Physical education method and system based on VR virtual reality

Abstract

The invention discloses a physical education method and system based on VR (virtual reality), and relates to the field of cross application of information technology, VR technology, computer-aided teaching and sports science. The method comprises the steps that kinematic chain data and physiological data of a student are acquired, and the fatigue state of the student is determined in real time through a biological-power chain fatigue model. The core of the invention lies in that in response to a fatigue state (for example, when fatigue is increased), the system cooperatively adjusts two parameters: on one hand, a task difficulty parameter (for example, ball speed) is reduced so as to reduce exercise load; and on the other hand, random disturbance parameters (such as visual tilt or tactile noise) are improved to actively promote neural adaptation and sensory reweighting in a fatigue state. According to the method, the defect that the difficulty can only be passively reduced in the prior art is overcome, passive adaptation is changed into active learning, and the training effect is remarkably improved.

Description

Technical Field

[0001] This invention relates to the interdisciplinary application of information technology, virtual reality (VR) technology, computer-aided instruction, and sports science. More specifically, this invention relates to an adaptive virtual reality system and method for sports teaching, competitive training, or sports rehabilitation. This system can dynamically and collaboratively adjust challenge and disturbance parameters in the virtual training environment based on real-time, multi-dimensional assessment of the trainee's biomechanical and physiological states. Background Technology

[0002] Virtual reality technology, with its ability to provide immersive, controllable, and repeatable training environments, is increasingly being applied in professional sports education, competitive training, and sports rehabilitation. These systems allow trainees to simulate high-intensity competition scenarios, repeat specific technical movements, and receive real-time data feedback in a safe virtual environment.

[0003] To optimize the training experience and prevent learners from feeling frustrated due to excessively difficult tasks or boredom due to overly easy tasks, some existing "exercise games" have introduced Dynamic Difficulty Adjustment (DDA) mechanisms. However, these DDA mechanisms have significant limitations in their technical implementation. On the one hand, they are mostly based on passive feedback adjustments made according to the learner's task performance (e.g., score, hit rate, completion time). For example, when the system detects consecutive mistakes by a learner, it might simply reduce the task difficulty, such as slowing down the movement speed of a virtual opponent or the flight speed of virtual sports equipment (e.g., a tennis ball). This simple adjustment method cannot distinguish the root cause of the learner's performance decline: whether it is due to insufficient skill, lack of concentration, or decreased motor control due to physiological fatigue.

[0004] On the other hand, some independent systems have emerged in this field attempting to monitor trainee fatigue. These systems may utilize eye-tracking sensors built into VR head-mounted displays (HMDs) to detect visual fatigue, or use sensors such as accelerometers to track head and hand movements to infer fatigue. Some studies even employ external biosensors, such as heart rate monitors or surface electromyography (sEMG) sensors, to assess trainees' physiological load. However, these fatigue monitoring systems are typically "open-loop," their primary function being to report fatigue levels or simply halt training when a certain threshold is reached.

[0005] Therefore, existing technologies have a fundamental flaw that urgently needs to be addressed: they generally treat "fatigue" as a purely negative state, and the only countermeasure is to passively reduce the load or stop training completely.

[0006] First, current technology fails to delve into the biomechanical level to understand fatigue. In sports, fatigue (especially core muscle fatigue) often leads to a breakdown in the athlete's kinetic chain. To maintain apparent athletic performance (such as ball speed), athletes unconsciously employ compensatory errors, such as overusing the shoulders and elbows to compensate for decreased trunk strength. This not only reduces training efficiency but is also a major cause of sports injuries. Existing task-performance-based dynamic monitoring (DDA) or isolated physiological monitoring cannot identify this hidden and dangerous biomechanical compensation.

[0007] Secondly, and most importantly, current technology fails to recognize the positive benefits of training under fatigue for elite athletes. A core element of high-level athletic skill is the ability to maintain efficient and stable motor control even when fatigue weakens bodily sensory signals (such as proprioception) and impairs central nervous system command transmission. Current technology reduces all challenges upon detecting fatigue, effectively depriving trainees of valuable opportunities to train their "fatigue resistance" and "sensory integration abilities."

[0008] In summary, there is an urgent need in this field for a novel technical solution that must: first, not only monitor heart rate or performance, but also accurately and multidimensionally (combining biomechanics and physiology) identify the true fatigue that leads to the collapse of movement patterns; second, after identifying this fatigue state, instead of simply stopping training, it should proactively and strategically adjust the VR environment to provide a new active training paradigm aimed at promoting neuroadaptability and sensory integration. Summary of the Invention

[0009] The purpose of this invention is to provide a VR-based physical education teaching method and system to address the shortcomings of existing VR training systems in fatigue management, as pointed out in the background art. These shortcomings include the inability to accurately identify biomechanical fatigue and the inability to passively reduce the difficulty when trainees are fatigued, rather than actively and proactively utilizing fatigue to promote deeper levels of motor learning.

[0010] In a first aspect, an embodiment of the present invention provides a physical education teaching method based on VR (Virtual Reality), comprising: Acquire trainees' kinetic chain data and physiological data; Based on the kinetic chain data and the physiological data, a bio-kinetic chain fatigue model is established, and the fatigue state of trainees is determined in real time using the model. In response to the fatigue state, at least two parameters in the VR virtual reality environment are adaptively adjusted simultaneously: A task difficulty parameter; and A random perturbation parameter.

[0011] Optionally, the step of adaptively adjusting at least two parameters in the VR virtual reality environment in response to the fatigue state includes: When the fatigue state indicates an increase in fatigue level, the task difficulty parameter is reduced and the random perturbation parameter is increased.

[0012] Optionally, the kinematic chain data includes: posture, angular velocity, or acceleration data of at least one joint of the trainee's body, acquired by at least one inertial measurement unit sensor.

[0013] Optionally, the physiological data includes: heart rate variability data and / or surface electromyography signal data.

[0014] Optionally, the bio-kinetic chain fatigue model includes: Based on the kinetic chain data, the biomechanical fatigue state of the trainee is determined; and Based on the physiological data, the physiological fatigue state of the trainee is determined; The fatigue state is determined based on the fusion of the biomechanical fatigue state and the physiological fatigue state.

[0015] Optionally, the task difficulty parameters include at least one of the following: the movement speed of the virtual opponent, the emission frequency of the virtual sports equipment, the flight speed of the virtual sports equipment, or the response time window for the trainee to complete the training task.

[0016] Optionally, the random perturbation parameters include: visual perturbations applied to the VR virtual reality environment, including random tilting of the virtual scene, transient noise in the field of view, or random positional offset of virtual objects.

[0017] Optionally, the random perturbation parameter further includes: proprioceptive perturbation applied to the trainee, the proprioceptive perturbation including non-periodic vibration applied by a haptic feedback device or random force feedback applied to virtual exercise equipment.

[0018] Optionally, the method further includes: Before implementing the teaching method, baseline kinetic chain data and baseline physiological data of the trainees when performing the corresponding sports in the real world are obtained. The bio-kinetic chain fatigue model determines the fatigue state based on the differences between the baseline data and the kinetic chain data and physiological data obtained in the VR virtual reality environment.

[0019] Secondly, an embodiment of the present invention provides a VR-based physical education teaching system, comprising: At least one sensor is used to acquire the trainee's kinetic chain data and physiological data; A processor; A memory storing a computer program that, when executed by the processor, implements the method as described in any one of the first aspects.

[0020] The present invention has achieved the following beneficial effects: It achieves accurate fatigue assessment from multiple dimensions and can identify hidden compensations.

[0021] Existing technologies either rely solely on a single physiological indicator (such as heart rate) or solely on kinematic performance (such as head movement). This invention constructs a "bio-kinetic chain fatigue model" that integrates biomechanics (based on kinetic chain theory) and physiology (central fatigue and local muscle fatigue), enabling earlier and more accurate identification of fundamental fatigue leading to decreased athletic performance. Crucially, this model can accurately identify dangerous biomechanical compensation patterns (i.e., kinetic chain collapse) that are currently imperceptible to existing technologies, allowing for intervention before damage occurs.

[0022] It enables active neural training under fatigue conditions, transforming "passive adaptation" into "active learning".

[0023] To address the shortcomings of the "fatigue-stop" approach in the prior art, this invention provides a revolutionary collaborative adjustment strategy. When the trainee is fatigued, the system reduces the explicit task difficulty (such as ball speed) to prevent motion overload; on the other hand, it (non-obviously) increases random perturbation parameters (such as visual tilt or tactile noise).

[0024] This collaborative strategy of "reducing difficulty and increasing perturbation" is based on the principle of "stochastic resonance." Fatigue weakens the trainee's internal sensory signals (such as proprioception). At this time, the random perturbation (noise) applied by the system does not increase the burden, but (counterintuitively) amplifies the brain's perception threshold for these weak internal signals.

[0025] At the same time, these perturbations (especially visual and tactile perturbations) force the brain to reduce its reliance on unreliable external signals and instead increase the processing weight of internal signals (proprioception, vestibular sensation), a process known as "sensory re-weighting".

[0026] The end result is that the trainees' nervous system is actively and safely trained in a state of fatigue, their neuroplasticity is stimulated, and they learn how to control their bodies more effectively when fatigued, thus solving the fundamental defect of the background technology that fatigue means stopping training.

[0027] This greatly improves the transferability of training results to the real world.

[0028] A common challenge with existing VR training is the difficulty in transferring its effects to the real world. This invention addresses this by acquiring baseline data of the learner's real-world movements before training and rigorously defining "fatigue" as a "performance drift"—a deviation of the learner's movement patterns during VR training from this "real-world baseline." This ensures that the system trains and optimizes only the learner's efficient movement patterns in the real world, rather than an isolated "virtual game score." This fundamentally solves the problems of effectiveness and transferability in VR training.

[0029] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0030] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0031] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a VR-based physical education teaching method according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a VR-based physical education teaching system according to an embodiment of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0033] Example 1: This embodiment provides a physical education teaching method based on VR (Virtual Reality). (Refer to...) Figure 1The process illustrated includes the following steps in a typical execution cycle: First, the system initiates a data acquisition process via sensors configured on the student's body or VR device. This process continuously captures multimodal data streams in real time as the student performs specific sports movements (e.g., tennis swing, golf swing, skiing, etc.) in an immersive VR environment. These data streams are designed to include at least two types of key information: kinetic chain data reflecting the kinematic characteristics of different parts of the student's body, and physiological data reflecting the student's internal physiological state.

[0034] After acquiring this data, the system's internal data processing unit (e.g.) Figure 2 The processor shown will invoke a pre-built or real-time calculated "bio-kinetic chain fatigue model". The specific construction and working principle of this model will be explained in detail in subsequent embodiments (especially Embodiment 5). The core function of this model is to receive and process the aforementioned kinetic chain data and physiological data, and determine the trainee's current fatigue state in real time and dynamically through a fusion analysis algorithm. This fatigue state can be a continuous quantitative fatigue index (e.g., 0 to 100) or a discrete classification fatigue level (e.g., State 1 "fresh / no fatigue", State 2 "moderate fatigue / compensation has appeared", State 3 "high fatigue / near exhaustion").

[0035] Next, the system enters an adaptive adjustment closed-loop feedback phase. The system control logic responds to the real-time fatigue state determined by the model. This "response" is one of the core features of the invention; it is not a simple "stop" command, but rather the execution of a coordinated (i.e., synchronous) adaptive adjustment. Specifically, the system will simultaneously (i.e., within one adjustment cycle) adaptively adjust at least two parameters with different properties in the VR virtual reality training environment.

[0036] The first parameter is a task difficulty parameter. This parameter defines the explicit challenge level or extrinsic load of the training task that the trainee needs to complete. For example, it determines the strength of the virtual opponent or the speed of the virtual equipment.

[0037] The second parameter is a random perturbation parameter. This parameter defines the level of unexpected, random interference or sensory challenge imposed on the learner by the training environment. For example, it determines the stability of the visual scene or the degree of interference from tactile feedback.

[0038] This closed-loop system, which takes the trainees' biomechanical and physiological states as input, judges them through a fusion model, and finally outputs a two-parameter (difficulty and perturbation) collaborative adjustment strategy, constitutes the basic framework of the method of this invention.

[0039] Example 2: This embodiment, based on the framework of Embodiment 1, elaborates in detail a preferred and highly innovative core working principle and logic of the adaptive adjustment step. In a preferred implementation, when the fatigue state determined by the bio-kinetic chain fatigue model indicates that the trainee's fatigue level is increasing (e.g., from "moderate fatigue" to "high fatigue", or the quantified fatigue index exceeds a preset higher threshold), the system control logic will execute a specific, non-intuitive collaborative adjustment strategy: on the one hand, the system reduces the task difficulty parameter; on the other hand, the system increases the random perturbation parameter.

[0040] The scientific principle and technical purpose behind this strategy of "reducing difficulty and increasing perturbation" is to solve the defect of "stopping when fatigued" in the background technology and realize active neural training in a state of fatigue.

[0041] First, the working principle of reducing task difficulty parameters is explained in detail. This is done to reduce the learner's explicit motor and cognitive load. When learners are highly fatigued, their muscle strength, explosive power, reaction speed, and fine motor control all decline significantly. Maintaining a high-difficulty task at this time (such as a high-speed virtual tennis ball) will lead to repeated failures, which not only generates a strong sense of frustration and severely damages training motivation, but more importantly, learners may enter a state of exhaustion in order to "force" themselves to complete the movement, leading to a complete breakdown of their movement patterns or a significant increase in the risk of sports injuries in the real world. Therefore, by (for example) reducing the speed of the virtual opponent or the ball speed (as detailed in Example 6), the system ensures that the learner's physical and cognitive load is reduced to a controllable range, allowing them to continue to execute and complete technical movements, keeping them within a "controllable challenge" zone.

[0042] Secondly, the working principle of improving the random perturbation parameters is explained in detail. This is the key innovation of this invention. Its purpose is not to further increase the burden on trainees or "punish" them, but to actively and strategically promote neural adaptation under fatigue. Its working principle is based on two major mechanisms in neuroscience: "random resonance" and "sensory reweighting".

[0043] "Random resonance" is a phenomenon observed in nonlinear systems, such as the human nervous system, where a suitable amount of random noise input of a specific intensity can enhance the system's ability to detect and process weak signals. In the context of this invention, the student's "fatigue" state, at the neural level, means that their internal sensory signals (especially proprioceptive signals from muscles, tendons, and joints) become weak, blurred, and the "signal-to-noise ratio" decreases. At this time, a controlled random perturbation (such as the non-periodic vibration detailed in Example 8) actively applied by the system through (for example) a tactile device acts as the "noise" required for the "random resonance" mechanism. This external noise interacts nonlinearly with the student's weak internal sensory signals, (counterintuitively) making it easier for the student's central nervous system to "capture" these weak proprioceptive signals, thereby temporarily improving body awareness and the precision of motor control during fatigue.

[0044] "Sensory reweighting" refers to the brain's ability to dynamically adjust the degree of dependence (i.e., "weighting") on information from each sensory channel when faced with multiple sensory inputs (visual, auditory, proprioceptive, and vestibular). When the random perturbation parameters are increased, especially visual perturbations (such as the random tilt of the scene detailed in Example 7), the system intentionally creates a sensory conflict. For example, random scene tilt makes the "balance" and "spatial" information provided by the learner's visual system unreliable. In order to maintain posture control and action execution, the learner's brain is forced to perform "sensory reweighting": that is, actively reducing dependence on the currently unreliable visual signals and increasing dependence on other sensory channels (such as vestibular sensation and proprioception enhanced by "random resonance").

[0045] In summary, this collaborative strategy of "reducing difficulty and increasing perturbation" achieves the following technical results: when trainees are fatigued, it reduces the challenge to their "muscle strength" and "reaction speed" (by reducing difficulty), but increases the challenge to their "neural control" and "sensory integration" (by increasing perturbation). This strategy actively induces the brain's neuroplasticity, safely and efficiently training trainees' core abilities to efficiently integrate sensory information and maintain motor control even when fatigued (i.e., when internal signals are weak and external signals are unreliable).

[0046] Example 3: This embodiment details the source and working principle of the "kinematic chain data" obtained in Embodiment 1. In a preferred implementation, the kinematic chain data is acquired by deploying at least one inertial measurement unit (IMU) sensor at at least one key joint point on the trainee's body.

[0047] These IMU sensors are small, microelectromechanical systems (MEMS) based electronic devices. They can be integrated into wearable straps, clothing, or VR trackers and attached to specific parts of the learner's body, such as (but not limited to): wrists, elbows, and shoulders (to analyze upper limb kinetic chains); or attached to the hips (e.g., on a belt) and ankles (to analyze lower limb kinetic chains); or attached to the sternum or upper back (to analyze core trunk movements).

[0048] The core internal components of each IMU sensor typically include at least one three-axis accelerometer and one three-axis gyroscope.

[0049] The triaxial accelerometer is used to measure the linear acceleration of the IMU sensor (i.e., the joint at which it is located) in three-dimensional space (orthogonal axes x, y, z). Its descriptive operating principle is as follows: it detects the displacement of a tiny, movable mass ("test mass") within the sensor housing relative to the sensor. As the sensor accelerates, this mass lags behind due to inertia, resulting in displacement; simultaneously, gravity also continuously causes this displacement. The accelerometer quantifies this displacement by measuring, for example, changes in capacitance or piezoresistive pressure, thus outputting a mixed signal that includes both kinetic and gravitational acceleration.

[0050] The triaxial gyroscope (or angular rate sensor) is used to measure the angular velocity (i.e., the speed and direction of rotation) of the IMU sensor about its three orthogonal axes (e.g., pitch, roll, and yaw). Its working principle is typically based on the Coriolis effect: a tiny internal structure (such as a tuning fork or vibrating ring) is driven to produce continuous vibration; when the sensor rotates, the Coriolis effect causes the plane of this vibration to deflect; the sensor calculates its angular velocity by detecting the degree of this deflection.

[0051] As the trainee moves, the IMU fixed at the joint moves accordingly. The microprocessor inside the sensor, or the data fusion algorithm that transmits the raw data stream to the system's main processor, performs data fusion and processing on the signals from the accelerometer and gyroscope (e.g., using Kalman filtering or complementary filtering algorithms).

[0052] The first step is to calculate the orientation of the sensor (i.e., the joint) by accumulating (i.e. integrating) the angular velocity signal output by the gyroscope over time. This means that the sensor is facing in three-dimensional space (e.g., pitch, roll, yaw angle).

[0053] The second step involves using the calculated attitude information to calculate the projection of the gravity vector onto the sensor coordinate system in real time.

[0054] The third step is to subtract this gravitational component from the total signal from the accelerometer to obtain the linear acceleration data caused purely by motion.

[0055] The fourth step is to obtain the velocity by accumulating (integrating) the linear acceleration once, and the displacement by accumulating it again (although in practical applications, due to the accumulation of integration errors, it usually depends more on attitude and acceleration data).

[0056] Therefore, the IMU sensor kit in this embodiment can provide real-time data streams of posture, angular velocity, and / or acceleration at one or more joints of the trainee's body. These high-frequency kinematic data collectively constitute "kinematic chain data," providing crucial biomechanical input for the "bio-kinetic chain fatigue model" in the subsequent embodiment five.

[0057] Example 4: This embodiment details the source and working principle of the "physiological data" obtained in Embodiment 1. In a preferred implementation, the physiological data is designed to simultaneously cover both central fatigue and local fatigue, specifically including heart rate variability (HRV) data and / or surface electromyography (sEMG) signal data.

[0058] First, the working principle of heart rate variability (HRV) data will be explained in detail. HRV data is primarily used to assess the central nervous system fatigue state and the balance state of the autonomic nervous system (ANS) of trainees. HRV does not refer to the speed of heart rate (i.e., the number of heartbeats per minute), but rather to the minute variations or irregularities in the time intervals between consecutive heartbeats (i.e., the RR interval).

[0059] Its working principle is as follows: The human heartbeat is precisely regulated by the autonomic nervous system (ANS). The two main branches of the ANS, namely the sympathetic nervous system (SNS, which is responsible for the "fight or flight" response, promoting an accelerated and "regular" heartbeat) and the parasympathetic nervous system (PNS, which is responsible for the "rest and digestion" response, promoting a slowed and "irregular" heartbeat), are in a state of continuous dynamic antagonism and balance.

[0060] When trainees are in good spirits and have recovered well, the regulation of the parasympathetic nervous system (PNS) is dominant, and the variability of the heart rate interval is high (i.e., the HRV value is high), indicating that the autonomic nervous system is highly adaptable and has sufficient reserves.

[0061] Conversely, when trainees are under high levels of physical or mental stress or central fatigue, the activity of the sympathetic nervous system (SNS) becomes dominant, while the regulatory capacity of the parasympathetic nervous system is suppressed. This leads to a more "rigid" and "rhythmic" heartbeat pattern, with a significantly reduced variability in heart rate intervals (i.e., low HRV values).

[0062] The system acquires a real-time stream of heart rate interval data via, for example, an electrocardiogram (ECG) sensor (e.g., a heart rate monitor) worn on the trainee's chest or a photoplethysmography (PPG) optical sensor worn on the wrist / inside a VR headset. The processor then calculates relevant time-domain metrics of HRV (e.g., RMSSD, the root mean square of the difference between adjacent RR intervals) and / or frequency-domain metrics (e.g., the ratio of high-frequency power (HF) to low-frequency power (LF)). By analyzing trends in these HRV metrics (e.g., a sustained decrease in RMSSD relative to baseline), the system can quantify the trainee's level of central fatigue.

[0063] Secondly, the working principle of surface electromyography (sEMG) signal data is explained in detail. The sEMG data is mainly used to assess the local muscle fatigue state of specific muscles (i.e., the main force-generating muscle groups that perform the movement). sEMG detects and records the bioelectrical signals generated by a muscle during contraction by using non-invasive electrodes placed on the skin surface of a specific muscle (e.g., the deltoid muscle or forearm muscles in a tennis swing, or the quadriceps muscle in a squat).

[0064] Its working principle is as follows: when the brain issues a movement command, motor neurons are activated, causing the muscle fibers they innervate to generate "action potentials" (a type of electrochemical signal). The spatiotemporal sum of action potentials from a large number of muscle fibers is conducted in muscle tissue and skin, ultimately forming a weak, complex sEMG signal that can be detected by surface electrodes.

[0065] When a muscle begins to fatigue, its sEMG signal undergoes a series of characteristic, quantifiable changes: Changes in signal amplitude: During sustained, non-exhaustive contractions, in order to maintain the same force output (to compensate for the contribution of some fatigued muscle fibers), the central nervous system recruits more motor units and increases their firing frequency. This is reflected in sEMG signals as a gradual increase in signal amplitude (e.g., calculated by the root mean square value RMS).

[0066] Changes in the signal spectrum: During muscle fatigue, the metabolic environment inside the muscle changes (e.g., the accumulation of metabolic products such as lactic acid), which slows down the conduction velocity (CV) of action potentials on the muscle fiber membrane. This slowdown in action potential conduction velocity manifests in signal processing as a "compression" or "shift" of the sEMG signal's power spectrum towards the lower frequency region. Specifically, the median frequency (MDF) or mean power frequency (MPF) of the sEMG signal will significantly decrease.

[0067] The system quantifies the local fatigue level of specific key muscle groups in trainees by analyzing the upward trend of the amplitude (e.g., RMS value) and / or the downward trend of the frequency (e.g., MDF value) of sEMG signals.

[0068] Example 5: This embodiment details a preferred implementation of the "biological-kinetic chain fatigue model" described in Embodiment 1. In this implementation, the model is constructed as a deep learning model of a hybrid convolutional neural network-long short-term memory network-attention (HybridCNN-LSTM-Attention). This model is pre-trained to receive multimodal temporal data (kinetic chain data and physiological data) as input and outputs a classification of the trainee's current fatigue state (e.g., Level 0: "fresh"; Level 1: "moderate fatigue / compensation"; Level 2: "high fatigue / near exhaustion").

[0069] 1. Data Acquisition and Feature Engineering Before the model can operate (i.e., determine fatigue state in real time), it needs to be trained using training data. Both the training data and the real-time input data include the following preprocessed and extracted feature vectors. The data sources are as described in Examples 3 (IMU) and 4 (HRV, sEMG).

[0070] To achieve a robust model capable of recognizing "power chain collapse", this embodiment preferably extracts the feature-engineered representation shown below.

[0071] First, regarding biomechanical or kinetic chain data, this data preferably comes from the inertial measurement unit (IMU) sensor described in Example 3. From this IMU data, the model extracts features for evaluating core (proximal) force exertion, such as peak angular velocity of trunk rotation; and features for evaluating distal velocity, such as peak angular velocity of elbow extension. More importantly, the model extracts a kinetic chain temporal feature, such as the time difference between the peak proximal trunk rotation and the peak distal elbow extension, which is used to evaluate kinetic chain efficiency (i.e., the "whiplash effect"). Simultaneously, other kinematic features, such as the root mean square (RMS) of triaxial linear or angular acceleration, or jerk, can also be extracted to evaluate movement smoothness and explosiveness.

[0072] Secondly, regarding physiological data, this data preferably comes from the surface electromyography (sEMG) sensor described in Example 4. From the sEMG signal, the model extracts time-domain and frequency-domain features. Time-domain features (e.g., integrated electromyography (iEMG) and root mean square (RMS) values) are used to assess the activation level of local muscles. Frequency-domain features (e.g., median frequency (MDF) and mean power frequency (MPF)) are used to assess local muscle fatigue, as a decrease in MDF is a recognized indicator of fatigue.

[0073] Similarly, heart rate variability (HRV) data from Example 4 were also used to extract features. For example, time-domain features such as the root mean square (RMSSD) of the difference between adjacent RR intervals and frequency-domain features such as the ratio of low-frequency power (LF) to high-frequency power (HF) (LF / HF) were extracted. These features were primarily used to assess the state of the autonomic nervous system and central fatigue.

[0074] Finally, the model also extracts a key performance feature, namely baseline drift. This feature quantifies the degree of "biomechanical compensation" by calculating the Euclidean distance or dynamic time warping (DTW) distance between the current kinetic chain features (such as the kinetic chain temporal features mentioned above) and the "real-world baseline" described in Example 9.

[0075] 2. Hybrid Fusion Model Architecture In this embodiment, the feature data streams from different sensors and processed by feature engineering (e.g., data sampled within a 200-millisecond sliding time window) are input into the hybrid model. The preferred architecture of the model is designed to process the time-series data sequentially, with the following design objectives: (1) CNN layers are used to automatically extract spatial (or local) features from the multimodal data; (2) LSTM layers are used to capture the temporal dependencies in the fatigue accumulation process; and (3) an Attention mechanism is used to assign higher weights to the "keyframes" that determine the fatigue state.

[0076] The specific architecture of the model preferably includes: data first passes through an input layer, the shape of which corresponds to the size of the time window and the number of features.

[0077] Next, the input data is first fed into a convolutional neural network (CNN) component to automatically extract spatial or local features from the multimodal data. This component preferably includes at least two one-dimensional convolutional (1D-CNN) layers. For example, the first 1D-CNN layer can be configured with 64 filters and a convolutional kernel of size 3, using the ReLU activation function; the second 1D-CNN layer can be configured with 128 filters and a convolutional kernel of size 3, also using the ReLU activation function, to extract higher-dimensional composite features. After the convolutional layers, a max-pooling layer, for example, with a pooling size of 2, is preferably set to reduce the dimensionality of the feature map and retain the most salient features.

[0078] The feature sequences processed by the CNN components are then fed into a Long Short-Term Memory (LSTM) layer. This LSTM layer is preferably configured, for example, to have 64 units and is set to return complete sequences (i.e., return_sequences=True). The function of this LSTM layer is to capture long-term temporal dependencies in the sequences, which is crucial for modeling the cumulative nature of fatigue.

[0079] The output sequence of the LSTM layer is then passed to a self-attention mechanism layer. The function of this attention mechanism is to dynamically assign different weights to time steps (i.e., "keyframes") in the sequence, so that the model can automatically and more focus on those time points that are most important for judging the fatigue state, such as fatigue inflection points or the moment of compensation.

[0080] The attention-weighted sequence features are then aggregated into a single feature vector by a global average pooling layer. This vector is then fed into a fully connected (dense) layer, for example, with 64 neurons and a ReLU activation function, for the final non-linear feature integration.

[0081] Finally, the model sets up an output layer, which is preferably a fully connected layer with, for example, 3 neurons (corresponding to three fatigue levels) and a Softmax activation function to output the probability distribution of the three fatigue levels (e.g., level 0: "fresh"; level 1: "moderate fatigue"; level 2: "high fatigue").

[0082] 3. Model Training and Implementation The model is trained using labeled data for supervised learning. The labeled data (i.e., “real fatigue state”) can be obtained, for example, by: (1) synchronously acquiring the “real-world baseline data” described in Example 9 as a “fresh” state; (2) requiring the trainee to perform exhaustive exercise (e.g., continuously swinging the racket at high intensity until the MDF value of their sEMG signal drops by more than 30%, or their kinetic chain timing deviates from the baseline by more than 3 standard deviations), and labeling the state as “high fatigue / compensation”.

[0083] Through the model architecture and feature engineering described above, the system can receive real-time data streams from Examples 3 and 4 and reliably and in real-time output (e.g., updated every second) the trainee's current fatigue state, thereby providing a clear, quantifiable, and biomechanically and physiologically integrated decision-making basis for subsequent adaptive adjustment steps. This solves the technical problem that existing technologies cannot quantify and identify "biomechanical compensation."

[0084] Example 6: This embodiment details the specific implementation of the "task difficulty parameter" described in Embodiment 2. This parameter is crucial for adjusting the explicit training load of trainees. Upon receiving an "increased fatigue" signal from the bio-kinetic chain fatigue model (Embodiment 5), the Dynamic Difficulty Adjustment (DDA) module in the system executes the instruction to "reduce the task difficulty parameter."

[0085] This module achieves this by modifying procedural parameters of the VR training environment. In a training scenario for a tennis, table tennis, or ice hockey goalkeeper (for example), the task difficulty parameters may include at least one of the following: Virtual opponent's movement speed: In competitive training, the system can reduce the lateral movement speed, starting speed, or reaction speed of a virtual AI opponent (e.g., a tennis opponent) on the court, thereby reducing the threat of their return shot.

[0086] Launch frequency of virtual sports equipment: In repetitive training (e.g., in the mode of simulating a ball machine), the system can reduce the launch frequency of virtual sports equipment (e.g., tennis balls, table tennis balls, ice hockey balls). For example, the launch interval of a "tennis launcher" can be increased from one ball every 3 seconds to one ball every 5 seconds, giving trainees more time for preparation and physiological recovery.

[0087] Flight speed of virtual sports equipment: This is one of the most direct ways to adjust the difficulty. The system can reduce the flight speed of the virtual ball (for example, reducing the average speed of an incoming tennis ball from 100 km / h to 80 km / h), giving trainees more time to judge the trajectory of the ball, prepare their steps, and organize their return movements.

[0088] The system can relax the time window for determining "effective shot" or "optimal hitting point". For example, in a "fresh" state, the system may require the student to hit the ball within 20 milliseconds before or after the ball reaches the optimal hitting point to be considered "perfect"; in a "fatigued" state, the system can relax this time window to 50 milliseconds, thereby reducing the student's time pressure and cognitive load.

[0089] By finely adjusting one or more of the above parameters, the explicit difficulty of the system can be reduced, thereby matching the decreased athletic and reaction abilities of trainees when they are fatigued, and preventing them from entering an "overload" state, as explained in detail in Example 2.

[0090] Example 7: This embodiment details a specific implementation of the "random perturbation parameter" described in Embodiment 2, namely, visual perturbation. Upon receiving an "increased fatigue level" signal from the fatigue model (Embodiment 5), the perturbation generation module in the system executes an instruction to "increase the random perturbation parameter." The purpose of this is to actively induce "sensory reweighting," as detailed in Embodiment 2.

[0091] Crucially, these visual disturbances must be random, non-periodic, and unpredictable. If the disturbances are regular and predictable (e.g., constant sinusoidal scene shaking), the learner's brain will quickly adapt to this pattern and "filter" it out of their perception, or learn to anticipate it, thus completely negating the training effect of "sensory reweighting."

[0092] The visual perturbation may include at least one of the following, which are generated and applied in real time by the system's graphics rendering engine: Random tilting of the virtual scene: The system applies a random, small-amplitude (e.g., a random value between 0 and 5 degrees) roll (tilt left and right) or pitch (tilt forward and backward) to the entire virtual scene rendered to the learner (i.e., the "field of vision" or "world coordinate system"). The timing, angle, direction, and duration of this tilt are all determined by a pseudo-random number generator. This creates a momentary conflict between the learner's visual input ("the world is tilting") and their vestibular input (the inner ear's perception of the direction of gravity has not changed), forcing the central nervous system to immediately suppress unreliable visual signals and instead rely more on vestibular and proprioceptive senses to maintain balance and orientation.

[0093] Transient noise in the visual field: The system randomly superimposes a layer of "visual noise" into the trainee's visual field (random time point, random duration, e.g., 50 to 200 milliseconds). This noise can manifest as simulated television snow, brief global or local blurring effects, or momentary brightness flicker. This transient loss of visual information "blinds" the trainee's hand-eye coordination loop, forcing the trainee to rely on their internal motion model and proprioception to predictively complete actions in the absence of clear visual feedback.

[0094] Random position offset of virtual objects: In a task such as catching or hitting a ball, the system distinguishes between the object's physical position (calculated by the physics engine for collision detection) and its rendered position (the visual image seen by the learner in VR). When a perturbation is applied, the system applies a small (but significant) random spatial position offset to the image of the "virtual ball" rendered to the learner, based on the "real" ball trajectory calculated by the physics engine. This makes purely visual "hand-eye coordination" unreliable, because the "ball seen by the eye" is not in its "real physical position." This also forces learners to rely more on proprioception of their own limb position and movement trajectory for predictive control rather than reactive control.

[0095] Example 8: This embodiment details another specific implementation of the "random perturbation parameter" described in Embodiment 2, namely, proprioceptive perturbation. This perturbation works in conjunction with the visual perturbation in Embodiment 7 to further enhance the effect of sensory conflict and more directly applies the "random resonance" principle described in Embodiment 2.

[0096] The random perturbation parameters may also include perturbations applied to the learner that are designed to directly stimulate the proprioceptive channel. These perturbations are implemented using specialized haptic feedback devices. These devices may be VR controllers held by the learner, specially designed (e.g., with vibration motors) motion equipment models (such as racket handles, golf club grips), wearable haptic vests, or limb straps.

[0097] The proprioceptive perturbation includes at least one of the following: Non-periodic vibrations applied by haptic feedback devices: This is a preferred implementation of "random resonance". The system controls the tactile device to generate sub-threshold (which is almost imperceptible to the learner) or near-threshold (which is slightly perceptible), non-periodic vibrations.

[0098] Unlike the strong, short, regular vibrations ("force feedback") used in traditional VR to simulate "collisions" or "shooting," this perturbation is not intended to simulate any specific event, but rather to be continuously applied as a "beneficial background noise."

[0099] The implementation is as follows: the processor does not simply send a "turn on (frequency, amplitude)" command to the vibration motor of the haptic unit, but instead utilizes buffered haptics. The processor generates a pseudo-random amplitude sequence (e.g., a byte stream consisting of random byte values ​​between 0 and 255) in real time and continuously pushes this data stream to the vibration buffer of the haptic device at a high frequency (e.g., 320 times per second). As the motor "plays" this random buffer, it produces a weak, "chaotic" or "irregular" vibration. This vibration continuously and slightly stimulates the mechanoreceptors (i.e., proprioceptors) of the skin and joints of the learner's hands, arms, or torso, thereby (as described in Example 2) amplifying the learner's central nervous system's ability to perceive its own weak internal motion signals (proprioception).

[0100] Random force feedback applied to virtual exercise equipment: If the learner holds a device with active force feedback (e.g., a specially designed racket handle or steering wheel with a built-in motor or electromagnetic actuator), the system can apply a small, random, and unintended torque or resistance along the critical path of the learner's movement (e.g., the acceleration phase of the swing). For example, a torsional torque of 1 Newton-meter could be randomly applied during the swing for 50 milliseconds. This direct physical disturbance (rather than just a sensory "vibration") forces the learner's neuromuscular system to react instantly, dynamically adjusting the activation patterns of relevant muscle groups (e.g., enhancing the "active stiffness" of the forearm pronator and wrist flexors) to maintain wrist stability and proper racket face orientation. This directly and intensively trains the learner's robust control under uncertainty (i.e., fatigue or disturbance).

[0101] Example 9: This embodiment details a pre-calibration step that is crucial to the method of the present invention (especially the fatigue model in Embodiment 5). This step aims to address the core problem of low "effect transferability" commonly found in VR training in the background art, namely, how to ensure that training results in VR can be effectively transferred to real-world athletic performance.

[0102] The method preferably includes a baseline acquisition initialization procedure. This procedure can be executed when a trainee first uses the system, or at the beginning of each new training cycle (e.g., when the trainee is physically fit and technically proficient) to update their baseline data.

[0103] In this “baseline acquisition” procedure, the system will ask the trainee to remove the VR headset and use real sports equipment (e.g., the trainee’s own racket or club) in the real world (e.g., on a real tennis court, running track or golf driving range).

[0104] Trainees are required to wear the full sensor kit used in the system of this invention (i.e., the IMU sensor of Embodiment 3 and the HRV / sEMG sensor of Embodiment 4) and, under guidance, perform a series of standardized real-world sports activities corresponding to the VR training content. For example, performing 10 maximum effort tennis forehand shots or running at 80% of maximum heart rate for 5 minutes.

[0105] While trainees perform these real-world movements, the system simultaneously acquires and stores baseline kinetic chain data (e.g., the timing and amplitude characteristics of their most efficient kinetic chain) and baseline physiological data (e.g., their HRV level in their "fresh" state, and sEMG response maps when performing standard movements) in their "real and optimal" state. This data is stored in the system's memory, forming the trainee's "gold standard" or "personal reference profile."

[0106] Subsequently, when the trainee puts on the VR device and begins to implement the teaching method of the present invention, a core working principle of the "bio-kinetic chain fatigue model" described in Example 5 is to continuously compare the kinetic chain data and physiological data acquired in real time in the VR environment with the trainee's real-world baseline data retrieved from the memory.

[0107] Here, the “fatigue state” of the present invention is operationally and rigorously defined as: the “deviation” or “drift” of the student’s current (in VR) data pattern relative to its “real-world baseline” data pattern.

[0108] The theoretical basis for this definition comes from the concepts of "performance fatigability" and "velocity decay" monitoring principles in modern sports science. For example: When the model (Example 5) detects that the trainee's kinetic chain pattern in VR (e.g., trunk-shoulder angular velocity ratio) begins to deviate from its true baseline kinetic chain pattern (i.e., compensation occurs), the model determines that "biomechanical fatigue" is occurring.

[0109] When the model detects that the physiological cost (e.g., sEMG amplitude) required for a learner to maintain a baseline movement pattern in VR is significantly higher than their baseline physiological cost, the model determines that "physiological fatigue" is occurring.

[0110] In this way, the present invention ensures that VR training remains anchored to the learner's real-world performance. The "fatigue" that the system combats and trains against is a "drift" in real-world movement patterns, rather than a decline in virtual game scores. This fundamentally guarantees that the neuroadaptations acquired in VR (e.g., through the strategy of Embodiment 2) can be transferred back to the learner's real-world athletic performance to the greatest extent possible.

[0111] Example 10: This embodiment provides a VR-based physical education teaching system for implementing the methods described in any of the above embodiments (Embodiments 1 to 9). (Refer to...) Figure 2 The system diagram shown can be a highly integrated VR all-in-one device or a distributed system consisting of a VR head-mounted display, a set of external sensors, and a host computer (e.g., a high-performance PC or cloud server).

[0112] The system preferably includes the following key components: At least one sensor: Physically, this component is a sensor suite used to comprehensively acquire the trainee's kinetic chain and physiological data. As shown in the figure, the sensor suite may include: Kinematic chain sensors: One or more inertial measurement units (IMUs) (as described in Example 3) are used to acquire attitude, angular velocity, and acceleration data at the joints. These IMUs can be partially integrated into the VR controller and headset, or partially worn as stand-alone sensors on the trainee's limbs and torso.

[0113] Physiological sensors: an electrocardiogram (ECG) sensor or photoelectric (PPG) sensor for acquiring heart rate variability (HRV) data (as described in Example 4); and a surface electromyography (sEMG) sensor for acquiring local muscle fatigue data (as described in Example 4).

[0114] One processor: The processor is the computing core of the system. It can be a central processing unit (CPU), a graphics processing unit (GPU), or a combination of AI acceleration chips (such as NPU or TPU) dedicated to running complex models. This processor is responsible for performing all computationally intensive tasks.

[0115] One memory: A memory is a non-volatile storage medium (e.g., a solid-state drive, RAM) that stores computer programs. When these computer programs are loaded and executed by a processor, they enable the system to implement the methods described in this invention.

[0116] The system may also include an output unit, for example: A display unit: This is typically an OLED or LCD screen in a VR headset, used to present an immersive virtual reality environment to learners.

[0117] A haptic unit: This can be a vibration motor or force feedback actuator integrated into a VR controller or used as a standalone device (such as a haptic vest, gloves, or straps).

[0118] The system's workflow during operation is as follows: Data acquisition: The at least one sensor continuously acquires the trainee's raw kinetic chain data and physiological data.

[0119] Data processing: Data is transmitted to the processor in real time.

[0120] Model execution: The processor executes a computer program stored in memory. This program contains at least: a. Baseline Management Module: Used to store and retrieve the real-world baseline data of trainees as described in Example 9.

[0121] b. Fatigue Model Module: This is the "Bio-kinetic Chain Fatigue Model" described in Example 5. This module receives real-time data from sensors and compares it with baseline data to determine the trainee's fatigue state in real time.

[0122] c. Adaptive Control Module: This is the "cooperative adjustment logic" described in Example 2. This module receives the fatigue state from the fatigue model as input and generates corresponding control commands based on preset logic (e.g., "when fatigue increases, reduce difficulty and increase disturbance").

[0123] Environmental feedback: The processor sends these control commands to the various output units to adaptively adjust the VR environment. a. Send instructions to the rendering pipeline of the display unit to adjust the task difficulty parameters (e.g., reduce the ball speed, as described in Example 6) and / or apply visual perturbations (e.g., randomly tilt the scene, as described in Example 7).

[0124] b. Send instructions to the haptic unit to apply a proprioceptive perturbation (e.g., play a non-periodic vibration buffer, as described in Example 8).

[0125] In this way, the system forms a complete, real-time closed loop from perception, modeling, decision-making to feedback, which can intelligently and adaptively realize the advanced sports teaching method as described in any one of Examples 1-9.

[0126] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A physical education teaching method based on VR (Virtual Reality), characterized in that, include: Acquire trainees' kinetic chain data and physiological data; Based on the kinetic chain data and the physiological data, a bio-kinetic chain fatigue model is established, and the fatigue state of trainees is determined in real time using the model. In response to the fatigue state, at least two parameters in the VR virtual reality environment are adaptively adjusted: a task difficulty parameter and a random perturbation parameter.

2. The method according to claim 1, characterized in that, The step of adaptively adjusting at least two parameters in the VR virtual reality environment in response to the fatigue state includes: When the fatigue state indicates an increase in fatigue level, the task difficulty parameter is reduced and the random perturbation parameter is increased.

3. The method according to claim 1, characterized in that, The kinematic chain data includes: posture, angular velocity, or acceleration data of at least one joint of the trainee's body, acquired by at least one inertial measurement unit sensor.

4. The method according to claim 1 or 3, characterized in that, The physiological data include: heart rate variability data and / or surface electromyography signal data.

5. The method according to claim 1, characterized in that, The bio-kinetic chain fatigue model includes: Based on the kinetic chain data, the biomechanical fatigue state of the trainee is determined; and Based on the physiological data, the physiological fatigue state of the trainee is determined; The fatigue state is determined based on the fusion of the biomechanical fatigue state and the physiological fatigue state.

6. The method according to claim 1 or 2, characterized in that, The task difficulty parameters include at least one of the following: the movement speed of the virtual opponent, the launch frequency of the virtual sports equipment, the flight speed of the virtual sports equipment, or the response time window for the trainee to complete the training task.

7. The method according to claim 1 or 2, characterized in that, The random perturbation parameters include: visual perturbations applied to the VR virtual reality environment, including random tilting of the virtual scene, transient noise in the field of view, or random positional offset of virtual objects.

8. The method according to claim 7, characterized in that, The random perturbation parameters also include: proprioceptive perturbations applied to the trainee, which include non-periodic vibrations applied by haptic feedback devices or random force feedback applied to virtual exercise equipment.

9. The method according to claim 1, characterized in that, The method further includes: Before implementing the teaching method, baseline kinetic chain data and baseline physiological data of the trainees when performing the corresponding sports in the real world are obtained. The bio-kinetic chain fatigue model determines the fatigue state based on the differences between the baseline data and the kinetic chain data and physiological data obtained in the VR virtual reality environment.

10. A VR-based physical education teaching system, characterized in that, include: At least one sensor is used to acquire the trainee's kinetic chain data and physiological data; A processor; A memory storing a computer program, which, when executed by the processor, implements the method as described in any one of claims 1-9.

Images