing is expressed as:

[0026]

[0027] Among them, p(a|s,a) is the probability distribution of executing grasping action a in the current state s, and w h is a weight matrix used to store information about the state and hidden layer; b is a bias vector used to control the nonlinear activation function during forward propagation; g(s') is a reward function used to evaluate the value of the current state and is adjusted according to the adaptive experimental results of the grasping feedback;

[0028] The improved deep Q network function in step S4 is expressed as:

[0029]

[0030] Among them, r(s,a) is the reward function, which is used to evaluate the operational perfection of the current material grasping and correction mechanism; γ is the operator parameter used for correction; q(s,a) is the actual grasping action probability distribution.

[0031] The invention provides an automatic deviation-correcting mechanism for material grabbing. The mechanism comprises a frame, an input conveyor belt and an output conveyor belt are installed inside the frame, movable guide rails are arranged on both sides of the upper part of the frame, a terminal moving platform is arranged between the movable guide rails, and symmetrical structures are arranged on both sides of the terminal moving platform to be installed and connected to the movable guide rails. The structure is that a connecting rotating rod and a parallelogram mechanism long rod are sequentially installed at both ends of the terminal moving platform, the parallelogram mechanism long rod is connected to the upper end of a T-shaped connecting piece, a connecting long rod and a connecting side rod are respectively installed at both sides of the lower end of the T-shaped connecting piece, the other end of the connecting long rod is installed on a front-end moving driving pair, and a connecting short rod and a rear-end moving driving pair are sequentially installed at the other end of the connecting side rod; the front-end moving driving pair and the rear-end moving driving pair on the same side are installed on the same moving guide rail, and a suction cup and a video detection module are installed on the lower side of the terminal moving platform; the mechanism operation algorithm adopts a deep Q network improved by the TensorFlow Lite hardware acceleration library to realize the orderly discharge of scattered food or medicine on the input conveyor belt to the output conveyor belt, which brings the following benefits:

[0032] 1. The automatic deviation-correcting mechanism for material grabbing has a simple structure and low cost. It is specially designed for the finishing and packaging of food and medicine, etc. It has high working efficiency and strong applicability.

[0033] 2. The automatic deviation correction mechanism for material grabbing adopts a video detection module, which can identify food and medicines with high precision, so as to achieve accurate grabbing and sorting and placement;

[0034] 3. The operating algorithm of the automatic deviation-correcting mechanism for material grabbing uses a deep Q network improved by the TensorFlow Lite hardware acceleration library to neatly discharge scattered food and medicine on the input conveyor belt to the output conveyor belt, with fast operation speed and high efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the overall structure of the present invention;

[0036] Figure 2 It is a schematic diagram of the working of the present invention;

[0037] Figure 3 is a system flow chart of the algorithm of the present invention;

[0038] Figure 4 This is a model training flow chart of the algorithm of the present invention;

[0039] Figure 5 This is an algorithm optimization architecture diagram of the algorithm of the present invention.

[0040] The markings in the figure are: 1. Frame; 2. Foot pad; 3. Input conveyor belt; 4. Output conveyor belt; 5. Moving guide rail; 6. Front end moving drive pair; 7. Rear end moving drive pair; 8. Connecting long rod; 9. Connecting short rod; 10. Connecting side rod; 11. T-shaped connector; 12. Parallelogram mechanism long rod; 13. End moving platform; 14. Suction cup; 15. Video detection module; 16. Connecting rotating rod; 17. Linear motor. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0042] like Figure 1 Figure 2 As shown: As shown is an automatic deviation correction mechanism for material grabbing, including a frame 1, a foot pad 2, an input conveyor belt 3, an output conveyor belt 4, a moving guide rail 5, a front end moving drive pair 6, a rear end moving drive pair 7, a connecting long rod 8, a connecting short rod 9, a connecting side rod 10, a T-shaped connector 11, a parallelogram mechanism long rod 12, an end moving platform 13, a suction cup 14, a video detection module 15, a connecting rotating rod 16 and a linear motor 17, as shown Figure 1As shown, the automatic deviation correction mechanism for material grabbing is provided with a frame 1, and the frame 1 is made of welded steel and has a frame structure. Foot pads 2 are provided at the bottom of the four groups of supporting feet of the frame 1, and the foot pads 2 are provided with adjusting bolts to facilitate the installation and adjustment of the frame 1. An input conveyor belt 3 and an output conveyor belt 4 are installed inside the frame 1. The input conveyor belt 3 and the output conveyor belt 4 can be newly installed or can be the conveyor belts of the original factory. Moving guide rails 5 are provided on both sides of the upper part of the frame 1, and an end moving platform 13 is provided between the two groups of moving guide rails 5. The end moving platform 13 is located on the upper part of the input conveyor belt 3 and the output conveyor belt 4. Symmetrical structures are provided on both sides of the end moving platform 13 and are connected to the moving guide rails 5. The structure at both ends of the end moving platform 13 is that a connecting rotating rod 16 is installed on the upper part of both ends of the end moving platform 13, and a parallelogram mechanism long rod 12 is installed on the connecting rotating rod 16. The other end of the parallelogram mechanism long rod 12 is installed on the upper end of the T-shaped connecting piece 11, and the two ends of the T-shaped connecting piece 11 are installed A connecting long rod 8 and a connecting side rod 10 are respectively installed at the ends, and a connecting short rod 9 is installed at the other end of the connecting side rod 10. One end of the connecting long rod 8 is connected and installed on the front end mobile drive pair 6, and one end of the connecting short rod 9 is connected and installed on the rear end mobile drive pair 7. The front end mobile drive pair 6 and the rear end mobile drive pair 7 are installed on the same moving guide rail 5. A linear motor 17 is installed on the outer side of the front end mobile drive pair 6 and the rear end mobile drive pair 7. By controlling the linear motor 17 of the front end mobile drive pair 6 and the rear end mobile drive pair 7 on the guide rail 5, the movement of the terminal moving platform 13 in the space above the input conveyor belt 3 and the output conveyor belt 4 is realized; a suction cup 14 and a video detection module 15 are installed on the lower side of the terminal moving platform 13. The suction cup 14 can easily absorb food or medicine, and then adjust the direction and position of the food or medicine. The video detection module 15 is provided with a high-definition camera, which can quickly identify the orientation information of food or medicine through the high-definition camera, obtain the data that needs to be adjusted, and realize rapid sorting.

[0043] like Figure 3-5 As shown in the figure, the operation algorithm of the automatic deviation correction mechanism for material grasping adopts the deep Q network improved by the TensorFlow Lite hardware acceleration library, which is used to neatly arrange the scattered food or medicine on the input conveyor belt to the output conveyor belt; the algorithm automatically adjusts the grasping force, grasping angle and other parameters through continuous trial and error and learning to achieve the best grasping effect; Figure 3 As shown, the specific steps include:

[0044] S1. Deploy data collection and detection equipment;

[0045] A data acquisition and detection device is deployed on the material grasping mechanism, and a cloud network management system is established to monitor the status of the monitoring device and data transmission. Specifically, sensors or cameras are installed on the grasping mechanism to monitor the motion trajectory, motion speed, motor speed and other information in real time during the grasping process. These data will be used as input for subsequent model training and algorithm improvement. Using Python's Pytorch package, data acquisition software or interfaces are developed to automatically collect these data sequences and store them in a database or file. Then, according to the set goals and scope, the number and diversity of samples required for the data are determined. Ensure that the data set contains enough samples to cover different design variable combinations and indicator ranges. To explain the content of data collection in more detail, consider the following aspects of data:

[0046] 1) Motion trajectory data: The motion trajectory of the object during the grasping process is obtained in real time through the sensor or camera installed on the grasping mechanism; the motion trajectory includes the position coordinates of the object or the contour information of the object on the two-dimensional plane;

[0047] 2) Movement speed data: By measuring the movement speed of the grasping mechanism, the movement state of the object during the grasping process can be understood; the movement speed will be obtained through the acceleration sensor;

[0048] 3) Motor speed data: By measuring the motor speed, the power output of the gripping mechanism can be understood; the motor speed will be obtained through the encoder device;

[0049] 4) Grasping force data: By measuring the grasping force of the grasping mechanism, we can understand the magnitude of the force applied during the grasping process; the grasping force will be obtained through the weighing sensor measurement equipment;

[0050] 5) Grasping time data: By measuring the time required for the grasping mechanism to complete a grasp, the efficiency of the grasping process can be understood; the grasping time will be obtained through a timer measurement device;

[0051] S2, data preprocessing;

[0052] By establishing a cloud storage system, the data streams of various monitoring devices in the material grasping mechanism are received and stored, thereby generating sequences such as motion data, motor speed data, grasping data, etc.; the sequence data is deeply learned by generating adversarial networks (GANs), and the batch normalization technology is used to optimize the data acquisition model of the automatic correction mechanism; the original sequence data of each device in the cloud storage system is processed to eliminate environmental interference and outliers; this application adopts the generative adversarial network (GAN) for sequence data processing; GAN is a powerful generative model used to generate synthetic data similar to real data; in grasping feedback regulation, GAN is used to generate simulated data with diversity and robustness; by training the GAN model, a series of simulated data of grasping scenarios will be generated to enhance the diversity and generalization ability of the data set; in the data preprocessing step, GAN can be used to model and optimize sequence data; the following is a detailed description of the entire process:

[0053] 1) Data cleaning, removing outliers and noise data;

[0054] 2) Feature extraction: extracting features related to grasp feedback regulation from the raw data;

[0055] 3) Data normalization: normalize the extracted features to make the numerical ranges of different features consistent;

[0056] 4) Data enhancement: transforming or expanding the original data through certain methods to increase the diversity and robustness of the data;

[0057] 5) Data partitioning: divide the preprocessed data into training set, validation set and test set for model training, tuning and evaluation;

[0058] Finally, TF-IDF technology is used for feature extraction and encoding; the encoded input sequence is input into the GAN model for model operation; appropriate features are extracted according to the characteristics of the problem; for each parameter of the material grabbing automatic deviation correction device proposed in this application, its motion trajectory data, motor speed, grasping force and other data features are extracted;

[0059] The algorithm uses batch normalization technology to optimize the GAN algorithm for the preprocessing optimization process of the overall correction mechanism data; the input sequence of the data is simply represented as x1, x2, x3, and the output sequence is represented as y1, y2, y3; this application proposes to use batch normalization technology to optimize the data acquisition model of the automatic correction mechanism; the loss function definition of the optimized GAN algorithm is given by the following formula:

[0060] L(D,G)=E[D(x)]-E[G(z)]+max(E[D(G(z))])

[0061] Among them, G represents the generator and D represents the discriminator, which are used to generate and judge various data sets sent by the correction mechanism; z~N(0,σ 2 ) is a random noise vector, σ is the standard deviation of the Gaussian distribution, ε is a small positive number used to prevent division by zero; E[] represents the expected value; max() represents the maximum value;

[0062] The loss function consists of two parts: the first part is the expected value of the probability distribution of the real data being correctly classified, and the second part is the expected value of the probability distribution of the generated data being misclassified. By minimizing this loss function, the purpose of improving the ability of the training generator G and the discriminator D to recognize data is ultimately achieved;

[0063] S3, model design and training;

[0064] like Figure 4 As shown, after obtaining the pre-processed grasping mechanism sequence data of each room, the deep Q network DQN algorithm is used to train the data model. The grasping process can be regarded as a Markov decision process (MDP), in which the state includes the current grasping position, grasping force, grasping angle, etc., the action includes adjusting the grasping force, adjusting the grasping angle, etc., and the reward includes feedback information in the two cases of grasping success and grasping failure. In order to achieve adaptive adjustment, a deep reinforcement learning algorithm can be used. This application uses a deep Q network, namely a DQN network. Before training, the pre-processed sequence data needs to be prepared. The S3 model design and training includes the following sub-steps:

[0065] S31, determining the state space and the action space;

[0066] S32, define the neural network structure;

[0067] In the adaptive adjustment of grasping feedback, the state space includes the current grasping position, grasping force, grasping angle, etc., and the action space may include adjusting the grasping force, adjusting the grasping angle, etc. This application converts these states and actions into discrete digital representations as the input of the DQN algorithm;

[0068] S33, define loss function and optimizer;

[0069] The core of the DQN algorithm is a neural network structure, which consists of an input layer, several hidden layers and an output layer. The input layer receives discrete digital representations of states or actions. The hidden layer is used to extract and transform features of states or actions. The output layer generates the next state-action decision based on these features and the current reward value.

[0070] S34, start training;

[0071] When training the DQN algorithm, we use experience replay technology to improve training efficiency. Specifically, we store some experiences first, and then continuously sample samples from these experiences during the training process. This can avoid repeated training of the same samples, thereby reducing training time. At the same time, the target network will also help us train better. It maps the current state-action decision to the next state-action decision, thereby guiding the model to learn.

[0072] S35, testing and optimization;

[0073] When the training is completed, use the test set to evaluate the performance of the DQN algorithm; if the performance is poor, try to adjust the neural network structure, loss function, optimizer and other parameters to optimize the model;

[0074] S36, model saving and loading;

[0075] During the training process, the model's weights and bias parameters are saved regularly to avoid the computational overhead of training the model from scratch each time. The saved parameter files are stored in a hard disk or other persistent storage medium. When using the trained model later, the saved parameter files can be loaded into the new model for inference.

[0076] S4, Improved deep Q network model construction;

[0077] Through the improved deep Q network model, we can better judge whether all the states of subsequent operation meet the perfect conditions. If they meet the conditions, they can be released, and if they do not meet the conditions, they can be adjusted immediately. In addition, this application proposes to use the TensorFlowLite hardware acceleration library to improve the performance of the DQN algorithm, and use the algorithm conversion tuning formula to improve the accuracy and efficiency of the calculation results. Finally, the improved version of the deep Q network function is used to complete the optimization of the adaptive adjustment of the capture feedback. Figure 5 As shown, for the improvement of the deep Q network, a hardware acceleration library will be used from the hardware level; this application decided to use the TensorFlowLite hardware acceleration library to improve the performance of the DQN algorithm; it is a deep learning framework specifically for mobile devices and embedded devices, supporting a variety of hardware accelerators, including CPU, GPU and TPU, etc.; through TensorFlow Lite, the DQN algorithm is converted into efficient code suitable for mobile devices, thereby improving performance and response speed, and then adaptively adjusting the accuracy of material grasping feedback faster; in this process, the algorithm conversion formula is as follows: The algorithm conversion tuning formula in S4 of the running algorithm of the automatic correction mechanism for material grasping is shown as:

[0078]

[0079] Among them, p(a|s,a) is the probability distribution of executing grasping action a in the current state s, and w h is a weight matrix used to store information about the state and hidden layer; b is a bias vector used to control the nonlinear activation function during forward propagation; g(s') is a reward function used to evaluate the value of the current state and is adjusted according to the adaptive experimental results of the grasping feedback;

[0080] The improved deep Q network function in step S4 is expressed as:

[0081]

[0082] Among them, r(s,a) is the reward function, which is used to evaluate the operational perfection of the current material grasping and correction mechanism; γ is the operator parameter used for correction; q(s,a) is the actual grasping action probability distribution.

[0083] The above description is only a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent change made based on the technical essence of the present invention still falls within the scope of protection required by the present invention.

[0084] The present invention proposes an automatic deviation-correcting mechanism for material grabbing and an operation control method. The mechanism is provided with a frame, an input conveyor belt and an output conveyor belt are installed inside the frame, movable guide rails are arranged on both sides of the upper part of the frame, a terminal moving platform is arranged between the movable guide rails, symmetrical structures are arranged on both sides of the terminal moving platform and are installed and connected to the movable guide rails, and the structure is that a connecting rotating rod, a parallelogram mechanism long rod and a T-shaped connecting piece are sequentially installed at both ends of the terminal moving platform, a connecting long rod and a connecting side rod are respectively installed at both ends of the T-shaped connecting piece, the other end of the connecting long rod is installed on a front-end moving driving pair, and a connecting short rod and a rear-end moving driving pair are sequentially installed at the other end of the connecting side rod, and a suction cup and a video detection module are installed on the lower side of the terminal moving platform; the operation algorithm of the mechanism adopts a deep Q network improved by a TensorFlow Lite hardware acceleration library to realize the neat discharge of scattered food or medicine on the input conveyor belt onto the output conveyor belt.

Claims

1. An operating method of an automatic deviation-correcting mechanism for material grabbing, characterized in that: The automatic deviation-correcting mechanism for material grabbing comprises a frame (1), a foot pad (2), an input conveyor belt (3), an output conveyor belt (4), a movable guide rail (5), a front-end movable driving pair (6), a rear-end movable driving pair (7), a connecting long rod (8), a connecting short rod (9), a connecting side rod (10), a T-shaped connecting piece (11), a parallelogram mechanism long rod (12), an end moving platform (13), a suction cup (14), a video detection module (15), a connecting rotating rod (16) and a linear motor (17). The automatic deviation-correcting mechanism for material grabbing is provided with a frame (1). ), foot pads (2) are arranged at the bottom of the four groups of supporting feet of the frame (1), an input conveyor belt (3) and an output conveyor belt (4) are installed inside the frame (1), movable guide rails (5) are arranged on both sides of the upper part of the frame (1), and an end moving platform (13) is arranged between the two groups of movable guide rails (5), and the end moving platform (13) is located on the upper part of the input conveyor belt (3) and the output conveyor belt (4), and completely symmetrical structures are arranged on both sides of the end moving platform (13) and connected to the movable guide rails (5); taking one side structure as an example, the upper part of both ends of the end moving platform (13) is respectively installed A connecting rod (16), on which a parallelogram mechanism long rod (12) is mounted, and a T-shaped connecting piece (11) is mounted on the other end of the parallelogram mechanism long rod (12), and a connecting long rod (8) and a connecting side rod (10) are mounted on both ends of the T-shaped connecting piece (11), and a connecting short rod (9) is mounted on the other end of the connecting side rod (10), the connecting long rod (8) is connected and mounted on a front end moving driving pair (6), and the connecting short rod (9) is connected and mounted on a rear end moving driving pair (7), and the front end moving driving pair (6) and the rear end moving driving pair (7) is installed on the same movable guide rail (5), and a linear motor (17) is installed on the outer side of the front movable drive pair (6) and the rear movable drive pair (7); a suction cup (14) and a video detection module (15) are installed on the lower side of the terminal movable platform (13); the automatic deviation correction mechanism for material grabbing is provided with a terminal movable platform (13), and a suction cup (14) and a video detection module (15) are installed on the lower side of the terminal movable platform (13), and the suction cup (14) can move or rotate with the terminal movable platform (13), and the video detection module (15) is provided with a high-definition camera; The operating method of the automatic deviation-correcting mechanism for material grabbing adopts a deep Q network improved by the TensorFlow Lite hardware acceleration library to neatly discharge scattered food or medicine on the input conveyor belt to the output conveyor belt; specifically, the following steps are included: S1. Deploy data collection and detection equipment; By deploying a data acquisition and detection device on the material grabbing mechanism and establishing a cloud network management system to monitor the status of the monitoring device and data transmission; S2, data preprocessing; By establishing a cloud storage system, the data streams of various monitoring devices in the material grabbing mechanism are received and stored, and then motion data, motor speed data, and grabbing data sequences are generated; the sequence data is deeply learned through the generative adversarial network GAN, and the batch normalization technology is used to optimize the data acquisition model of the automatic correction mechanism; S3, model design and training; After obtaining the pre-processed grabbing mechanism sequence data of each room, the deep Q network DQN algorithm is used to train the data model; the S3 model design and training includes the following sub-steps: S31, determining the state space and the action space; S32, define the neural network structure; S33, define loss function and optimizer; S34, start training; S35, testing and optimization; S36, model saving and loading; S4, Improved deep Q network model construction; Through the improved deep Q network model, we can determine whether all subsequent running states meet the perfect conditions. If they meet the conditions, they can be released. If not, they can be adjusted immediately. The TensorFlow Lite hardware acceleration library is used to improve the performance of the DQN algorithm, and the algorithm conversion tuning formula is used to improve the accuracy and efficiency of the calculation results. Finally, the improved deep Q network function is used to complete the capture feedback adaptive adjustment optimization work.

2. The operating method of the automatic deviation-correcting mechanism for material grabbing according to claim 1 is characterized in that: The GAN loss function formula in S2 in the operation method of the automatic deviation correction mechanism for material grabbing is expressed as: L(D,G)=E[D(x)]-E[G(z)]+max(E[D(G(z))]) Among them, G represents the generator and D represents the discriminator, which are used to generate and judge various data sets sent by the correction mechanism; z~N(0,σ 2 ) is a random noise vector, σ is the standard deviation of the Gaussian distribution, ε is a very small positive number used to prevent division by zero; E[] represents the expected value; max() represents the maximum value.

3. The operating method of the automatic deviation-correcting mechanism for material grabbing according to claim 1 is characterized in that: The algorithm conversion and tuning formula in S4 in the operation method of the automatic deviation correction mechanism for material grabbing is expressed as: Among them, p(a|s,a) is the probability distribution of executing grasping action a in the current state s, and w h is a weight matrix used to store information about the state and hidden layer; b is a bias vector used to control the nonlinear activation function during forward propagation; g(s') is a reward function used to evaluate the value of the current state and is adjusted according to the adaptive experimental results of the grasping feedback; The improved deep Q network function in step S4 is expressed as: Among them, r(s,a) is the reward function, which is used to evaluate the operational perfection of the current material grasping and correction mechanism; γ is the operator parameter used for correction; q(s,a) is the actual grasping action probability distribution.

Images