Vehicle seat system and method

Abstract

The application discloses a vehicle-mounted seat system and method, belonging to the technical field of vehicle-mounted seats. The system includes a multi-source data acquisition module, an artificial intelligence reasoning module deployed on a vehicle edge computing platform, a seat execution module, and a human-computer interaction module. The artificial intelligence reasoning module is built-in with an artificial intelligence dynamic skeleton model pre-trained by a full-age spinal health data set, fuses real-time physiological and body data collected by high-density sensors in the vehicle and personal spinal health data authorized by the user, generates a personalized seat adjustment scheme in real time, drives the seat to perform support and massage actions, and displays spinal health assessment information to the occupant. The application realizes personalized spinal health management covering all age groups, all health data are processed locally at the vehicle end, eliminating the privacy leakage risk of cloud transmission from the architecture, and the existing vehicle-mounted seat hardware architecture can be reused to realize functional upgrade with low hardware cost increment, facilitating mass production and mounting.

Description

Technical Field

[0001] This application relates to the field of vehicle seat technology, and more specifically, to a vehicle seat system and method for personalized spinal support and health management based on artificial intelligence. Background Technology

[0002] Health adjustment functions for car seats have gradually become standard features in mid-to-high-end models. Currently, the existing technologies closest to this application mainly fall into two categories. The first category is the mainstream solution represented by zero-gravity seats. These solutions use a few pressure sensors built into the seat to capture the pressure points on the spine by preset fixed ergonomic support angles, and adjust the massage mode based on fixed rules, thus only achieving a general relaxation function. The second category is the advanced health solution adopted by some high-end models. This solution uses cameras and biosensors to collect physiological data such as driver fatigue and heart rate, and provides simple proactive interventions such as vibration reminders or meditation guidance for fatigue and emotional stress. However, it does not provide personalized treatment and long-term management for spinal health.

[0003] The existing solutions described above reveal three core flaws in practical applications. First, the adjustment logic is based on generalized presets, failing to adapt to the personalized spinal health status of different users, resulting in insufficient function utilization. Second, they only collect real-time in-vehicle data, failing to integrate users' spinal health data across different scenarios, and lack coverage of postural characteristics across all age groups, making it difficult to adapt to special groups such as children and the elderly. Third, the reliance on cloud transmission for processing user health data poses a risk of privacy leaks, leading to low user acceptance.

[0004] Therefore, there is an urgent need for a vehicle seat spinal health management solution that can achieve personalized adaptation for all age groups, integrate multi-source health data, and ensure privacy and security. Summary of the Invention

[0005] One objective of this application is to provide an in-vehicle seat system and method for AI-powered personalized spinal support and health management in driving scenarios, at least to address the problems of existing in-vehicle seats being unable to adapt to the personalized spinal care needs of users of different ages and health conditions, insufficient cross-domain data collaboration, and the risk of leakage of user health privacy.

[0006] To achieve the above objectives, some embodiments of this application provide the following aspects:

[0007] In a first aspect, some embodiments of this application provide an AI-powered personalized spinal support and health management in-vehicle seat system for driving and riding scenarios, comprising: a multi-source data acquisition module for acquiring real-time physiological and postural sensor data of in-vehicle occupants and receiving user-authorized personal spinal health data; an AI inference module deployed on the vehicle's local edge computing platform, incorporating an AI dynamic skeletal model pre-trained on a full-age-group spinal health dataset, for fusing the real-time physiological and postural sensor data and the personal spinal health data to generate a personalized seat adjustment scheme adapted to the current spinal state of the occupant; a seat execution module for adjusting the seat's support structure, seat posture, and massage actions according to the personalized seat adjustment scheme; and a human-computer interaction module for displaying spinal health assessment information generated based on the AI ​​dynamic skeletal model to the occupant.

[0008] Optionally, the multi-source data acquisition module includes a flexible tactile sensor array integrated within the seat, wherein the sensing point density of the flexible tactile sensor array is not less than 200 points / m². This high-density sensor array can accurately capture changes in the occupant's spinal curvature and body pressure distribution, providing a precise data foundation for personalized adjustments.

[0009] Optionally, the AI ​​dynamic skeletal model is pre-trained using an labeled dataset containing no fewer than 100,000 sets of spinal health data across all age groups. Pre-training with a large-scale dataset covering all age groups enables the model to recognize spinal characteristics of different groups, including children, adults, and the elderly, thereby achieving accurate adaptation for users of all ages.

[0010] Optionally, the artificial intelligence inference module further includes a multimodal heterogeneous data fusion unit for fusing the real-time physiological and posture sensing data and the personal spinal health data. Through data fusion processing, real-time in-vehicle data and the user's historical health data are effectively integrated, improving data consistency and usability.

[0011] Optionally, the AI ​​inference module completes all inference processes on the vehicle side, and the real-time physiological and posture sensing data and personal spinal health data do not leave the vehicle side; model parameters are only uploaded anonymously with user permission. Deploying the entire AI inference process on the vehicle-side edge computing platform avoids the risks of cloud transmission of users' original health data from the underlying architecture, significantly improving user privacy and acceptance.

[0012] Optionally, the AI ​​inference module is configured to retrieve corresponding processing strategies based on the occupant's identity and the driving / riding scenario. Specifically, in a family scenario with underage occupants, the AI ​​dynamic skeletal model generates a seat adjustment plan in conjunction with a database of adolescent spinal growth. In a long-distance business trip scenario, a segmented massage and wellness plan is generated based on the occupant's historical spinal health data and real-time driving / riding duration. Employing differentiated processing strategies for different scenarios and user identities enables the system to provide more precise and practical health services, enhancing the system's scenario adaptability and user engagement.

[0013] Secondly, this application also provides a management method for an artificial intelligence-based in-vehicle seat system, applied to the system described in any of the first aspects above, comprising: acquiring real-time physiological and posture sensing data of in-vehicle occupants, and user-authorized personal spinal health data; using an artificial intelligence dynamic skeletal model deployed on an in-vehicle edge computing platform and trained with spinal health data from all age groups, fusing and reasoning the acquired real-time physiological and posture sensing data and the personal spinal health data to generate personalized seat control parameters including personalized massage modes, support curves, and seat motion planning parameters; driving the seat to perform corresponding posture adjustment and massage actions according to the personalized seat control parameters; and generating and displaying a spinal health visualization report to the occupant.

[0014] Optionally, the real-time physiological and posture sensing data is acquired using a flexible tactile sensor array with a density of no less than 256 points / m². Employing a high-density sensor array allows for more precise capture of dynamic changes in spinal curvature, ensuring the accuracy and richness of the inference input data.

[0015] Optionally, the method further includes: collecting post-use data, including user feedback information and the latest spinal curvature data, after the occupant has used the device, and inputting the post-use data into the AI ​​dynamic skeletal model for incremental training. By collecting post-use data for incremental training, the model can continuously adapt to changes in the user's spinal condition, thereby continuously improving the accuracy of personalized solutions.

[0016] Optionally, the AI ​​dynamic skeletal model is pre-trained using at least 100,000 sets of labeled spinal health data across all age groups, and its inference process is entirely completed on the vehicle side. Training based on large-scale labeled data ensures the model's basic capabilities, while the vehicle-side inference architecture ensures that user health data remains within the vehicle throughout the entire process, balancing the requirements of high performance and high security.

[0017] Compared with related technologies, the solution provided in this application, by constructing a vehicle-side localized artificial intelligence inference architecture and integrating real-time data from high-density in-vehicle sensors with user-authorized multi-source spinal health data, achieves personalized spinal health management covering all age groups and all health states for the first time. This solution completely solves the problem of "functional fragmentation" caused by traditional seats relying on fixed rules and failing to adapt to individual differences, elevating the health function of seats from general relaxation to personalized care. Simultaneously, all health data is processed on the vehicle side, fundamentally eliminating the risk of privacy leaks during cloud transmission. Furthermore, this solution can reuse existing in-vehicle seat hardware architecture, achieving rapid mass production with a lower BOM cost, demonstrating significant technological advancement and commercial viability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a structural block diagram of a vehicle seat system provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0022] In this disclosure, the terms "upper," "lower," "inner," "middle," "outer," "front," and "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for better description of the embodiments of this disclosure and their implementations, and are not intended to limit the indicated devices, elements, or components to having a specific orientation, or to require them to be constructed and operated in a specific orientation. Furthermore, some of the aforementioned terms may be used to indicate other meanings besides orientation or positional relationship; for example, the term "upper" may in some cases indicate a dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in the embodiments of this disclosure according to the specific circumstances.

[0023] Furthermore, the terms "set up," "connect," and "fix" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this disclosure according to the specific circumstances.

[0024] Unless otherwise stated, the term "multiple" means two or more.

[0025] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0026] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.

[0027] It should be noted that, unless otherwise specified, the embodiments and features described in the present disclosure can be combined with each other.

[0028] Example 1

[0029] Combination Figure 1 As shown in this embodiment, an AI-powered personalized spinal support and health management vehicle seat system for driving scenarios is provided, comprising: a multi-source data acquisition module, an AI inference module, a seat execution module, and a human-computer interaction module. In this embodiment, the above four modules work collaboratively to form a complete closed loop from data perception and intelligent decision-making to execution feedback and information presentation. The entire system uses a vehicle-side edge computing platform as its central hub, focusing on the personalized spinal health needs of occupants, and achieves an integrated solution for real-time monitoring, precise analysis, and proactive intervention, aiming to upgrade traditional vehicle seats from passive support devices to active spinal health management terminals.

[0030] A multi-source data acquisition module is used to acquire real-time physiological and postural sensor data of in-vehicle occupants and receive user-authorized personal spinal health data. Specifically, the multi-source data acquisition module may include an in-vehicle smart camera, a steering wheel bioelectrical impedance sensor, and a flexible tactile sensor array integrated into the seat. The in-vehicle smart camera is used to collect biometric data such as driver facial fatigue and eye movement; the steering wheel bioelectrical impedance sensor is used to collect user heart rate and skin conductance data; and the flexible tactile sensor array is used to collect real-time data on seat pressure distribution and changes in spinal curvature. Personal spinal health data can be obtained from the user's mobile terminal, cloud-based health platform, or medical institution interface through the in-vehicle communication module after user authorization. This data includes, but is not limited to, the user's historical medical records, spinal bone data monitored by wearable devices, and publicly available spinal health research data for different age groups.

[0031] The design of this multi-source data acquisition module follows the principle of heterogeneous data complementarity. A single type of sensor can only reflect one aspect of an occupant's state. For example, cameras primarily capture facial and eye features, bioelectrical impedance sensors focus on physiological electrical signals, while flexible tactile sensor arrays directly measure the physical interaction of the human-chair interface. By collaboratively acquiring information from these different physical principles, sampling frequencies, and data formats, the system can comprehensively depict the occupant's real-time state from multiple dimensions, including visual, physiological, and mechanical perspectives. This provides information redundancy and cross-validation possibilities for subsequent artificial intelligence inference, overcoming the inherent limitations of single sensors, such as susceptibility to environmental interference and limited information dimensions. For instance, when a vehicle travels over bumpy roads, body pressure distribution data may experience instantaneous fluctuations. By combining this with occupant posture images captured by cameras, the system can distinguish whether the fluctuation is a passive change caused by road surface excitation or an active change caused by the occupant actively adjusting their posture, thus avoiding misjudgment.

[0032] The AI ​​inference module, deployed on the vehicle's local edge computing platform, incorporates a dynamic skeletal model pre-trained with a full-age-range spinal health dataset. This model integrates real-time physiological and postural sensor data with the individual's spinal health data to generate a personalized seat adjustment scheme that adapts to the current occupant's spinal condition. The AI ​​inference module utilizes an automotive-grade edge computing chip, and all inference calculations are performed on the vehicle side. The user's original health data remains within the vehicle; model parameters are only anonymized and uploaded with the user's permission.

[0033] From a working principle perspective, the AI ​​inference module is the decision-making center of this system. Its core AI dynamic skeletal model underwent pre-training using a large-scale labeled dataset before deployment to the vehicle. During pre-training, the model learned the complex mapping relationship between input features (including body pressure distribution matrix, spinal key point coordinates, user age and other demographic information, historical health records, etc.) and output decisions (optimal support curve, massage points and intensity) through supervised learning. Internally, the model employs a deep neural network architecture, automatically extracting hierarchical features from the input data through multi-layer nonlinear transformations. Once deployed to the vehicle, the model enters the inference phase. Each time a passenger sits down, the multi-source data acquisition module transmits real-time data to the AI ​​inference module. The model performs forward propagation calculations, outputting personalized seat adjustment parameters within milliseconds. This inference process requires no cloud connection and is entirely completed using the vehicle's computing power, ensuring real-time service and data privacy and security.

[0034] The seat execution module is used to adjust the seat's support structure, seat posture, and massage actions according to the personalized seat adjustment scheme. The seat execution module may include actuators such as a seat back support airbag array, a seat cushion support adjustment mechanism, massage airbags or massage motors, etc. By adjusting the inflation volume, position, and action mode of each component, personalized support and massage for the occupant's spine can be achieved.

[0035] The seat's operating principle is based on a combination of pneumatic adjustment and electromechanical control. Multiple independently controlled airbag units are distributed within the backrest and seat cushion. Each airbag is connected to an air pump via a solenoid valve. Control parameters from the artificial intelligence inference module are interpreted as the switching sequence of each solenoid valve and the duty cycle of the air pump, thus achieving precise control over the inflation of each airbag. When the support height of a certain area needs adjustment, the corresponding airbag inflates; when a massage action is needed, adjacent airbags alternately inflate and deflate according to a predetermined rhythm, creating a wave-like or point-pressure massage effect. Compared to traditional integrated adjustment mechanisms, this distributed airbag array solution can achieve centimeter-level or even millimeter-level precision in local support adjustment, allowing the seat's support surface to accurately match the individual's spinal curvature.

[0036] The human-computer interaction module is used to display spinal health assessment information generated based on the aforementioned artificial intelligence dynamic skeletal model to occupants. The module can display current spinal status analysis, historical trends, and improvement suggestions to occupants in the form of visual charts via the vehicle's central control screen or dashboard display.

[0037] Optionally, in some embodiments, the multi-source data acquisition module includes a flexible tactile sensor array integrated within the seat, wherein the sensing point density of the flexible tactile sensor array is not less than 200 points / m². Specifically, the flexible tactile sensor array is distributed in the seat back and seat cushion areas, and each sensing point can independently detect pressure values. This high-density layout enables refined acquisition of body pressure distribution on the occupant's back and spinal curvature. When the occupant sits down, the sensor array scans the pressure values ​​at each point at a preset frequency, generating a real-time body pressure distribution matrix. This matrix data serves as one of the inputs to the artificial intelligence inference module.

[0038] The introduction of the aforementioned features brings significant technical benefits. Those skilled in the art will understand that the spatial resolution of body pressure distribution measurement directly determines the upper limit accuracy of spinal curvature inversion. Traditional vehicle seats typically place pressure sensors in only a few locations, resulting in sparse sampling points. This allows for only a rough assessment of the occupant's approximate stress area and cannot reconstruct the continuous curvature of the spine. This embodiment increases the sensing point density to at least 200 points / m², reducing the spacing between adjacent sensing points to within approximately 7 cm. This enables the formation of a dense sequence of sampling points along the spinal axis on the backrest. By interpolating and curve fitting these discrete pressure values, the system can reconstruct a sagittal plane curvature profile of the spine that closely approximates reality. Furthermore, the high-density array can capture local abnormal peaks in the body pressure distribution. These peaks are often associated with compensatory postures caused by problems such as scoliosis and herniated discs, providing crucial clues for the AI ​​inference module to identify potential health risks. Consequently, when generating personalized support curves, the AI ​​dynamic skeletal model can provide targeted support to specific vertebral segments, rather than simply adjusting the tilt angle of the entire backrest, achieving true personalization of the support solution.

[0039] In other embodiments, the AI ​​dynamic skeletal model is pre-trained using an labeled dataset containing no fewer than 100,000 sets of spinal health data across all age groups. This dataset covers spinal morphology data for different age groups, including children, adolescents, adults, and the elderly, and includes data on normal spinal curvature, scoliosis, and characteristic data of common spinal problems such as lumbar disc herniation. The data annotations include information such as the location of key spinal points, curvature type, and health assessment level. Through pre-training, the AI ​​dynamic skeletal model acquires the ability to recognize the spinal characteristics of occupants of different age groups and output corresponding adaptation solutions.

[0040] The reason this embodiment emphasizes the "all-age group" characteristic of the dataset is that there are fundamental differences in spinal morphology among different age groups. Children and adolescents' spines are in the growth and development stage, the vertebral epiphyseal plates have not yet closed, and the spinal curvature has not yet been formed. Poor posture can easily induce developmental problems such as scoliosis.

[0041] While the adult spine is relatively stable, prolonged sitting can lead to lumbar disc degeneration. The spines of older adults face degenerative changes such as osteoporosis, vertebral compression fractures, and disc height loss. If AI models are trained using only adult data, they may misinterpret developmental or degenerative features as abnormalities when dealing with children or elderly passengers, leading to inappropriate adjustment strategies. By introducing labeled data covering all age groups during the pre-training phase, the model can learn the inherent statistical distribution patterns of spinal characteristics in different age groups during training, forming age-related prior knowledge.

[0042] In actual reasoning, by incorporating occupant age information, the model can automatically switch to the parameter space corresponding to the age group for decision-making, avoiding a one-size-fits-all approach. Therefore, this solution achieves full lifecycle coverage from children to the elderly, overcoming the limitation of traditional car seats that are only suitable for standard adult users.

[0043] Furthermore, in some embodiments, the artificial intelligence inference module further includes a multimodal heterogeneous data fusion unit for fusing the real-time physiological and posture sensing data and the personal spinal health data. The multimodal heterogeneous data fusion unit receives multi-source heterogeneous data, including body pressure data from a flexible tactile sensor array, visual data from a camera, vital sign data from a bioelectrical impedance sensor, and user-authorized historical health data. Through time synchronization, spatial alignment, and feature extraction, it uniformly converts data of different formats and sampling frequencies into standardized feature vectors for inference by the artificial intelligence dynamic skeletal model. During the fusion process, this unit also incorporates vehicle motion state information to compensate for sensor data fluctuations caused by dynamic factors such as vehicle bumps, acceleration, and deceleration.

[0044] The technical effectiveness of this multimodal heterogeneous data fusion unit is reflected in the following two aspects.

[0045] First, it solves the alignment problem of multi-source data in time and space. For example, a camera captures images at a frame rate of 30Hz, while a tactile sensor scans pressure values ​​at a frequency of 100Hz, resulting in inconsistent sampling times. Simultaneously, there is a spatial transformation relationship between the camera coordinate system and the seat coordinate system. The fusion unit ensures, through hardware timestamp synchronization and coordinate system calibration, that all data involved in inference reflects the same state of the occupant at the same moment, eliminating system errors introduced by spatiotemporal misalignment.

[0046] Secondly, it effectively suppresses noise interference in dynamic driving environments. During vehicle acceleration, deceleration, and cornering, inertial forces cause displacement and pressure changes in the occupant's body relative to the seat. These changes are not caused by the occupant's active posture adjustments; if not filtered out, they will lead to misjudgments of the spinal state by the AI ​​inference module. The fusion unit, by introducing motion parameters such as acceleration and angular velocity provided by the vehicle's CAN bus, constructs a dynamic compensation model to separate the pressure change component caused by inertial forces from the total measurement values, extracting the quasi-static pressure component that is only related to the occupant's body state. After this processing, the signal-to-noise ratio of the input data to the AI ​​inference module is significantly improved, laying a solid foundation for subsequent accurate decision-making.

[0047] In some embodiments, the AI ​​inference module completes all inference processes on the vehicle side, and the real-time physiological and posture sensing data and personal spinal health data do not leave the vehicle side; model parameters are only uploaded anonymously with user permission. Specifically, after completing model inference on the vehicle-side edge computing platform, the AI ​​inference module generates a personalized seat adjustment plan and directly sends it to the seat execution module. The user's original health data is stored in a secure local storage area on the vehicle side and is not transmitted to an external server via the network. With user permission, the system only uploads the anonymized and desensitized model gradient parameters to the cloud for global model updates; the upload process does not include any original health information traceable to a specific individual.

[0048] From a technical architecture perspective, the vehicle-side local inference design in this embodiment resolves the long-standing conflict between health data applications and privacy protection. Health data is highly sensitive personal information, and cloud transmission and storage inevitably introduce security risks such as data leakage and unauthorized access, which is one of the main reasons for low user acceptance of in-vehicle health functions. By confining all inference computations within the physical boundary of the vehicle-side edge computing platform, the lifecycle of the user's original health data is completed entirely in a closed loop locally in the vehicle, fundamentally eliminating the risk of leakage in the transmission link and cloud storage.

[0049] Meanwhile, to accommodate the continuous evolution of AI models, this solution adopts a simplified implementation of the federated learning concept—uploading only the anonymized model gradient parameters, rather than the original data. Gradient parameters are intermediate mathematical quantities required for model parameter updates; after processing with techniques such as differential privacy, it is impossible to infer any individual information from them. This design, while protecting user privacy, still leverages collective intelligence to continuously optimize model performance, achieving an organic balance between privacy protection and functional evolution.

[0050] Furthermore, in some embodiments, the AI ​​inference module is configured to retrieve corresponding processing strategies based on occupant identity and driving scenario. The system identifies occupants through facial recognition via in-vehicle cameras or by logging in with a user account. When the scenario involves a minor family member, the AI ​​dynamic skeletal model, combined with a database of adolescent spinal growth, generates a seat adjustment plan. This plan focuses on adaptively adjusting support strength to suit the postural characteristics of adolescents during their skeletal development. When the scenario involves a long-distance business trip, the system generates a segmented massage and therapy plan based on the occupant's historical spinal health data and the current real-time driving duration. During the long journey, the system automatically switches between massage modes and support postures at preset time intervals. After the trip, a health analysis report is pushed to the user through the human-computer interaction module.

[0051] The introduction of scenario-based strategies upgrades the system from a one-size-fits-all massage program to an intelligent health service that is tailored to the specific time and individual needs. In the scenario of family members with minors, the adolescent spinal growth database contains reference data on vertebral body size, intervertebral disc thickness, and normal ranges of physiological curvature development for adolescents of various ages. An AI-powered dynamic skeletal model compares the currently collected postural data with age-appropriate reference standards in this database. If it detects a deviation in local pressure distribution from the normal range, it adjusts the airbag support force in the corresponding area to provide appropriate corrective force, preventing the impact of prolonged poor posture on the developing spine. In the scenario of long-distance business travel, the generation of segmented massage therapy programs is based on a quantitative modeling of the cumulative fatigue patterns of long-distance driving. The system continuously monitors changes in body pressure distribution along the passenger's spine, assesses the fatigue level of various muscle groups, and automatically initiates the massage program at appropriate times, rather than waiting until the passenger feels significant discomfort before manually activating it. This proactive and predictive intervention strategy can more effectively delay the accumulation of driving fatigue and reduce the risk of cumulative spinal damage during long-distance driving. The health analysis report after the trip provides a basis for long-term health management, helping users to establish continuous attention and scientific understanding of their own spinal health.

[0052] This disclosure also provides an artificial intelligence-based method for managing spinal health in vehicle seats, applied to the system described above, comprising the following steps:

[0053] Step S201: Acquire real-time physiological and postural sensor data of the occupants, as well as user-authorized personal spinal health data. After the system is started, the flexible tactile sensor array in the seat, the in-vehicle camera, the bioelectrical impedance sensor on the steering wheel, etc., begin to collect real-time data of the occupants. At the same time, the system requests user-authorized personal spinal health data through the wireless communication module.

[0054] During system startup, the method first initializes and establishes handshakes for the data acquisition channels. Each sensor module performs self-tests and calibrations to ensure consistency of measurement benchmarks. The flexible tactile sensor array performs zero-point calibration, returning the pressure values ​​to zero under no-load conditions to eliminate the influence of seat structural stress on subsequent measurements. The camera performs automatic exposure and white balance adjustments to adapt to the in-vehicle lighting environment. The wireless communication module, based on the user-preset authorized range, initiates data synchronization requests to the bound mobile terminal or cloud-based health platform to obtain the user's latest uploaded spinal health records. This initialization process ensures that all subsequent inference calculations are based on accurate and complete data.

[0055] Step S202 involves using an AI dynamic skeletal model, trained on spinal health data across all age groups and deployed on an on-vehicle edge computing platform, to fuse and infer the acquired real-time physiological and posture sensing data and the individual spinal health data. This generates personalized seat control parameters, including personalized massage modes, support curves, and seat motion planning parameters. The AI ​​dynamic skeletal model first preprocesses the input data through a multimodal data fusion unit to generate standardized feature vectors. Then, based on the pre-trained model, it performs forward inference to output the currently optimal seat support curve parameters, massage point selection, massage intensity and frequency, and posture adjustment parameters for the seat back and seat cushion.

[0056] Step S202 is the core computational step of this method. The sequence of operations performed in the preprocessing stage of the multimodal data fusion unit includes: aligning the timestamps of each sensor's data to synchronize asynchronous sampling data to a unified time base; registering the spatial data to a coordinate system, transforming the coordinates of key occupant posture points in the camera coordinate system to the seat body coordinate system; filtering and denoising the pressure distribution data to separate static body pressure components and dynamic interference components; and feature encoding the historical health data, converting unstructured medical records into standardized feature vector representations. After preprocessing, the generated feature vectors are fed into the artificial intelligence dynamic skeletal model. During the model's forward inference, the input features flow sequentially through multiple hidden layers, with each layer's neurons performing weighted summation and nonlinear activation on the output of the previous layer, ultimately generating a multidimensional control parameter vector at the output layer.

[0057] The vector is parsed into a command format recognizable by the seat execution module, including but not limited to: the target pressure value sequence of each support airbag, the action timing diagram of the massage airbag, the target angle of the backrest tilt servo motor, and the target position of the seat cushion fore-and-aft sliding. The entire inference process runs in hard real-time mode on the vehicle-side edge computing platform, with the end-to-end latency from data input to parameter output controlled within 100 milliseconds, ensuring that the seat can respond almost instantly to changes in occupant posture.

[0058] Step S203: Drive the seat to perform corresponding posture adjustment and massage actions according to the personalized seat control parameters. After receiving the control parameters, the seat execution module parses the parameters into drive signals for each execution element, controls the inflation amount of the support airbag to achieve spinal curvature adaptation, and controls the start, stop, intensity, and rhythm of the massage elements to execute the personalized massage program.

[0059] In the execution drive phase of step S203, the parsing of control parameters follows the principle of hierarchical mapping. The top-level control parameters define the macroscopic functional goals of each region, such as "enhanced lumbar support" and "shoulder massage activation"; the middle layer translates them into the target states of each actuator, such as "inflate L3-L5 region airbags to 80%" and "shoulder massage airbags alternately operate at a frequency of 2Hz"; the bottom-level drive layer converts the target states into specific electrical signals such as the opening and closing sequence of the solenoid valve, the PWM duty cycle of the air pump motor, and the position loop control command of the servo motor.

[0060] The layered architecture design ensures that the decisions of the upper-layer AI inference module are not affected by changes in the underlying actuator hardware. When the hardware configuration of the seat execution module changes, only the intermediate layer mapping table needs to be updated, without modifying the AI ​​inference algorithm, which significantly improves the system's hardware compatibility and platform migration capabilities.

[0061] Step S204: Generate and display a spinal health visualization report to the occupants. During the reasoning process, the artificial intelligence inference module simultaneously generates a spinal health visualization report that includes the current spinal curvature status, body pressure distribution, comparative analysis with previous measurement results, and long-term trends. This report is then displayed on the in-vehicle display screen through the human-computer interaction module.

[0062] Step S204 embodies the innovative concept of this invention, which combines immediate adjustment with long-term health management.

[0063] Unlike traditional chairs that only focus on immediate seating comfort, this method simultaneously accumulates spinal health data assets during each use. The generation of the visualization report is handled by the auxiliary output branch of the artificial intelligence inference module. This branch operates in parallel with the main control parameter output, and it structures and organizes the high-level semantic features extracted from the intermediate layers of the model during the inference process (such as the estimated Cobb angle of spinal curvature, the left-right pressure asymmetry index, and the degree of pelvic tilt), and compares them longitudinally with past records in the historical database.

[0064] The report is presented in intuitive charts on the in-vehicle central control screen. For example, it uses color temperature graphs to map body pressure distribution, curve graphs to compare the sagittal curvature of the spine at different times, and radar charts to display the comprehensive scores of multidimensional health indicators. This visualized health feedback mechanism transforms the originally abstract and intangible state of spinal health into information that users can perceive and understand. This helps to increase users' awareness of their own spinal health and encourages them to actively adopt healthier sitting habits while driving, forming a positive cycle of "monitoring-feedback-improvement".

[0065] Optionally, in some embodiments, real-time physiological and posture sensing data are acquired through a flexible tactile sensor array with a density of not less than 256 points / m². This sensor array operates in a row-column scanning manner, with a scan cycle of not less than 50 milliseconds per frame, enabling real-time capture of dynamic changes in spinal curvature caused by changes in occupant posture. The acquired data includes the pressure value and coordinate position of each sensing point.

[0066] The technological gains from employing a high-resolution sensor array with a density of no less than 256 points / m² are significant. With a scan cycle of no less than 50 milliseconds, the sensor can continuously output body pressure distribution images at a frame rate exceeding 20Hz. This temporal resolution is sufficient to capture dynamic changes in spinal curvature caused by occupants' breathing and slight adjustments in posture. High spatial resolution ensures that the body pressure distribution images contain sufficient spatial detail, enabling the spinal midline extraction algorithm to locate the projected position of the spinous processes with sub-centimeter precision.

[0067] Comparative experiments show that when the density of sensing points is increased from about 50 points / m² in the traditional scheme to 256 points / m², the root mean square error of spinal curvature inversion is reduced by about 55%, and the accuracy of lumbar load estimation is increased from about 70% to about 96%, providing fundamental data quality assurance for the effectiveness of subsequent personalized adjustment schemes.

[0068] In some embodiments, the method further includes step S205: after occupant use, collecting post-use data including user feedback information and the latest spinal curvature data, and inputting the post-use data into the artificial intelligence dynamic skeletal model for incremental training. Users can rate the massage effect or select preference tags through the human-computer interaction module, and the system associates and stores user feedback with the latest spinal curvature data collected. After accumulating a certain amount of post-use data, the artificial intelligence inference module uses this data for incremental model training during vehicle idle periods (such as when the vehicle is parked and charging at night), updating the model parameters.

[0069] The incremental training mechanism in step S205 is a crucial step in achieving long-term personalized adaptation of the model. Its working principle is based on the theoretical framework of transfer learning and incremental learning. The pre-trained model already possesses general spinal health knowledge, but its parameters are statistically averaged solutions obtained by optimization on group data, which may not be optimal for a specific individual. In the incremental training phase, the model is initialized with the pre-trained weights and fine-tuned using the user's historical usage data. During fine-tuning, only some layer parameters of the model are updated, while the weights of the lower-level feature extraction layers remain unchanged. This allows the higher-level decision-making layers to adapt to the user's individual characteristics while preserving general knowledge.

[0070] To prevent catastrophic forgetting—the loss of general knowledge acquired during pre-training when learning new user features—fine-tuning employs a small learning rate and early stopping strategy, ensuring that model parameters are adjusted only to a limited extent around the pre-training weights. Incremental training computations are scheduled during vehicle idle periods, fully utilizing the idle window of vehicle-side computing power and not affecting real-time inference performance during normal driving. With increased usage, the model's personalized adaptation accuracy for the user continuously improves. Real-world testing data shows that after approximately 20 uses of incremental training, the user's subjective satisfaction score increases by an average of at least 30%, and the deviation between the model's predictions and the user's manually adjusted preferences gradually converges.

[0071] Furthermore, in some embodiments, the AI ​​dynamic skeletal model is pre-trained using at least 100,000 sets of labeled spinal health data across all age groups, and its inference process is entirely completed on the vehicle side. The pre-training process is completed in the cloud or on an offline server, and the pre-trained model parameters are then deployed to the vehicle-side edge computing platform. In actual use, all inference calculations are completed by the vehicle-side edge computing platform, without relying on a network connection; the system can still provide all functions normally even in a network-free environment.

[0072] The aforementioned "cloud-based pre-training + vehicle-side inference" separate deployment architecture has multiple advantages in engineering implementation. The pre-training phase has extremely high demands on computing resources, typically requiring high-performance GPU clusters for training for several days or even weeks. Completing this process in the cloud is the most economical and efficient approach.

[0073] After pre-training, the resulting model parameter files are compressed, quantized, and converted into a format supported by the vehicle-mounted inference chip, and then deployed to vehicles via OTA (Over-The-Air). The vehicle-mounted inference chip focuses on the forward propagation computation of the model, with computational complexity far lower than that of training; therefore, automotive-grade chips are sufficient. This architecture allows the system to continue providing services even when the network is disconnected, without relying on mobile network coverage, making it suitable for scenarios with no or unstable network access, such as underground parking garages and remote areas.

[0074] Meanwhile, independent inference on the vehicle side eliminates the impact of network transmission latency on system response time, which is especially important for scenarios requiring real-time response, such as seat posture adjustment and massage control. Furthermore, this architecture inherently supports user privacy protection because the user's raw data has no way to leave the vehicle after it leaves the factory, conforming to the design paradigm of "privacy computing."

[0075] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An AI-powered personalized spinal support and health management in-vehicle seat system for driving and riding scenarios, characterized in that, include: The multi-source data acquisition module is used to acquire real-time physiological and postural sensor data of occupants in the vehicle, and to receive personal spinal health data authorized by the user. The artificial intelligence inference module is deployed on the vehicle's local edge computing platform. It has a built-in artificial intelligence dynamic skeletal model pre-trained with a full-age spinal health dataset. It is used to integrate the real-time physiological and posture sensing data and the personal spinal health data to generate a personalized seat adjustment scheme that adapts to the current spinal status of the occupant in real time. The seat execution module is used to adjust the seat's support structure, seat posture, and massage actions according to the personalized seat adjustment scheme. The human-computer interaction module is used to display spinal health assessment information generated based on the artificial intelligence dynamic skeletal model to the occupants.

2. The system according to claim 1, characterized in that, The multi-source data acquisition module includes a flexible tactile sensor array integrated into the seat, wherein the sensing point density of the flexible tactile sensor array is not less than 200 points / m².

3. The system according to claim 1, characterized in that, The AI-powered dynamic skeletal model is pre-trained using an labeled dataset containing no fewer than 100,000 sets of spinal health data across all age groups.

4. The system according to claim 1, characterized in that, The artificial intelligence inference module also includes a multimodal heterogeneous data fusion unit, which is used to fuse the real-time physiological and posture sensing data and the personal spinal health data.

5. The system according to claim 1, characterized in that, The AI ​​inference module completes all inference processes on the vehicle side, and the real-time physiological and posture sensing data and personal spinal health data do not leave the vehicle side, but are only uploaded as model parameters with the user's permission.

6. The system according to claim 1, characterized in that, The artificial intelligence reasoning module is configured to retrieve corresponding processing strategies based on the passenger's identity and the driving scenario; in the scenario of a family with underage passengers, the artificial intelligence dynamic skeletal model generates a seat adjustment scheme in combination with a database of adolescent spinal growth; in the scenario of a long-distance business trip, a segmented massage and recuperation scheme is generated based on the passenger's historical spinal health data and real-time driving duration.

7. An artificial intelligence-based method for managing spinal health in vehicle seats, applied to the system described in any one of claims 1 to 6, characterized in that, include: Acquire real-time physiological and postural sensor data of occupants in the vehicle, as well as personal spinal health data authorized by the user; Using an AI dynamic skeletal model trained on spinal health data across all age groups and deployed on an on-vehicle edge computing platform, the acquired real-time physiological and posture sensing data and the individual spinal health data are fused and inferred to generate personalized seat control parameters, including personalized massage modes, support curves, and seat motion planning parameters. The seat is driven to perform corresponding posture adjustment and massage actions according to the personalized seat control parameters. Generate and display a visual report on spinal health to occupants.

8. The method according to claim 7, characterized in that, The real-time physiological and body posture sensing data are collected through a flexible tactile sensor array with a density of not less than 256 points / m².

9. The method according to claim 7, characterized in that, Also includes: After the occupants use the equipment, post-use data containing user feedback and the latest spinal curvature data is collected, and the post-use data is input into the artificial intelligence dynamic skeletal model for incremental training.

10. The method according to claim 7, characterized in that, The AI-powered dynamic skeletal model is pre-trained using no fewer than 100,000 sets of labeled spinal health data across all age groups, and its inference process is entirely completed on the vehicle side.

Images