Parameter identification method of intelligent tactile perception compliant grasping system

Abstract

The application discloses a parameter identification method of an intelligent tactile perception compliant grasping system, wherein the perception system adopts array piezoresistive sensors, piezoelectric sensors and 3D Hall sensors and is arranged in layers in imitation of the distribution of finger neurons. In order to realize seamless integration and collaborative work of the sensors, an artificial intelligence-based decoupling algorithm is adopted: a self-adaptive three-dimensional force decoupling model is constructed by using a deep learning algorithm (Transformer structure), the model is trained by a large amount of experimental data, the model can automatically learn complex nonlinear relationships, and the decoupling precision is improved; multi-modal information fusion is combined with multiple modal information such as pressure, vibration and stress direction to provide more comprehensive input data for the decoupling algorithm, and through collaborative analysis of the multi-modal information, the accuracy and reliability of the decoupling are further improved.

Description

Technical Field

[0001] This invention relates to the field of robot tactile sensing technology, specifically to a parameter identification method for an intelligent tactile perception compliant gripping system. Background Technology

[0002] Human hands are capable of dexterously grasping and manipulating various objects, primarily relying on the coordinated action of tactile sensation, muscle feedback, and the nervous system. This multimodal sensory ability allows humans to adapt to different shapes, materials (such as soft and brittle objects), and dynamic environments. In contrast, traditional robotic hands, lacking high-precision tactile feedback, are limited in their performance in complex tasks (such as grasping easily deformable objects).

[0003] As a core component of human-computer interaction, tactile sensors can detect multi-dimensional information such as force, vibration, sliding, and texture. Although microelectronics technology has driven the development of sensors such as capacitive, piezoresistive, and piezoelectric sensors, existing robotic arms are still far inferior to human hands in terms of dexterity and sensory sensitivity.

[0004] Different types of sensors (piezoresistive sensors, piezoelectric sensors, 3D Hall sensors) differ significantly in physical characteristics, signal types, and response speeds. Traditional modeling methods struggle to accurately describe their nonlinear characteristics, leading to insufficient decoupling accuracy. Furthermore, real-time decoupling requires rapid processing of large amounts of data, placing high demands on the computational efficiency and response speed of the algorithms. Achieving seamless integration and collaborative operation of these sensors is a crucial issue. Summary of the Invention

[0005] This invention provides a parameter identification method for an intelligent tactile sensing compliant gripping system, which solves the problems of low multimodal information decoupling accuracy, poor real-time performance, excessive size, and insufficient environmental adaptability in existing robotic gripper tactile systems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a parameter identification method for an intelligent tactile sensing compliant gripping system, wherein the sensing system employs an array of piezoresistive sensors, piezoelectric sensors, and 3D Hall sensors, arranged in a layered manner mimicking the distribution of neurons in a finger. 1) An artificial intelligence-based decoupling algorithm is adopted, and an adaptive three-dimensional force decoupling model is constructed using deep learning algorithms. The model is trained with a large amount of experimental data so that it can automatically learn complex nonlinear relationships and improve decoupling accuracy. 2) Multimodal information fusion combines multiple modal information such as pressure, vibration, and force direction to provide more comprehensive input data for the decoupling algorithm. Through the collaborative analysis of multimodal information, the accuracy and reliability of decoupling are further improved.

[0007] Furthermore, to ensure the detection efficiency of the multi-mode sensor, the following design is adopted. Layered data processing: By adopting layered data processing technology, signal processing and perception information identification are processed serially through two devices, reducing the computational burden of a single device and improving the efficiency of data processing during the detection process; Lightweight Neural Network Design: By adopting a lightweight network design, the complexity of the network structure is reduced and the network computation time is reduced while ensuring the accuracy of perceptual information recognition. Distillation technique: By using distillation techniques, the number of parameters in a lightweight neural network can be reduced, thereby improving the efficiency of model operation.

[0008] Furthermore, in the above-mentioned hierarchical data processing, signal preprocessing is performed through median filtering and wavelet transform. The bottom layer uses a microcontroller to perform median filtering and wavelet transform, while the upper-layer industrial control computer uses the Transformer model to perform multimodal feature fusion.

[0009] Furthermore, the Transformer model compressed by knowledge distillation technology is decoupled in three-dimensional forces in real time. The distillation technology improves the real-time performance of the algorithm through model compression, with a response latency of <10ms.

[0010] Furthermore, the intelligent decoupling algorithm is based on an improved Transformer with a lightweight network parameter size of <1MB.

[0011] Furthermore, during the training phase, 500 sets of pressure-vibration-three-dimensional force data were collected and input into the Transformer model for training; during the inference phase, the microcontroller preprocessed the data and transmitted it to the industrial control computer via USB 3.0, with a decoupling error of <±0.1N.

[0012] In summary, the intelligent tactile sensing compliant gripping system of this invention combines multiple high-performance sensors to achieve comprehensive perception of the gripping process by the robotic gripper. The control method of this invention utilizes deep learning algorithms to construct an adaptive decoupling model, which can automatically learn complex nonlinear relationships, significantly improving decoupling accuracy and the system's intelligence level. Employing hierarchical data processing technology, the data is initially processed and simple feature extraction is performed on an embedded device. Then, deep learning combined with a GPU device is used in a computer to identify tactile sensing information, enabling the system to achieve high-precision gripping position recognition and three-dimensional force decoupling, while also possessing rapid response capabilities to meet the needs of industrial automation. Attached Figure Description

[0013] Figure 1 This is a diagram of the overall system architecture of the present invention; Figure 2 This is a schematic diagram of the layered layout of the sensor components of the present invention; Figure 3 This is an electrical connection diagram of the tactile sensor of the present invention; Figure 4This is a structural diagram of the Transformer model of the present invention; Figure 5 shows the data of the 3D Hall sensor before and after filtering in a static state; Figure 6 shows the information extracted by wavelet transform. Detailed Implementation

[0014] The following is in conjunction with the appendix Figures 1 to 6 The specific implementation method of the parameter identification method of the intelligent tactile perception and compliant gripping system of the present invention will be further described in detail.

[0015] This invention employs a variety of high-performance sensors to achieve comprehensive perception of the grasping process. Specifically, these include: Array-type piezoresistive sensor: The M0404S distributed piezoresistive sensor is used to detect the pressure distribution on the contact surface between the object and the gripper during the grasping process, thereby achieving accurate identification of the grasping position. By arranging multiple piezoresistive sensor units on the gripper contact surface, the system can perceive the shape and positional deviations of the object in real time, providing a basis for the adaptive adjustment of the mechanical grasping system.

[0016] Piezoelectric sensor: An LDT0-028K piezoelectric sensor is selected to detect vibration signals generated during the gripping process. Vibration information is crucial for determining whether the object is stably gripped and whether there is a risk of slippage. By analyzing the vibration frequency and amplitude, the system can adjust the gripping force in a timely manner to prevent the object from slipping.

[0017] 3D Hall Sensor: Utilizing a composite structure of magnet, MLX90393 Hall sensor, and silicone, this sensor accurately detects the three-dimensional force direction applied to the hand's gripper during grasping. Through MEMS technology, it integrates triaxial force sensing and temperature sensing, enabling real-time monitoring of forces in the x, y, and z directions while simultaneously transmitting temperature information to the sensing system, providing multi-dimensional and precise feedback data for force control algorithms.

[0018] This multi-functional sensor module integrates a piezoresistive sensor, a piezoelectric sensor, and a 3D Hall sensor, and connects to a computer via a unified interface to identify tactile sensor parameters. The module uses an Arduino nano microcontroller to perform basic signal processing.

[0019] The control method of the present invention includes the following: 1. Signal Processing and Feature Extraction The raw signals acquired by sensors typically contain noise and interference, requiring preprocessing to extract useful information. This system employs the following methods for signal processing: Median filtering: used to remove spike noise from a signal, smooth the signal curve, and retain the main characteristics of the signal, as shown in Figure 5.

[0020] Wavelet transform: Wavelet decomposition is performed on the preprocessed signal to extract its time-frequency features. Wavelet transform can effectively separate high-frequency vibration information and low-frequency trend information in a signal, providing a foundation for subsequent feature extraction and analysis, as shown in Figure 6.

[0021] Feature extraction: Key features such as pressure peaks, vibration frequencies, and force directions are extracted from the processed signal. These features will serve as input to artificial intelligence algorithms to achieve intelligent perception and information identification.

[0022] 2. Application of Artificial Intelligence Algorithms To achieve intelligent perception of the system, this invention employs artificial intelligence algorithms, leveraging their powerful nonlinear fitting capabilities to avoid the complexity and inaccuracies of traditional modeling methods. Through a Transformer structure that mixes multimodal information, the system can decouple three-dimensional forces, vibrations, and other information during the grasping process in real time, providing the necessary information for precise and dexterous operation of the end effector, as shown in Figure 4.

[0023] 3. Electrical Design of Tactile Sensors The electrical hardware of the tactile sensor consists of an array of piezoresistive sensors, a piezoelectric sensor, a 3D Hall sensor, signal conditioning circuitry, a microcontroller, and a computer. The electrical connections are shown in Figure 3. The computer is a reliable industrial PC with rich software support; the microcontroller chosen is the small, easy-to-program, and cost-effective Arduino Nano; the array of piezoresistive sensors uses the M0404S distributed piezoresistive sensor, which has the advantages of small size and high sensing sensitivity; the piezoelectric sensor uses the LDT0-028K piezoelectric sensor, which features high detection accuracy and rapid response; the MLX90393 Hall sensor can detect the direction of force in real time and perform three-dimensional force decoupling.

[0024] 4. Multi-mode sensor ensures detection accuracy Different types of sensors (piezoresistive sensors, piezoelectric sensors, 3D Hall sensors) differ significantly in physical characteristics, signal types, and response speeds. Traditional modeling methods struggle to accurately depict their nonlinear characteristics, leading to insufficient decoupling accuracy. Furthermore, real-time decoupling requires rapid processing of large amounts of data, placing high demands on the computational efficiency and response speed of the algorithm. To achieve seamless integration and collaborative operation of these sensors, the following solution is adopted: (1) Artificial intelligence-based decoupling algorithm: Using deep learning algorithm (Transformer structure), an adaptive three-dimensional force decoupling model is constructed. The model is trained with a large amount of experimental data so that it can automatically learn complex nonlinear relationships and improve decoupling accuracy.

[0025] (2) Multimodal information fusion: Combining multiple modal information such as pressure, vibration, and force direction provides more comprehensive input data for the decoupling algorithm. Through collaborative analysis of multimodal information, the accuracy and reliability of decoupling are further improved.

[0026] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A parameter identification method for an intelligent tactile sensing compliant gripping system, wherein the sensing system employs an array of piezoresistive sensors, piezoelectric sensors, and 3D Hall sensors, arranged in a layered manner mimicking the distribution of neurons in a finger, characterized in that: 1) An artificial intelligence-based decoupling algorithm is adopted, and an adaptive three-dimensional force decoupling model is constructed using deep learning algorithms. The model is trained with a large amount of experimental data so that it can automatically learn complex nonlinear relationships and improve decoupling accuracy. 2) Multimodal information fusion combines multiple modal information such as pressure, vibration, and force direction to provide more comprehensive input data for the decoupling algorithm. Through the collaborative analysis of multimodal information, the accuracy and reliability of decoupling are further improved.

2. The parameter identification method for the intelligent tactile sensing compliant gripping system according to claim 1, characterized in that: To ensure the detection efficiency of the multimode sensor, the following design is adopted. 1) Layered data processing: The layered data processing technology is adopted to process signal processing and perception information identification in sequence through two devices, which reduces the computational burden of a single device and improves the efficiency of data processing during the detection process; 2) Lightweight Neural Network Design: A lightweight network design is adopted to reduce the complexity of the network structure and reduce the network computation time while ensuring the accuracy of perceptual information recognition. 3) Distillation technique: Distillation technique reduces the number of parameters in a lightweight neural network and improves the efficiency of model operation.

3. The parameter identification method for the intelligent tactile sensing compliant gripping system according to claim 2, characterized in that: In the above-mentioned hierarchical data processing, signal preprocessing is performed through median filtering and wavelet transform. The bottom layer uses a microcontroller to perform median filtering and wavelet transform, while the upper-layer industrial control computer uses the Transformer model to perform multimodal feature fusion.

4. The parameter identification method for the intelligent tactile sensing compliant gripping system according to claim 2, characterized in that: The knowledge distillation technique compresses the Transformer model for real-time decoupling of three-dimensional forces. The distillation technique improves the real-time performance of the algorithm through model compression, with a response latency of <10ms.

5. The parameter identification method for the intelligent tactile sensing compliant gripping system according to claim 1, characterized in that: The intelligent decoupling algorithm is based on an improved Transformer with a lightweight network parameter size of <1MB.

6. The parameter identification method for the intelligent tactile sensing compliant gripping system according to claim 5, characterized in that: During the training phase, 500 sets of pressure-vibration-3D force data were collected and input into the Transformer model for training. During the inference phase, the microcontroller preprocessed the data and transmitted it to the industrial control computer via USB 3.0, with a decoupling error of <±0.1N.

Images