Unmanned equipment contact sense sensor device based on large model

Abstract

The invention discloses an unmanned equipment contact sense sensor device based on a large model, which is applied to the technical field of artificial intelligence and unmanned equipment, and comprises a sensor module used for collecting a physical signal when the unmanned equipment is in contact with an external object; the data transmission module is used for transmitting the acquired data to the large model processing module; the large model processing module is used for carrying out deep analysis and processing on the transmitted data based on a large model, generating a corresponding decision instruction and sending the decision instruction to the control module; wherein the large model is constructed on the basis of a GPT-NeoX model of a Transform architecture, and the input of the model is obtained by performing feature extraction on acquired data on the basis of a CNN and RNN combined form; and the control module is used for converting the decision instruction into a control signal and sending the control signal to an execution structure of the unmanned equipment for action control. According to the invention, the perception and intelligent decision-making capability of unmanned equipment in a complex environment is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and unmanned equipment technology, and in particular to an unmanned equipment contact sensor device based on a large model. Background Art

[0002] The intelligence level of unmanned systems depends largely on their ability to perceive and interact with their surroundings. Contact sensors, as key components in achieving this, play a crucial role. Contact sensors enable unmanned systems to detect contact with external objects, as well as the magnitude and direction of contact force, enabling them to make more accurate and intelligent decisions in complex environments.

[0003] In industrial manufacturing, collaborative robots rely on contact sensors to collaborate safely and efficiently with human operators. When a robot makes unexpected contact with an operator while performing a task, the contact sensor can quickly sense the situation and trigger the appropriate safety mechanism to avoid harm to personnel. It also prevents equipment from being damaged by collisions, ensuring the smooth progress of the production process. In the logistics and warehousing industries, automated guided vehicles (AGVs) rely on contact sensors to accurately grasp and move goods. By sensing contact force and position information with the goods, AGVs can stably grasp goods and accurately transport them in complex warehouse environments, improving logistics efficiency. In the medical field, surgical robots use contact sensors to gently manipulate tissue, avoiding damage to surrounding healthy tissue and improving the accuracy and safety of surgery.

[0004] With the rapid development of artificial intelligence (AI), big models, as a key breakthrough, are profoundly changing the landscape of development across various fields. These models possess powerful learning and generalization capabilities, enabling them to learn and analyze vast amounts of data, extracting features and patterns within the data. Combining big models with contact sensors in unmanned vehicles presents unprecedented opportunities for enhancing the intelligence of unmanned systems. Big models can deeply analyze and process the vast amounts of data collected by contact sensors, enabling more accurate identification of object attributes such as shape, material, and hardness, as well as determining the interaction between the unmanned system and its surroundings. This enables unmanned systems to better adapt to complex and changing environments and accomplish more complex and sophisticated tasks.

[0005] Unmanned robots performing search and rescue missions in unknown wilderness environments can flexibly navigate complex terrain and obstacles through the collaborative work of contact sensors and a large model. Based on the contact information collected by the sensors, the large model can adjust the robot's action strategy in real time, guiding it to circumvent obstacles, find the optimal path, and quickly locate trapped individuals, improving rescue efficiency. In smart home scenarios, intelligent robot assistants leverage contact sensors and a large model to understand user intentions and needs. When users interact with the robot, the contact sensors detect touch, gestures, and other information, and the large model analyzes and interprets this information to provide corresponding services, such as helping users find items and control home appliances, providing users with a more convenient and intelligent living experience.

[0006] Research on large-scale contact sensor devices for unmanned equipment not only advances the development of unmanned equipment technology, improving its adaptability and mission execution capabilities in complex environments, but also offers broad application prospects and significant socioeconomic value. In the civilian sector, it can promote the development of smart logistics, smart homes, healthcare, and other industries, improving people's quality of life and productivity. Therefore, research in this area has both important practical and far-reaching strategic significance.

[0007] However, despite progress in the application of contact sensors and large models in unmanned systems, some shortcomings remain. Existing contact sensors still need to be further improved in terms of sensitivity, resolution, and stability to meet the needs of more complex, high-precision tasks in unmanned systems. Furthermore, the integration of large models with contact sensors is still in its early stages, requiring further research and improvement in data fusion methods, model training optimization, real-time performance, and reliability.

[0008] To this end, how to provide a large-model-based unmanned equipment contact sensor device that can effectively combine the knowledge and reasoning capabilities of the large model with the perception data of the contact sensor to achieve more efficient and intelligent decision-making is an urgent problem that technical personnel in this field need to solve. Summary of the Invention

[0009] In view of this, the present invention proposes an unmanned equipment contact sensor device based on a large model.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions: A large-model-based unmanned equipment contact sensor device comprises: a sensor module, a data transmission module, a large-model processing module and a control module; Sensor module: used to collect physical signals when unmanned equipment comes into contact with external objects; Data transmission module: used to transmit the data collected by the sensor module to the large model processing module; Large model processing module: This module is used to perform in-depth analysis and processing of the transmitted data based on the large model, and generate corresponding decision instructions to send to the control module. The large model is built based on the GPT-NeoX model of the Transformer architecture, and the input of the GPT-NeoX model is obtained by extracting features from the collected data based on a combination of CNN and RNN. Control module: used to convert decision instructions into control signals and send them to the execution structure of the unmanned equipment to control the movement of the unmanned equipment.

[0011] Optionally, the sensor module uses a contact sensor based on a quantum tunneling composite material, and optimizes the contact sensor from both structural and material aspects, specifically: Micro-nanostructure manufacturing technology is used in structural design to construct a sensor surface structure with a micro-pyramid array. Finite element simulation analysis is used to compare the effects of different micro-pyramid heights, spacings, and cone angles on sensor sensitivity, and to determine the optimal parameters of the micro-nanostructure. Nano-silver particles and carbon nanotubes are introduced into the material, and the content and dispersion state of the nano-silver particles and carbon nanotubes are regulated, and the optimal parameters are obtained through long-term stability testing.

[0012] Optionally, the sensor module further includes: a signal conditioning circuit and a data acquisition circuit; The signal conditioning circuit is used to amplify, filter and linearize the weak electrical signal output by the sensor. Specifically: The sensor output signal is first amplified by a low-noise amplifier. The amplified signal then passes through a bandpass filter to remove high-frequency noise and low-frequency interference signals. A linearization circuit is added to the signal conditioning circuit to fit and compensate the sensor's characteristic curve, ensuring a good linear relationship between the output signal and the contact force. The data acquisition circuit is used to convert the conditioned analog signal into a digital signal, specifically: An analog-to-digital converter is used to sample the analog signal, and a sample-and-hold circuit is added to the input of the analog-to-digital converter to keep the signal stable at the sampling moment.

[0013] Optionally, the data transmission module uses a high-speed Ethernet interface as the main communication method. The Ethernet controller is connected to the external network through an RJ45 interface, and an isolation transformer and lightning protection circuit are added to the Ethernet interface. At the same time, the data transmission module also supports multiple communication protocols such as TCP / IP and UDP.

[0014] Optionally, the GPT-NeoX model can be optimized, including designing a dedicated data encoding module at the input layer of the GPT-NeoX model to convert the multi-dimensional data collected by the contact sensor into a vector representation suitable for model input, specifically: A method based on the combination of position encoding and feature encoding is adopted to map the data of each dimension into a specific vector space, and position encoding is used to distinguish data at different time points and spatial locations.

[0015] Optionally, optimize the GPT-NeoX model, including converting the weights and activation values ​​in the model from high-precision data types to low-precision data types through model quantization and removing unimportant connections or neurons in the model through model pruning.

[0016] Optionally, before training the GPT-NeoX model, the collected data is preprocessed, including data denoising based on median filtering and Kalman filtering, and data normalization based on minimum-maximum normalization and Z-score normalization.

[0017] Optionally, feature extraction is performed on the collected data based on a combination of CNN and RNN, specifically: The spatial features extracted by CNN are used as the input of RNN. RNN processes the time series data through a cyclic structure and outputs a feature vector containing spatiotemporal features. An attention mechanism is introduced between CNN and RNN to make the model focus on important features.

[0018] Optionally, the training process of the GPT-NeoX model also includes: introducing integrated learning during the model inference process, fusing the inference results of multiple differently initialized models, and obtaining the final result through weighted averaging.

[0019] Optionally, the control module is also used to monitor and manage the operating status of the entire device, including working status monitoring of the sensor module, data transmission module and large model processing module, as well as fault diagnosis and alarm functions.

[0020] As can be seen from the above technical solution, compared to existing technologies, the present invention proposes a large-scale contact sensor device for unmanned equipment. By utilizing a sensor module based on quantum tunneling composite materials with optimized structure and materials, combined with a design including signal conditioning and data acquisition circuitry, this device achieves high-precision detection of sub-millinewton contact forces and stable signal output. A high-speed Ethernet interface and a data transmission module with lightning protection and anti-interference design ensure real-time and reliable data transmission. The GPT-NeoX model, based on the Transformer architecture, undergoes optimization through input layer encoding, quantization pruning, and combines CNN and RNN fusion feature extraction with an attention mechanism to significantly improve the recognition accuracy of object material, shape, and contact state (reaching over 90%), while also enhancing inference reliability through ensemble learning. The control module not only accurately converts decision commands into control signals but also monitors the status of each module in real time and provides fault diagnosis and alarms. Overall, this device offers significant advantages over traditional sensors in terms of accuracy (error ≤ 2%), stability (drift ≤ 3%), response speed (average 6ms), and reliability (significantly reduced failure rates), effectively enhancing the perception and intelligent decision-making capabilities of unmanned equipment in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0022] Figure 1 Schematic diagram of the device structure of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] Example 1

[0025] Embodiment 1 of the present invention discloses a large-scale model-based unmanned equipment contact sensor device, such as Figure 1 As shown, it includes: a sensor module, a data transmission module, a large model processing module and a control module; Sensor module: used to collect physical signals when unmanned equipment comes into contact with external objects.

[0026] In terms of sensor selection, it is crucial to fully consider the specific application scenarios and performance requirements of unmanned equipment. For the precision assembly tasks of industrial robots, high-precision, high-sensitivity contact sensors are required to ensure that the contact state and force magnitude can be accurately sensed even under the action of small forces. Therefore, after comparing and analyzing the performance of various sensors, the sensor module of the present invention adopts a contact sensor based on quantum tunnel composite material. Quantum tunnel composite material is a new type of intelligent material with a unique characteristic of resistance changing with pressure. When subjected to small pressure, the conductive particles inside it will be rearranged, causing the resistance to drop sharply, thereby achieving high-sensitivity detection of pressure. The sensitivity of this sensor is several times higher than that of traditional conductive rubber sensors, and it can detect contact force changes at the sub-millinewton level, meeting the strict requirements of industrial robot precision assembly for contact force perception accuracy.

[0027] In order to further optimize the performance of the sensor, the touch sensor was optimized from two aspects: structure and material. Specifically: Micro-nanostructure manufacturing technology is used in the structural design to construct a sensor surface structure with a micro-pyramid array. This micro-nanostructure can increase the contact area between the sensor and the object being measured, improve the pressure transmission efficiency, and thus enhance the sensitivity of the sensor. Finite element simulation analysis is used to compare the effects of different micro-pyramid heights, spacings, and cone angles on the sensor sensitivity to determine the optimal parameters of the micro-nanostructure.

[0028] Nanosilver particles and carbon nanotubes are introduced into the material. Nanosilver particles have good conductivity and stability, which can improve the conductivity of the composite material; carbon nanotubes have excellent mechanical properties and conductivity, which can enhance the mechanical strength and electron transmission capacity of the composite material. The content and dispersion state of nanosilver particles and carbon nanotubes are regulated, and the optimal parameters are obtained through long-term stability testing. The optimized quantum tunneling composite material has significantly improved stability and durability while maintaining high sensitivity.

[0029] The sensor module also includes: a signal conditioning circuit and a data acquisition circuit.

[0030] The signal conditioning circuit is used to amplify, filter and linearize the weak electrical signal output by the sensor. Specifically: The sensor's output signal is first amplified by a low-noise amplifier (LNA) to increase the signal's amplitude and reduce the impact of noise on the signal. The selected low-noise amplifier has an extremely low noise figure and a high gain-bandwidth product, enabling it to maintain signal purity while amplifying the signal. The amplified signal then passes through a bandpass filter to remove high-frequency noise and low-frequency interference signals, ensuring that the signal's frequency range meets the requirements. A second-order Butterworth bandpass filter is used, with cutoff frequencies set to 1Hz and 100Hz, respectively, to effectively remove electromagnetic interference and low-frequency drift signals in the environment. To compensate for the sensor's nonlinear characteristics, a linearization circuit is added to the signal conditioning circuit to fit and compensate the sensor's characteristic curve, resulting in a good linear relationship between the output signal and the contact force.

[0031] The data acquisition circuit is used to convert the conditioned analog signal into a digital signal, specifically: A high-precision 16-bit analog-to-digital converter (ADC) samples the analog signal, with a sampling rate of up to 100kHz, meeting the requirements for acquiring rapidly changing tactile signals. A high-precision reference voltage source is used as the ADC reference voltage to ensure conversion accuracy. To improve the stability and reliability of data acquisition, a sample-and-hold circuit is added to the ADC input to maintain signal stability at the sampling instant, preventing conversion errors caused by signal fluctuations.

[0032] Data transmission module: used to transmit the data collected by the sensor module to the large model processing module.

[0033] Considering the real-time and reliability requirements of data transmission, the data transmission module uses a high-speed Ethernet interface as its primary communication method. The Ethernet controller utilizes a high-performance, low-power chip and connects to the external network via an RJ45 interface. To ensure stable data transmission, an isolation transformer and lightning protection circuit are incorporated into the Ethernet interface to effectively prevent damage to the device from external electromagnetic interference and lightning strikes. The data transmission module also supports multiple communication protocols, including TCP / IP and UDP, which can be selected and configured according to actual application needs to achieve seamless communication with various external devices. To ensure data transmission reliability, data checking and error correction technologies, such as CRC and Hamming code error correction, are implemented during the data transmission process to ensure the integrity and accuracy of the transmitted data. When the sensor module collects large amounts of data, the data transmission module dynamically adjusts the transmission rate and mode based on the data volume and real-time requirements to meet the real-time data needs of the large model processing module.

[0034] Large model processing module: used to perform in-depth analysis and processing of the transmitted data based on the large model, and generate corresponding decision instructions and send them to the control module.

[0035] In selecting a large-scale model, after in-depth analysis and comparison of various mainstream large-scale models, and considering the characteristics and requirements of unmanned equipment contact sensor data processing, we ultimately selected the GPT-NeoX model based on the Transformer architecture. The GPT-NeoX model has strong language understanding and generation capabilities, and performs well in natural language processing tasks. Its open-source nature also allows us to customize and optimize it for specific application scenarios. The GPT-NeoX model utilizes the multi-head self-attention mechanism of the Transformer architecture, which effectively processes long sequences of data and captures complex dependencies within the data. When processing contact sensor data, this data typically contains both time series information and spatial location information, such as contact force variations at different moments and the distribution of contact points on the sensor surface. The GPT-NeoX model's multi-head self-attention mechanism can simultaneously focus on data at different time points and spatial locations, thereby better understanding the characteristics and patterns in the contact data. The model's multi-layered structure enables deep feature extraction, mining higher-level semantic information from raw contact signals, providing strong support for subsequent object attribute recognition and contact state determination.

[0036] In order to make the GPT-NeoX model better adapt to the contact sensor data, a series of targeted optimizations and adjustments were made to the model as follows: A dedicated data encoding module is designed at the input layer of the GPT-NeoX model to convert the multi-dimensional data collected by the contact sensor (such as contact force, contact position, contact time, etc.) into a vector representation suitable for model input. Specifically: A method based on combining positional encoding and feature encoding is used to map the data of each dimension into a specific vector space. Positional encoding is used to distinguish data at different time points and spatial locations. In this way, the model can accurately capture the temporal and spatial characteristics of the data when processing the input data.

[0037] To improve the model's inference efficiency and real-time performance, model quantization and pruning techniques were employed. Model quantization converts the model's weights and activation values ​​from high-precision data types to low-precision data types, for example, converting 32-bit floating-point numbers to 8-bit integers. This reduces the model's storage space and computational complexity, improving inference speed without significantly degrading model performance. Pruning also removes unimportant connections or neurons in the model, simplifying the model structure and further reducing computational complexity. The combined use of model quantization and pruning significantly improves the model's inference efficiency while maintaining model accuracy, enabling it to meet the stringent real-time requirements of unmanned equipment.

[0038] Before training the GPT-NeoX model, the collected data is preprocessed, including data denoising using median filtering and Kalman filtering, and data normalization using min-max normalization and Z-score normalization. Median filtering effectively removes isolated noise points by replacing the value of each point in the data sequence with the median of its neighborhood. Kalman filtering, an optimal estimation method based on a state-space model, accurately estimates the true value of a signal in the presence of noise. After filtering, the data is normalized to a specific interval, such as [0, 1] or [-1, 1], to eliminate dimensional differences between data of different dimensions and make the data comparable.

[0039] Feature extraction involves extracting key features from preprocessed data that reflect object properties and contact states, providing input for subsequent model inference. This paper employs a combination of CNN (convolutional neural network based on deep learning) and RNN (recurrent neural network) to extract features from the collected data, which serve as input for the GPT-NeoX model. CNN effectively extracts spatial features from data and has excellent capabilities for extracting information such as contact position distribution from contact sensor data. By designing convolution kernels of varying sizes and layers, the data is convolved to extract spatial features at different levels. RNN excels at processing time series data, capturing temporal dependencies within the data and effectively modeling information such as the temporal variation of contact force. The spatial features extracted by CNN serve as input to the RNN, which processes the time series data through a recurrent structure and outputs a feature vector containing spatiotemporal features. To further enhance feature extraction, an attention mechanism is introduced between the CNN and RNN, enabling the model to focus on important features in the data and improve feature representativeness.

[0040] Model inference uses a trained large model to analyze and judge the extracted features to derive the object's properties and contact status. The feature vectors output by feature extraction are input into the optimized GPT-NeoX model, which then infers and predicts the features based on the learned knowledge and patterns. During inference, the model generates a series of intermediate representations based on the input feature vectors. These intermediate representations contain preliminary judgments about the object's properties and contact status. Further processing and analysis of these intermediate representations ultimately yields information about the object's material, shape, hardness, and other properties, as well as contact status information such as stable contact, sliding, and collision. To improve the accuracy and reliability of inference, ensemble learning methods are employed during model inference, fusing the inference results of multiple models with different initializations and deriving the final result through weighted averaging and other methods.

[0041] Result output converts the results of model inference into a form that is easy to understand and apply, and then outputs them to the unmanned vehicle's control system or other application modules. Object attributes and contact status information are output as text or numbers, for example, indicating an object's material as "metal," shape as "cylindrical," hardness as "high," and contact status as "stable contact." To facilitate interaction with the unmanned vehicle's control system, the results can also be output as specific control instructions, such as adjusting the speed and force of the unmanned vehicle's movements based on the contact status. When outputting the results, a credibility assessment is also performed, assessing their reliability by calculating metrics such as the confidence level of the model inference results. If the confidence level of the results is low, the system automatically triggers retesting or further analysis to ensure the accuracy and reliability of the output results.

[0042] Control Module: This module converts decision instructions into control signals and sends them to the unmanned vehicle's execution structure to control its movements. For example, if the large model processing module determines that the unmanned vehicle is unstable while grasping an object, the control module will immediately adjust the actuator's movements, such as adjusting the gripping force or changing the gripping position, to ensure a stable grasp.

[0043] The control module is also used to monitor and manage the operating status of the entire device, including working status monitoring of the sensor module, data transmission module and large model processing module, as well as fault diagnosis and alarm functions.

[0044] The collaborative working mechanism between the various functional modules is based on event-driven and task-scheduling principles. When the sensor module detects a contact event, it triggers a data collection event, sending the collected data to the large model processing module via the data transmission module. Upon receiving the data, the large model processing module immediately initiates the data processing task, analyzes and processes the data according to preset algorithms and models, and generates decision instructions. The decision instructions are transmitted to the actuators of the unmanned equipment through the control module, and the actuators perform the corresponding actions according to the instructions. Throughout the entire process, the control module monitors the operating status of each module in real time to ensure stable system operation. If a module fails, the control module promptly issues an alarm signal and takes appropriate troubleshooting measures, such as switching to a backup module or performing fault recovery operations.

[0045] In order to deeply evaluate the performance advantages of the unmanned equipment contact sensor device based on the large model, the present invention conducts a comprehensive experimental comparison of the results with the traditional contact sensors.

[0046] In terms of accuracy, traditional contact sensors have certain measurement errors when measuring contact force due to their inherent physical properties and limitations in signal processing methods. When measuring tiny contact forces, traditional sensors have low resolution, making it difficult to accurately perceive tiny force changes. When measuring contact forces less than 1N, the relative error of traditional sensors is typically around 5%-10%. However, contact sensor devices based on large models can more accurately measure contact forces by learning and training from large amounts of contact data and using advanced data processing algorithms. Under the same experimental conditions, the relative error of sensors based on large models can be controlled to within 2% for contact force measurements less than 1N, significantly improving accuracy. This is because the large model can learn the complex nonlinear relationship between contact force and sensor output signal, allowing for more precise calibration and compensation of measured values.

[0047] In terms of response time, the signal processing flow of traditional tactile sensors is relatively simple, typically using hardware circuits for signal amplification, filtering, and conversion. The response time is primarily determined by the delay of the hardware circuits. Generally, the response time of traditional sensors is between 5 and 10 ms. Although the tactile sensor device based on the large model introduces the data processing process of the large model, its response time is not significantly increased through the coordinated optimization of hardware and software, as well as efficient algorithm implementation. In experiments, the average response time of sensors based on the large model is around 6 ms, which is comparable to that of traditional sensors. This is due to the use of a high-performance hardware computing platform and an optimized algorithm architecture, which enables the large model to complete the analysis and processing of sensor data in a short period of time.

[0048] In a stability comparison, traditional tactile sensors are prone to measurement drift during long-term use due to environmental factors (such as changes in temperature and humidity) and the aging of their own components. After eight hours of continuous operation, the measurement value drift of traditional sensors may reach 5%-10%. The tactile sensor device based on the large model utilizes the adaptive learning capabilities of the large model to monitor and compensate for performance drift caused by environmental changes and component aging in real time. Under the same experimental conditions of eight hours of continuous operation, the measurement value drift of the sensor based on the large model can be controlled within 3%, with significantly better stability than traditional sensors. The large model can automatically adjust the data processing model based on changes in environmental parameters and historical sensor data to maintain measurement accuracy.

[0049] In terms of reliability, the hardware structure of traditional tactile sensors is relatively simple, and their failure modes are primarily due to damage to sensor components and faulty circuit connections. During experiments, traditional sensors experienced an average of 2-3 failures for every 100 hours of operation. While the hardware and software systems of tactile sensor devices based on large models are more complex, they improve system reliability through the use of redundant design, fault diagnosis, and self-repair technologies, as well as the large model's ability to identify and process abnormal data. Within the same operating time, the average number of failures for sensors based on large models can be reduced to less than 1. The large model can monitor and analyze sensor data in real time. Once abnormal data is detected, it can promptly determine whether it is due to a sensor failure and take appropriate measures to address it, such as switching to a backup sensor or performing fault repair.

[0050] Recognition accuracy is a significant advantage of contact sensor devices based on large models. Traditional contact sensors can typically only detect the presence or absence of contact and simple force levels, and their ability to identify an object's material, shape, and other properties is very limited. However, sensor devices based on large models, by learning and training on contact data from a large number of objects of different materials and shapes, can accurately identify the object's properties and contact state. In experiments identifying objects of various materials and shapes, traditional sensors were almost unable to accurately identify the object's material and shape, while sensor devices based on large models achieved recognition accuracy exceeding 90%. The large model can extract key features from complex contact data and match and infer them with learned knowledge, thereby accurately determining the object's properties and contact state.

[0051] Unmanned equipment contact sensor systems based on large models offer significant advantages over traditional contact sensors in terms of precision, stability, and recognition accuracy. While their response time is comparable to that of traditional sensors, the coordinated optimization of hardware and software prevents a significant increase in response time due to the introduction of large models. In terms of reliability, the sensor system based on large models has also achieved good performance through the application of various technical means.

[0052] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0053] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A large-scale model-based unmanned equipment contact sensor device, characterized in that: include: Sensor module, data transmission module, large model processing module and control module; The sensor module is used to collect physical signals when the unmanned equipment comes into contact with external objects; The data transmission module is used to transmit the data collected by the sensor module to the large model processing module; The large model processing module is used to perform in-depth analysis and processing on the transmitted data based on the large model, and generate corresponding decision instructions to send to the control module; wherein the large model is constructed based on the GPT-NeoX model of the Transformer architecture, and the input of the GPT-NeoX model is obtained by extracting features from the collected data in the form of a combination of CNN and RNN; The control module is used to convert the decision instruction into a control signal and send it to the execution structure of the unmanned equipment to control the action of the unmanned equipment.

2. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: The sensor module uses a contact sensor based on quantum tunneling composite materials and is optimized in terms of both structure and material. Specifically: Micro-nanostructure manufacturing technology is used in structural design to construct a sensor surface structure with a micro-pyramid array. Finite element simulation analysis is used to compare the effects of different micro-pyramid heights, spacings, and cone angles on sensor sensitivity, and to determine the optimal parameters of the micro-nanostructure. Nano-silver particles and carbon nanotubes are introduced into the material, and the content and dispersion state of the nano-silver particles and carbon nanotubes are regulated, and the optimal parameters are obtained through long-term stability testing.

3. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: The sensor module further includes: a signal conditioning circuit and a data acquisition circuit; The signal conditioning circuit is used to amplify, filter and linearize the weak electrical signal output by the sensor, specifically: The sensor output signal is first amplified by a low-noise amplifier. The amplified signal then passes through a bandpass filter to remove high-frequency noise and low-frequency interference signals. A linearization circuit is added to the signal conditioning circuit to fit and compensate the sensor's characteristic curve, ensuring a good linear relationship between the output signal and the contact force. The data acquisition circuit is used to convert the conditioned analog signal into a digital signal, specifically: An analog-to-digital converter is used to sample the analog signal, and a sample-and-hold circuit is added to the input of the analog-to-digital converter to keep the signal stable at the sampling moment.

4. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: The data transmission module uses a high-speed Ethernet interface as the main communication method. The Ethernet controller is connected to the external network through an RJ45 interface, and an isolation transformer and lightning protection circuit are added to the Ethernet interface. At the same time, the data transmission module also supports multiple communication protocols such as TCP / IP and UDP.

5. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: Optimizing the GPT-NeoX model includes designing a specialized data encoding module at the input layer of the GPT-NeoX model to convert the multi-dimensional data collected by the tactile sensor into a vector representation suitable for model input, specifically: A method based on the combination of position encoding and feature encoding is adopted to map the data of each dimension into a specific vector space, and position encoding is used to distinguish data at different time points and spatial locations.

6. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: Optimize the GPT-NeoX model, including converting the weights and activation values ​​in the model from high-precision data types to low-precision data types through model quantization and removing unimportant connections or neurons in the model through model pruning.

7. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: Before training the GPT-NeoX model, the collected data is preprocessed, including data denoising based on median filtering and Kalman filtering, and data normalization based on minimum-maximum normalization and Z-score normalization.

8. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: Feature extraction of collected data is performed based on the combination of CNN and RNN, specifically: The spatial features extracted by CNN are used as the input of RNN. RNN processes the time series data through a cyclic structure and outputs a feature vector containing spatiotemporal features. An attention mechanism is introduced between CNN and RNN to make the model focus on important features.

9. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: The training process of the GPT-NeoX model also includes: introducing integrated learning in the model inference process, fusing the inference results of multiple differently initialized models, and obtaining the final result through weighted averaging.

10. The large-scale model-based unmanned equipment contact sensor device according to claim 1, characterized in that: The control module is also used to monitor and manage the operating status of the entire device, including working status monitoring of the sensor module, data transmission module and large model processing module, as well as fault diagnosis and alarm functions.

Images