Digital human with real individual continuous growth and construction method

Abstract

This invention discloses a native digital human that continuously grows alongside a real individual and its construction method, comprising: establishing a mapping relationship between a target activity task and multiple predetermined human functional indicators; collecting measurement values ​​and actual performance values ​​of the target individual over multiple time periods to form an individual functional dataset; extracting high-order statistical values ​​from a homogeneous database that are in the top P percentile of a reference group as a top-tier benchmark; generating a first difference between the predicted performance value and the actual performance value, and a second difference between the individual measurement value and the top-tier benchmark; generating intervention target values ​​based on the first and second differences and encoding them into a structured training scheme; continuously injecting new periodic measurement values ​​generated after the real individual executes the training scheme into the dataset and triggering iterative updates, so that the native digital human evolves synchronously with the training results. This invention can construct a native digital human that continuously grows alongside a real individual, enabling lossless experiments in the virtual world and supporting team collaborative simulations.

Description

Technical Field

[0001] This invention relates to the technical fields of high-value skills training, such as sports training, rehabilitation medicine, health management, pilot training, fire and rescue training, and surgical skills enhancement. More specifically, this invention relates to a native digital human that continuously grows alongside a real individual and its construction method. Background Technology

[0002] In fields such as sports training, rehabilitation medicine, health management, fire and rescue, aviation, and high-risk operations, human function assessment and capability enhancement are the core foundation for developing personalized plans, judging intervention effects, and ensuring mission safety. With the development of digital twin technology, constructing individualized virtual human models has become possible, providing a new technological path for precise assessment and dynamic intervention. However, current human function assessment and digital twin technologies often suffer from static, isolated, and closed-loop deficiencies, making it difficult to meet the demands for precise, dynamic, and collaborative applications, and presenting many problems that urgently need to be addressed. For example: First, it cannot achieve continuous growth alongside the real individual. Current digital twin technologies mostly focus on static modeling or one-way simulation, with the core purpose of "reproduction" rather than "companionship." Once established, the model remains fixed and cannot evolve autonomously based on continuous input of individual data, failing to form a partnership of "co-growth" with the user. Every training result and every functional improvement of the individual cannot be synchronously absorbed by the digital model, causing the gap between the virtual model and the real individual to widen over time.

[0003] Second, it's impossible to conduct lossless testing in the virtual world. For critical vocational skills training such as pilot training, firefighter training, and surgical training, the cost of trial and error in the real world is extremely high, even life-threatening. Current digital twin technology lacks the ability to allow digital avatars to enter the virtual environment beforehand to verify solutions, simulate actions, and identify risks, failing to provide real individuals with the safety guarantee of "try before you act." Individuals can only directly bear risks in the real world; mission plans cannot be fully validated in the virtual environment.

[0004] Third, it cannot achieve collaborative team simulations. Existing technologies mostly focus on the digital modeling of single individuals, lacking the ability to create digital avatars for each member of a team and conduct group collaborative drills in a virtual environment. For scenarios requiring high levels of coordination, such as fire and rescue teams, sports teams, flight formations, and surgical teams, it is impossible to pre-rehearse the effectiveness of different personnel combinations, tactical plans, and collaboration modes in the virtual world. Improving team effectiveness still relies on repeated practice in reality, which is inefficient and risky.

[0005] Therefore, it is necessary to design a technical solution that can overcome the above-mentioned defects. Summary of the Invention

[0006] One objective of this invention is to provide a native digital human that grows alongside a real individual and a method for constructing such a human, enabling the human to grow alongside a real individual, take risks on behalf of the real individual, and provide collaborative deduction capabilities for the team.

[0007] To achieve these objectives and other advantages of the present invention, according to one aspect of the present invention, a native digital human that grows alongside a real individual is provided, comprising: a task association model storing mapping relationships between target activity tasks and multiple predetermined human function indicators; an individual function dataset storing human function indicator measurements of the target individual collected over multiple time periods, and corresponding actual performance values ​​of the target activity tasks; a homogeneous reference interface configured to extract high-order statistical values ​​from a reference group that shares at least the same occupation, gender, age, height, and weight as the target individual from an external homogeneous database, representing the top p-threshold as a benchmark; and a difference analyzer connected to the individual function dataset and the homogeneous reference interface, respectively, for... The system generates a first difference between the predicted and actual performance values ​​based on data from the individual functional dataset, and a second difference based on the individual measurements and the top benchmark. An intervention target generator, connected to the difference analyzer, generates intervention target values ​​for at least one human functional indicator based on the first and second differences, and encodes these intervention target values ​​into a structured training scheme. This structured training scheme instructs real individuals to perform corresponding training. An evolution engine, connected to the individual functional dataset, the difference analyzer, and the intervention target generator, continuously injects new periodic measurements generated after real individuals perform the structured training scheme into the individual functional dataset, and triggers iterative updates in the difference analyzer and the intervention target generator, enabling the native digital human to evolve synchronously with the training results of real individuals.

[0008] Furthermore, the predetermined human function indicators include at least physical, cognitive, and psychological dimensions; among them, the physical dimension includes at least strength, speed, endurance, flexibility, and agility indicators; the cognitive dimension includes at least attention, reaction time, decision-making accuracy, and spatial perception indicators; and the psychological dimension includes at least stress tolerance, emotional stability, self-confidence, and arousal level indicators. The mapping relationship stored in the task association model is the performance prediction value Y=Σ(Wi×fi(Xi))+C, where Xi is the current measurement value of the i-th human function indicator, fi is a pre-set nonlinear monotonic function for the human function indicator, Wi is the corresponding weight, and C is a constant; the corresponding weights are pre-set according to the target activity task through biomechanical analysis, cognitive task analysis, and psychological load analysis.

[0009] Furthermore, the corresponding weights are determined in advance through the following methods: For the target activity task, biomechanical data, cognitive task performance data, and psychophysiological response data of multiple individuals in the reference group are collected when performing the target activity task; measurement values ​​of multiple individuals in the reference group on each indicator of physical, cognitive, and psychological dimensions are collected; regression analysis is performed on the biomechanical data, cognitive task performance data, and psychophysiological response data with the measurement values ​​of each dimension to obtain the contribution of each human function indicator to the performance level of the target activity task; the contribution is normalized and used as the corresponding weight.

[0010] Furthermore, the evolution engine is also equipped with a weight update unit, which updates the corresponding weights according to the following rule: Wi(t+1)=Wi(t)+η×Δ(t)×(∂Y) actual / ∂Xi), where Wi(t) is the weight of the i-th indicator in the current period, η is the preset learning rate, Δ(t) is the first difference, and Y actual For actual performance values, ∂Y actual / ∂Xi represents the rate of change of the actual performance value relative to the measured value of the i-th indicator within the current period.

[0011] Furthermore, the actual performance values ​​in the individual functional dataset are provided by a task performance evaluation model. This model generates actual performance values ​​using at least one of the following evaluation modes based on the type of the target activity task: For activity tasks with standard action templates, video data is generated by recording the complete process of the target individual performing the target activity task using an image acquisition device. Frame-by-frame human skeletal keypoint detection is performed on the video data to generate a three-dimensional skeletal point temporal sequence of the target individual within the target activity task execution cycle. Standard three-dimensional skeletal point temporal sequences of a reference group performing the target activity task, at least of the same gender, age, height, and weight as the target individual, are extracted from a homogeneous database. The dynamic time warping distance between the three-dimensional skeletal point temporal sequence and the standard three-dimensional skeletal point temporal sequence is calculated. This dynamic time warping distance is normalized and mapped to a preset scoring interval, and the action score is output as the actual performance value. For tasks based on results... For target-oriented activities, objective performance indicators are collected from the individual during the execution of the target activity. These objective performance indicators include, but are not limited to, task completion time, accuracy, success rate, number of errors, reaction time, and decision-making accuracy. The collected objective performance indicators are compared with the performance indicator distribution of a reference group in a homogeneous database on the same task, normalized, mapped to a preset scoring range, and the performance score is output as the actual performance value. For high-cost, high-value, or difficult-to-repeat activities, the individual's native digital human is loaded into the virtual simulation environment of the target activity. The native digital human performs the target activity in the virtual simulation environment, and the simulation performance data of the digital human during the virtual execution is collected. The simulation performance data is normalized, mapped to a preset scoring range, and the simulation score is output as the actual performance value. The simulation score is used to replace the actual performance value of the real individual in reality, so as to achieve non-destructive testing in high-risk scenarios.

[0012] Furthermore, the intervention target generator is configured to: determine at least one indicator to be intervened from multiple predetermined human functional indicators based on the sign and absolute value of the first difference; obtain the indicator deviation value corresponding to the indicator to be intervened in the second difference; calculate the intervention target value of the indicator to be intervened based on the indicator deviation value and the corresponding value of the top benchmark on the indicator to be intervened; extract historical training data from a homogeneous database from a reference group that has the same occupation, gender, age, height, and weight as the target individual, for the indicator to be intervened from its current measurement value to the intervention target value, the historical training data including weekly training frequency, single training intensity range, and training duration in weeks; and encode the intervention target value, weekly training frequency, single training intensity range, and training duration in weeks into a structured training scheme.

[0013] Furthermore, the evolution engine is configured to: generate time-series variation curves of various human functional indicators based on the current measurement values ​​of the target individual over multiple consecutive periods; perform similarity matching between the time-series variation curves and the typical variation curves of the corresponding reference group in the homogeneous database; when the similarity exceeds a preset threshold, output the predicted value of the target individual's human functional indicators at a specified future time point based on the extrapolated value of the time-series variation curves; if the similarity is less than the preset threshold, mark the target individual as a developmental shift type, store the time-series variation curves in the homogeneous database to form a new reference subgroup, and simultaneously pause the output of the predicted value of the human functional indicators.

[0014] Furthermore, the evolution engine is configured to: construct native digital humans for multiple target individuals and conduct group collaborative simulations in a virtual environment; the group collaborative simulations include: generating multiple simulation schemes for preset team task scenarios, including at least one of personnel matching schemes, tactical schemes, and collaborative strategy schemes; loading each simulation scheme into the virtual environment, and having each native digital human execute simulation tasks according to the simulation scheme; the evolution engine collects group collaboration indicators during the execution of each simulation scheme, including at least one of team task completion efficiency, individual load distribution, key event success rate, and number of collaboration errors; the evolution engine compares the group collaboration indicators of each simulation scheme according to preset optimization rules, selects the optimal scheme, and outputs the team optimization scheme.

[0015] According to another aspect of the present invention, a method for constructing a native digital human that grows alongside a real individual is also provided, comprising: S1: establishing a mapping relationship between a target activity task and multiple predetermined human functional indicators; S2: collecting the human functional indicator measurements of the target individual and the corresponding actual performance values ​​of the target activity task at multiple time periods to form an individual functional dataset; S3: extracting high-order statistical values ​​from a reference group that has at least the same gender, age, height, and weight as the target individual from an external homogeneous database, and using these values ​​as a top-tier benchmark; S4: determining performance based on the mapping relationship and the current measurement values. S5: Generate a first difference value representing the deviation between the predicted value and the actual performance value; S6: Generate a second difference value based on the individual measurement value and the top benchmark; S7: Generate an intervention target value for at least one human function indicator based on the first and second differences, and encode the intervention target value into a structured training scheme, which is used to instruct real individuals to perform corresponding training; S8: Continuously inject the new periodic measurement values ​​generated after the real individuals perform the structured training scheme into the individual function dataset, and repeatedly generate the first difference value, the second difference value, and the intervention target value based on the updated individual function dataset, so that the native digital human evolves synchronously with the training results of real individuals.

[0016] The present invention has at least the following beneficial effects: This invention continuously collects data from real individuals and compares it with the "top-percentile" (P-threshold) of individuals with similar physiological conditions, generating a Δ difference value representing the gap between the individual and the top performers. Based on this difference value, the system generates personalized, structured training plans to guide real individuals in targeted training. New periodic measurements generated after training are continuously injected into the individual's functional dataset, triggering iterative updates to the difference analyzer and intervention target generator, ensuring that the model parameters of the native digital human evolve synchronously with the training results of the real individual. Thus, this invention constructs a complete closed loop of "data collection—difference analysis—plan generation—training execution—data reinjection," achieving "co-growth" and "real-time companionship" between the digital avatar and the real individual, solving the fundamental defects of existing digital twin technology models that are rigid and unable to evolve synchronously with the individual.

[0017] The native digital human constructed in this invention can replace a real individual for preliminary practice in a virtual environment. Taking competitive sports as an example, the digital human can try out new moves in a virtual environment, provide early warnings of injury risks, and simulate different tactics in a virtual arena. The risks and efficiency of each plan can be evaluated based on the digital human's execution results. This "digital avatar first, real person then executes" model significantly improves the safety and execution efficiency of high-risk professions (such as fire rescue, competitive sports, aviation, surgery, and high-risk operations), minimizing the real-world trial-and-error costs for critical professions such as pilots, firefighters, and surgeons.

[0018] This invention can be extended to team scenarios, creating an independent native digital avatar for each member of the team. Taking basketball as an example, digital avatars can be created for five athletes, forming a virtual team in a virtual environment to conduct batch drills on different personnel combinations and tactical plans. By analyzing the data from the virtual drills, the optimal personnel combinations and action plans can be identified, which real-world fire brigades can then use to organize training and competitions. Thus, this invention achieves a leap from individual digital avatars to team collaborative simulation capabilities, freeing team effectiveness from repeated real-world adjustments and significantly reducing the cycle and risks of collaborative training.

[0019] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0020] Figure 1 This is a flowchart of one embodiment of this application. Detailed Implementation

[0021] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0022] It should be understood that terms such as "having," "comprising," and "including" used in the embodiments of this application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of components in a specific posture. If the specific posture changes, the directional indication will also change accordingly. When an element is referred to as "fixed to" or "set on" another element, it can be directly on the other element or may have an intervening element present. When an element is referred to as "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element through an intervening element. Descriptions involving "first," "second," etc., in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0023] It should be noted that the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0024] In one embodiment, a native digital human that grows alongside a real individual can be used for physical fitness assessment and training optimization of firefighters, as well as for technical improvement and team collaboration simulations for competitive athletes. This native digital human includes a task association model, which stores mapping relationships between target activities and multiple predetermined human functional indicators. For example, it can quantitatively correlate a task of carrying equipment up stairs with indicators such as lower limb strength, cardiorespiratory endurance, and core stability; or it can quantitatively correlate a basketball three-point shooting task with indicators such as lower limb explosive power, upper limb strength, core stability, concentration, and stress tolerance. The mapping relationship can be implemented in various forms, including but not limited to preset function mappings, neural network models, rule engines, or combinations thereof. For example, a function based on expert knowledge can be used to describe the quantitative relationship between indicators and task performance; the mapping relationship can be implemented using a neural network, taking human functional indicator measurements as input and task performance predictions as output, constructing an end-to-end nonlinear mapping through a data-driven approach. Those skilled in the art can select or combine different mapping implementation methods according to task characteristics, data conditions, and application requirements; this invention does not limit this. The individual functional dataset stores human functional index measurements of the target individual collected over multiple time periods. Examples include bi-weekly measurements of maximum squat weight, maximum oxygen uptake, and plank duration, along with corresponding actual performance values ​​for target activities, such as the time to climb ten flights of stairs while carrying equipment, or weekly measurements of vertical jump height, grip strength, and three-point shooting percentage. The homogeneous reference interface is configured to extract high-order statistical values ​​from a reference group with the same occupation, gender, age, height, and weight as the target individual from an external homogeneous database. These values ​​represent the top P percentiles of the group and serve as the benchmark. For example, P can be set to 10, meaning data from the top 10% of performers within the same group is extracted. The difference analyzer connects to both the individual performance dataset and the homogeneous reference interface. It generates a first difference based on the data in the individual performance dataset, which represents the deviation between the predicted performance value and the actual performance value. For example, the difference between the predicted stair climbing time and the actual stair climbing time based on the task association model, or the difference between the predicted shooting percentage and the actual shooting percentage. At the same time, it generates a second difference based on the individual measurement value and the top benchmark, such as the difference between the individual's maximum weight for a weighted squat and the corresponding indicator in the top benchmark, or the difference between the individual's vertical jump height and the corresponding indicator in the top benchmark.The intervention target generator connects to the difference analyzer and generates intervention target values ​​for at least one human functional indicator based on the first difference and the second difference. The intervention target values ​​are then encoded into structured training programs. For example, when the first difference shows that the predicted stair climbing time is slower than the actual time and the second difference shows that the lower limb strength is insufficient, the lower limb strength is set as the intervention indicator, and an intervention target value is generated to increase the maximum weight of the weighted squat from the current 80 kg to 90 kg. This target value is then encoded into a squat training program of three times a week, eight to twelve times per set. Alternatively, when the first difference shows that the predicted shooting percentage is lower than the actual shooting percentage and the second difference shows that the lower limb explosive power is insufficient, the lower limb explosive power is set as the intervention indicator, and an intervention target value is generated to increase the vertical jump height from the current 60 cm to 65 cm. This target value is then encoded into an augmented training program of four times a week, six to eight times per set. The evolutionary engine connects to the individual functional dataset, the differential analyzer, and the intervention target generator. It continuously injects new periodic measurements generated after real individuals execute a structured training program into the individual functional dataset. For example, the maximum weight for a weighted squat and the time to climb stairs, measured again after two weeks, are input into the dataset; or the vertical jump height and shooting percentage, measured again after one week, are input. The evolutionary engine triggers the differential analyzer to recalculate the first and second differences, and triggers the intervention target generator to adjust the training program for the next period based on the updated data, ensuring that the model parameters of the native digital human evolve synchronously with the training results of the real individuals. In team collaborative simulation applications, independent native digital humans can be built for each member of the team. For example, digital avatars can be built for each member of a rescue team consisting of ten firefighters. These digital humans form a virtual fire brigade in a virtual environment, conducting batch drills on different fire scenarios, personnel combinations, and tactical plans. The system analyzes the virtual drill data to find the optimal personnel combination and action plan, and feeds the optimal plan back to the real fire brigade. Simultaneously, each member's digital human continues to evolve synchronously with the training results of their respective real individuals according to the aforementioned mechanism.

[0025] In existing technologies, human function assessment systems typically employ a one-time test, comparing individual indicators with the average of a general population to provide a static assessment conclusion. This approach fails to dynamically adjust assessment parameters and intervention directions based on changes in individual performance, and it cannot achieve collaborative simulations at the team level. This embodiment, by constructing a digital human system comprising a task association model, individual function datasets, a homogeneous reference interface, a difference analyzer, an intervention target generator, and an evolutionary engine, achieves continuous accumulation and iterative analysis of multi-period data. It can dynamically generate and update training schemes based on the deviation between individual performance and predicted values, as well as the deviation from the top benchmark of a homogeneous group. This allows the native digital human to continuously evolve with the accumulation of real individual training results. Furthermore, by constructing digital avatars for team members and conducting collaborative simulations in a virtual environment, it achieves a leap in capabilities from individual to team level, overcoming the shortcomings of traditional assessment systems that are static, singular, lack feedback, and cannot perform team collaborative simulations.

[0026] In one embodiment, the difference signal output by the evolution engine, in addition to being used to update the parameters of the native digital human, can also be connected to a multimodal feedback interface as a raw driving signal. This interface can convert the difference into auditory, visual, or tactile guidance signals according to preset encoding rules. For example, it can map the difference in the physical dimension to pitch changes and the difference in the cognitive dimension to rhythm density, generating real-time auditory feedback so that the digital human can convey information to real individuals in a more intuitive way.

[0027] In one embodiment, multiple predetermined human functional indicators may include physical, cognitive, and psychological dimensions. Physical dimensions may include strength indicators such as grip strength or maximum weight in a weighted squat, speed indicators such as 10-meter sprint time, endurance indicators such as 12-minute running distance, flexibility indicators such as sit-and-reach distance, and agility indicators such as T-run completion time. Cognitive dimensions may include attention indicators such as reaction time in a sustained attention test, reaction time indicators such as milliseconds in a simple reaction time test, decision accuracy indicators such as the correct judgment rate in a simulated task, and spatial perception indicators such as the error value in a 3D spatial positioning test. Psychological dimensions may include stress tolerance indicators such as the magnitude of heart rate variability before and after a stressful task, emotional stability indicators such as scores on a self-rating emotion scale, confidence indicators such as scores on a self-efficacy questionnaire, and arousal level indicators such as skin conductance levels. In addition to physical, cognitive, and psychological dimensions, the human functional indicators may also include extended dimensions as needed for the application scenario, including but not limited to occupation type, task level, circadian rhythm, environmental factors, and personal preferences, to further enhance the personalization of the native digital human. The mapping relationship stored in the task association model can be represented as Y=Σ(Wi×fi(Xi))+C, where Xi is the current measured value of the i-th individual's human functional index, fi is a pre-defined nonlinear monotonic function for the human functional index, Wi is the corresponding weight, and C is a constant. For example, the strength index can use a logarithmic function to reflect the law that strength increases significantly in the early stages of task performance but decreases marginally in the later stages. The speed index can use a power function to characterize the nonlinear relationship between speed improvement and task performance. The weights Wi are pre-defined based on the target activity task through biomechanical analysis, cognitive task analysis, and psychological load analysis. For example, for fire rescue tasks, by analyzing the force contribution of each muscle group in the stair climbing action, the cognitive decision-making needs during task execution, and the impact of psychological pressure on performance, the weight of lower limb strength can be set to 0.3, the weight of decision accuracy to 0.25, and the weight of stress tolerance to 0.2, with the weights of other indicators allocated accordingly. The constant C can be an empirical correction value, such as 0 or 5, used to make the predicted value closer to the actual measurement range.

[0028] In existing technologies, human function assessments often focus solely on physical fitness indicators, neglecting the impact of cognitive abilities and psychological states on task performance. This leads to significant discrepancies between assessment results and actual task performance. This embodiment incorporates cognitive and psychological dimensions into the human function indicator system, sets nonlinear monotonic functions for different dimensions, and scientifically assigns weights based on biomechanical analysis, cognitive task analysis, and psychological load analysis. This allows performance predictions to more comprehensively reflect an individual's combined physical, cognitive, and psychological abilities, thereby improving prediction accuracy.

[0029] In one optional embodiment, the task association model is implemented using a neural network. Specifically, human functional index measurements (including physical, cognitive, and psychological indicators) collected over multiple time periods are used as input feature vectors, and the corresponding actual performance values ​​are used as training labels to train a multi-layer neural network (such as a deep neural network, convolutional neural network, or recurrent neural network) to learn the end-to-end mapping relationship from functional indicators to performance prediction values. After training, the neural network can directly output performance prediction values ​​without pre-setting function forms and weights. Those skilled in the art can choose a suitable network architecture based on the data scale and task complexity; this invention does not limit this choice.

[0030] In one embodiment, the corresponding weights of the mapping relationship can be predetermined in the following way. For a target activity task, such as carrying equipment up stairs in fire rescue, biomechanical data, cognitive task performance data, and psychophysiological response data of multiple individuals in the reference group are collected when performing the task. For example, a 3D motion capture system and force table are used to record the joint angles and ground reaction forces during the stair climbing process; an eye tracker is used to record the fixation point distribution and saccade path to assess attention allocation; and a heart rate monitor and skin conductance sensor are used to record heart rate variability and skin conductance levels during the task to assess psychological load. At the same time, measurements of these individuals in the reference group on various indicators of physical, cognitive, and psychological dimensions are collected, such as maximum weight for weighted squats, 10-meter sprint time, reaction time of sustained attention test, simple reaction time, and scores of self-rating emotion scales. Regression analysis is performed on biomechanical data, cognitive task performance data, and psychophysiological response data with measurements from various dimensions. For example, multiple linear regression or partial least squares regression can be used to obtain the contribution of each human functional indicator to the performance level of the target activity task. For instance, the regression coefficient for lower limb strength on stair climbing time is -0.35, the regression coefficient for decision accuracy is -0.28, and the regression coefficient for stress tolerance is -0.20, indicating that higher values ​​for these indicators result in shorter task completion times. The contribution is normalized and used as the corresponding weights. That is, the contribution of each indicator is divided by the sum of the contributions of all indicators to make the sum of the weights equal to 1. For example, after normalization, the weight of lower limb strength can be set to 0.35 divided by the total contribution sum to get 0.4, the weight of decision accuracy can be set to 0.28 divided by the total contribution sum to get 0.32, and the weight of stress tolerance can be set to 0.20 divided by the total contribution sum to get 0.23.

[0031] In existing technologies, the weighting of mapping relationships often relies on subjective experience or simple statistics, lacking in-depth analysis of the multi-dimensional influencing factors of physical, cognitive, and psychological aspects in specific tasks, leading to a disconnect between weights and actual task requirements. This embodiment collects biomechanical data, cognitive performance data, and psychophysiological response data of a reference group during task execution, and performs regression analysis with multi-dimensional indicator measurements. This transforms objective contributions into weights, enabling the weights to truly reflect the actual impact of each dimension indicator on task performance, thereby improving the scientific rigor and relevance of the mapping relationship.

[0032] In one embodiment, the evolutionary engine is further configured with a weight update unit, which updates the corresponding weights according to the following rule: Wi(t+1)=Wi(t)+η×Δ(t)×(∂Y) actual / ∂Xi), where Wi(t) is the weight of the i-th indicator in the current period, η is the preset learning rate, Δ(t) is the first difference, and Y actual For actual performance values, ∂Y actual / ∂Xi represents the rate of change of the actual performance value relative to the measured value of the i-th indicator within the current period. The learning rate can be 0.01 or 0.02, used to control the step size of each weight adjustment. The first difference is the difference between the predicted and actual performance values ​​for the current period, such as the predicted stair-climbing time minus the actual stair-climbing time. The rate of change can be approximated by recording the changes in actual performance after fine-tuning of individual functional indicators within the current period, for example, by comparing the differences between two tests, or by obtaining the slope through linear regression of indicator measurements and actual performance values ​​over a period of time. For example, in fire rescue training, if the weight of lower limb strength in the current cycle is 0.4, the first difference of -0.2 indicates that the predicted climbing time is faster than the actual climbing time, i.e., the prediction is too high. The rate of change of 0.5 indicates that for every unit increase in lower limb strength, the actual climbing time decreases by 0.5 seconds. The learning rate is 0.01. Then the updated weight is 0.4 plus 0.01 multiplied by -0.2 and then multiplied by 0.5, which equals 0.399. The weight is slightly reduced, indicating that the contribution of this indicator to the prediction being too high has been adjusted.

[0033] It should be noted that the weights predetermined through regression analysis are used as the initial weights for the mapping relationship in the task association model, and are used for performance prediction in the early stages of building the native digital human. As individual training data continues to accumulate, the weight update unit configured in the evolution engine will dynamically optimize the weights based on the difference between the individual's actual performance and the predicted value, so that the mapping relationship gradually adapts to the personalized characteristics of the target individual.

[0034] In existing technologies, the weights of the mapping relationship are fixed once determined, and cannot be adaptively adjusted according to changes in individual performance. As the training process progresses, the relationship between indicators and performance may change, and fixed weights will lead to a gradual increase in prediction bias. This embodiment introduces a weight update mechanism, which enables the weights to be dynamically optimized based on the difference between the individual's actual performance and the prediction, as well as the actual impact of indicator changes on performance. This allows the native digital human to continuously learn individual characteristics, improving the long-term accuracy of predictions.

[0035] In one embodiment, the actual performance values ​​in an individual's functional dataset can be provided by a task performance evaluation model. This model generates actual performance values ​​based on the type of the target activity task using an appropriate evaluation mode. For activity tasks with standard action templates, such as gymnastic movements or basketball shooting, video data can be generated by recording the complete process of the target individual performing the task using image acquisition equipment. For example, two high-speed cameras can be used to capture the video data from the front and side. Frame-by-frame human skeletal keypoint detection is performed on the video data. A pose estimation algorithm, such as OpenPose, is used to extract the three-dimensional coordinates of keypoints such as shoulders, elbows, wrists, hips, knees, and ankles in each frame, generating a three-dimensional skeletal point temporal sequence within the task execution cycle. Standard three-dimensional skeletal point temporal sequences of reference groups with the same occupation, gender, age, height, and weight as the target individual performing the same task are extracted from a homogeneous database. The dynamic time warping distance between the target individual's temporal sequence and the standard temporal sequence is calculated. This distance quantifies the similarity between the two on the time axis; the smaller the distance, the closer the movement is to the standard. The dynamic time warping distance is normalized and mapped to a preset scoring range, such as 0 to 100 points, and the action score is output as the actual performance value. For outcome-oriented tasks such as demolition operations in fire rescue, objective performance indicators can be collected on the target individual during task execution, such as task completion time, demolition accuracy, number of errors, and decision-making accuracy. These objective performance indicators are then compared with the performance indicator distribution of a reference group in a homogeneous database on the same task. After normalization, the data is mapped to a preset scoring range, and the performance score is output as the actual performance value. For high-cost, high-value, or difficult-to-repeat critical tasks (such as pilot emergency response, fire rescue, and complex surgical procedures), i.e., tasks with high repetition costs or difficulty in repetition, such as real fire rescue, the target individual's native digital human can be loaded into the virtual simulation environment of the target task. The native digital human performs the task in the virtual simulation environment, and simulation performance data is collected during the virtual execution process, such as the movement path in the virtual fire, physical exertion, simulated psychological load, and task completion time. This simulation performance data is normalized and mapped to a preset scoring range, and the simulation score is output as the actual performance value. This simulation score is used to replace the actual performance value of the real individual in reality, thereby achieving non-destructive testing in high-risk scenarios.

[0036] In existing technologies, obtaining actual performance values ​​often relies on manual scoring or simple result measurements such as hit rate or completion time, which cannot accurately assess the performance quality of different task types and lacks safe and effective assessment methods for high-risk scenarios. This embodiment uses a task performance evaluation model, employing action quality evaluation, task performance evaluation, or virtual simulation evaluation modes for different task types. This model can comprehensively, objectively, and safely reflect the quality and effectiveness of an individual's task completion, providing accurate actual performance values ​​for difference analysis. In particular, the virtual simulation evaluation mode makes non-destructive testing possible in high-risk scenarios.

[0037] In one alternative embodiment, the native digital human can be constructed using a temporal predictive neural network. For example, a Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Transformer model, or a combination thereof can be trained using a sequence of historical performance vectors (including physical, cognitive, and psychological dimensions) of the target individual over multiple time periods as input, and the predicted performance vector or optimization suggestions for the next period as output. This neural network model undergoes incremental learning or fine-tuning by continuously injecting new data, achieving synchronous evolution with the real individual. Those skilled in the art can choose a suitable network architecture based on the data sequence length and prediction accuracy requirements; this invention does not limit this choice.

[0038] In one optional embodiment, the task performance evaluation model can employ a neural network to perform end-to-end motion quality scoring on video data. Specifically, for activity tasks with standard motion templates, the video frame sequence of the target individual performing the task is directly input into a 3D convolutional neural network (C3D), a video transformer, or a two-stream network. The network outputs a motion quality score as the actual performance value, without the need for explicit skeletal keypoint detection and dynamic time warping distance calculation. This approach is particularly suitable for task scenarios with complex motion patterns that are difficult to accurately describe using traditional feature engineering. Those skilled in the art can choose a suitable network architecture based on task characteristics and computational resources; this invention does not limit this choice.

[0039] In one embodiment, the intervention target generator can also be configured to perform the following operations: First, determine at least one indicator to be intervened from a plurality of predetermined human functional indicators based on the sign and absolute value of a first difference. For example, if the first difference is positive and the absolute value is greater than 3, it indicates that the predicted performance is better than the actual performance, i.e., the actual performance is lower than expected, and it is necessary to identify the weak indicator that caused the poor performance. If the first difference is negative and the absolute value is greater than 3, it indicates that the predicted performance is lower than the actual performance, which may be due to the underestimation of individual indicators, but usually priority is given to cases where the actual performance is lower than expected. Then, obtain the indicator deviation value corresponding to the indicator to be intervened in the second difference. For example, if the second difference of lower limb strength is -10 kg, it means that the individual's maximum weight for a weighted squat is lower than the corresponding value in the top benchmark. Next, calculate the intervention target value based on the indicator deviation value and the corresponding value of the top benchmark on the indicator to be intervened. For example, if the lower limb strength in the top benchmark is 100 kg and the deviation is -10 kg, the intervention target value can be set to 95 kg, i.e., slightly higher than the current value or directly targeting the top benchmark. From a homogeneous database, historical training data is extracted from a reference group that shares the same occupation, gender, age, height, and weight as the target individual. This data corresponds to the improvement of the intervention indicator from the current measurement value to the intervention target value. This data can include weekly training frequency, single training intensity range, and training duration in weeks. For example, in the reference group, improving from 80 kg to 95 kg typically involves squat training three times a week at 70% to 81% of maximum repetitions, for six weeks. Finally, the intervention target value, weekly training frequency, single training intensity range, and training duration are encoded into a structured training plan, such as generating an executable plan that includes specific training movements, sets, repetitions, and rest intervals.

[0040] In existing technologies, intervention goals are often set based on general recommendations or coaching experience, lacking individualized quantitative basis, and there is no clear correlation between training plans and goals. This embodiment combines the first and second differences to comprehensively determine the intervention indicators and intervention target values ​​from two dimensions: individual performance deviation and group apex deviation. It also uses historical training data from a homogeneous group to generate a structured training plan that matches the goals, making the intervention plan more targeted and operable.

[0041] In one embodiment, the evolutionary engine can also be configured to generate time-series curves of various human function indicators based on the target individual's current measurements over multiple consecutive periods. For example, it can connect indicators such as maximum squat weight, maximum oxygen uptake, and reaction time measured weekly over the past eight weeks into separate curves. Then, this time-series curve is matched for similarity with typical curves of a corresponding reference group in a homogeneous database, for example, using dynamic time warping algorithms or Pearson correlation coefficients to calculate the similarity. When the similarity exceeds a preset threshold, such as 0.8 or 0.85, it indicates that the individual's development trend is consistent with the homogeneous group. Based on the extrapolated values ​​of the time-series curves, predicted values ​​of the target individual's human function indicators at a specified future time point can be output, for example, using linear regression or exponential smoothing methods to predict the maximum squat weight three months later. If the similarity is less than a preset threshold, the target individual is marked as a developmental deviation type, indicating that its change trajectory is different from that of the general population and may be affected by injury, special training programs or other factors. The time series change curve of this individual is stored in a homogeneous database to form a new reference subgroup, providing a more accurate reference for subsequent individuals with similar change trajectories. At the same time, the output of human function index prediction values ​​is suspended to avoid inaccurate predictions interfering with decision-making.

[0042] In existing technologies, digital twin systems typically assess a person's condition based solely on data from the current moment, lacking the ability to analyze long-term trends and identify individuals with developmental deviations. This embodiment, through time series analysis, not only predicts an individual's future functional level but also identifies individuals with developmental deviations through similarity matching, supplementing their data into a new reference subgroup within the database. This enriches the diversity of the homogeneous database and provides a more accurate reference for the subsequent assessment of similar individuals.

[0043] In one embodiment, the evolutionary engine can construct corresponding native digital humans for multiple target individuals and perform group collaborative simulations in a virtual environment. Specifically, when an optimization plan needs to be developed for a team task scenario, the evolutionary engine first obtains the human functional index measurements, historical performance data, and current model parameters of each real individual in the team, and constructs a corresponding native digital human for each individual based on this data. These native digital humans retain the same physical, cognitive, and psychological dimensions as their real individuals and can independently perform tasks in the virtual environment. The evolutionary engine then generates multiple simulation plans for preset team task scenarios, such as high-rise building search and rescue missions in fire rescue or offensive and defensive tactical missions in basketball games. The simulation plans may include personnel allocation plans, such as specifying which members form the advance team and which members form the support team; they may include tactical plans, such as route selection in fire rescue or defensive formation selection in basketball games; and they may also include collaborative strategy plans, such as communication frequency, task handover timing, or offensive and defensive transition rhythm. In one possible implementation, the number of simulation plans can be set to five or ten depending on computing resources. The evolutionary engine loads these simulation scenarios into a virtual environment, which can be built based on a physics engine to simulate terrain, obstacles, time pressure, and other conditions in the task scenario. Once each scenario is loaded, the corresponding native digital human executes the simulation task according to the role assignments and action instructions set in the scenario. For example, in a virtual fire, they might move along a designated route and collaboratively complete a breaching operation, or on a virtual sports field, they might execute an offensive maneuver according to specified tactics. During the simulation task execution, the evolutionary engine records the execution data of each native digital human through the data acquisition interface built into the virtual environment. This data is processed into group collaboration indicators, which may include team task completion efficiency, such as the total time from task start to goal achievement; individual workload distribution, such as the dispersion of each member's physical exertion or cognitive load during the task; key event success rate, such as the proportion of specified actions completed within a specific time window; and the number of collaboration errors, such as communication delays, uncoordinated actions, or overlapping tasks. The evolutionary engine compares these group collaboration indicators according to preset optimization rules. These rules can be set based on task objectives; for example, in fire rescue scenarios, priority is given to the solution with the shortest task completion time and the most balanced distribution of individual workloads, while in competitive sports scenarios, priority is given to the solution with the highest success rate for critical events. In one possible implementation, the optimization rules can use a weighted summation method, normalizing multiple indicators, multiplying them by their respective weights, and then summing them. The weight values ​​can be set to 0.3 or 0.5 based on task priority.The evolution engine comprehensively compares the metrics of various simulation scenarios, selects the optimal solution that meets the optimization rules, and outputs it as a team optimization plan. This plan may include suggestions for adjusting specific personnel combinations, role assignments, action sequences, or collaboration strategies. This team optimization plan can be used to guide real teams in personnel allocation and tactical deployment during actual tasks.

[0044] In existing technologies, team collaborative training typically relies on repeated drills in real-world scenarios. Coaches or commanders develop personnel pairings and tactical plans based on experience, which are then executed by real personnel in a real environment to observe the effects. Each adjustment to the plan requires reorganizing personnel, venues, and equipment, resulting in high trial-and-error costs and low efficiency. In this embodiment, the evolutionary engine constructs native digital humans for each team member, generating multiple simulation plans in batches within a virtual environment and simulating their execution. This allows for parallel verification and quantitative comparison of different personnel pairings, tactical choices, and collaborative strategies without involving real personnel, venues, or risks. Simulation tasks in the virtual environment are not limited by factors such as venue, time, or security, allowing for multiple repetitions and rapid adjustments. The collected group collaboration indicators objectively reflect the actual effectiveness of each plan. The evolutionary engine compares the indicators based on preset optimization rules and outputs the optimal plan, enabling the team to obtain validated optimized plans before entering real-world task scenarios. This significantly reduces the cycle and risk of collaborative training, achieving a shift from experience-driven to data-driven team capability improvement.

[0045] In one embodiment, such as Figure 1As shown, the method for constructing a native digital human that grows alongside a real individual includes the following steps: Step 1: Establish a mapping relationship between the target activity task and multiple predetermined human functional indicators. For example, mapping the task of carrying equipment up stairs in fire rescue to indicators such as lower limb strength, cardiorespiratory endurance, core stability, decision-making accuracy, and stress tolerance. Step 2: Collect the target individual's human functional indicator measurements and corresponding actual performance values ​​for the target activity task over multiple time periods to form an individual functional dataset. For example, every two weeks, collect measurements of maximum weight for weighted squats, maximum oxygen uptake, reaction time in sustained attention tests, and scores on self-rating emotion scales, and record the corresponding stair-climbing completion time or virtual simulation score. Step 3: Extract high-order statistical values ​​from a reference group with the same occupation, gender, age, height, and weight as the target individual from an external homogeneous database. These values ​​fall within the top P percentile as a benchmark, where P can be 5 or 10. Step 4: Determine the predicted performance value based on the mapping relationship and current measurements, and generate a first difference value representing the deviation between the predicted and actual performance values. For example, calculate the difference between the predicted stair-climbing time and the actual stair-climbing time. The fifth step generates a second difference based on the individual's measurements and the top benchmark, for example, calculating the difference between the individual's maximum weight for a weighted squat and the corresponding indicator in the top benchmark. The sixth step generates intervention target values ​​for at least one human function indicator based on the first and second differences, and encodes these intervention target values ​​into a structured training program. This structured training program instructs real individuals to perform corresponding training. For example, when the first difference shows the predicted speed is faster than the actual speed and the second difference shows insufficient lower limb strength, a lower limb strength improvement target is set, and a corresponding squat training plan is generated. The seventh step continuously injects the new periodic measurements generated after the real individual performs the structured training program into the individual function dataset, and repeatedly generates the first difference, second difference, and intervention target values ​​based on the updated individual function dataset. This allows the native digital human to evolve synchronously with the training results of real individuals. For example, after two weeks, newly measured data and actual performance values ​​are input into the system, and the differences and intervention targets are recalculated, forming a closed loop of continuous optimization.

[0046] In existing technologies, the construction of digital twins is mostly a one-time static modeling process, lacking a mechanism for continuous evolution and unable to automatically optimize as individual data accumulates. This embodiment, through multi-period data collection, comparison with a top-tier benchmark of a homogeneous group, generation of differences, setting intervention goals, and iterative iteration, enables the native digital human to possess the ability to continuously evolve. It can continuously adjust model parameters and intervention strategies based on individual performance and group references, achieving closed-loop management of human function assessment and intervention.

[0047] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A native digital human that grows alongside a real individual, characterized by: include: The task association model stores the mapping relationship between the target activity task and multiple predetermined human functional indicators; Individual functional datasets are used to store human functional index measurements of target individuals collected over multiple time periods, as well as the corresponding actual performance values ​​of target activity tasks; The homogeneous reference interface is configured to extract high-order statistical values ​​from a reference group that has at least the same occupation, gender, age, height, and weight as the target individual from an external homogeneous database, and these values ​​are in the top P percentiles, serving as the top-tier benchmark. The difference analyzer connects to the individual functional dataset and the homogeneous reference interface, respectively, and is used to generate a first difference between the predicted and actual performance values ​​based on the data in the individual functional dataset, and a second difference between the individual measurements and the top benchmark. An intervention target generator, connected to a difference analyzer, is used to generate intervention target values ​​for at least one human functional index based on a first difference and a second difference, and to encode the intervention target values ​​into a structured training program, which is used to instruct real individuals to perform corresponding training. The evolution engine connects to the individual functional dataset, the differential analyzer, and the intervention target generator, respectively. It continuously injects new periodic measurement values ​​generated by real individuals after performing structured training programs into the individual functional dataset and triggers the differential analyzer and intervention target generator to perform iterative updates, so that the native digital human evolves in sync with the training results of real individuals.

2. The native digital human that continuously grows alongside a real individual as described in claim 1, characterized in that, Multiple predetermined human functional indicators include at least physical, cognitive, and psychological dimensions; Among them, the physical fitness dimension includes at least strength, speed, endurance, flexibility, and agility indicators; the cognitive dimension includes at least attention, reaction time, decision-making accuracy, and spatial perception indicators; and the psychological dimension includes at least stress tolerance, emotional stability, self-confidence, and arousal level indicators. The mapping relationship stored in the task association model is the performance prediction value Y=Σ(Wi×fi(Xi))+C, where Xi is the current measurement value of the i-th human functional index, fi is a nonlinear monotonic function pre-set for the human functional index, Wi is the corresponding weight, and C is a constant; the corresponding weight is pre-set according to the target activity task through biomechanical analysis, cognitive task analysis and psychological load analysis.

3. The native digital human that continuously grows alongside a real individual as described in claim 2, characterized in that, The corresponding weights are determined in advance in the following way: For the target activity task, we collected biomechanical data, cognitive task performance data, and psychophysiological response data from multiple individuals in the reference group when they performed the target activity task. The study collected measurements of multiple individuals in the reference group across various indicators in the physical, cognitive, and psychological dimensions. Regression analysis was performed on exercise biomechanics data, cognitive task performance data, psychophysiological response data and measurement values ​​of each dimension to obtain the contribution of each human function index to the performance level of the target activity task. The contribution level is normalized and then used as the corresponding weight.

4. The native digital human that continuously grows alongside a real individual as described in claim 2, characterized in that, The evolution engine is also equipped with a weight update unit, which updates the corresponding weights according to the following rules: Wi(t+1)=Wi(t)+η×Δ(t)×(∂Y actual / ∂Xi), where Wi(t) is the weight of the i-th indicator in the current period, η is the preset learning rate, Δ(t) is the first difference, and Y actual For actual performance values, ∂Y actual / ∂Xi represents the rate of change of the actual performance value relative to the measured value of the i-th indicator within the current period.

5. The native digital human that continuously grows alongside a real individual as described in claim 1, characterized in that, The actual performance values ​​in the individual functional dataset are provided by the task performance evaluation model, which generates actual performance values ​​using at least one of the following evaluation modes, based on the type of the target activity task: For activities with standard action templates, video data is generated by recording the complete process of the target individual performing the target activity task through image acquisition equipment. The video data is then used to perform frame-by-frame human skeleton key point detection to generate a three-dimensional skeleton point time sequence of the target individual within the execution cycle of the target activity task. Standard three-dimensional skeleton point time sequence of the target activity task is extracted from a homogeneous database from a reference group that has at least the same occupation, gender, age, height, and weight as the target individual. The dynamic time warping distance between the three-dimensional skeleton point time sequence and the standard three-dimensional skeleton point time sequence is calculated. The dynamic time warping distance is then normalized and mapped to a preset scoring interval, and the action score is output as the actual performance value. For outcome-oriented tasks, objective performance indicators are collected from the individual performing the task. These objective performance indicators include, but are not limited to, task completion time, completion accuracy, success rate, number of errors, reaction time, and decision accuracy. The collected objective performance indicators are compared with the performance indicator distribution of a reference group in a homogeneous database on the same task. After normalization and mapping to a preset scoring range, the performance score is output as the actual performance value. For critical activities and tasks that are high-cost, high-value, or difficult to repeat, a native digital human of the target individual is loaded into a virtual simulation environment of the target activity and task. The native digital human performs the target activity and task in the virtual simulation environment, and simulation performance data of the digital human during the virtual execution process is collected. The simulation performance data is normalized and mapped to a preset scoring range, and the simulation score is output as the actual performance value. The simulation score is used to replace the actual performance value of the real individual in reality, so as to realize non-destructive testing in high-risk scenarios.

6. The native digital human that continuously grows alongside a real individual as described in claim 1, characterized in that, The intervention target generator is also configured as follows: Based on the sign and absolute value of the first difference, at least one indicator to be intervened is determined from multiple predetermined human functional indicators. Obtain the indicator deviation value corresponding to the indicator to be intervened in the second difference; Based on the indicator deviation value and the corresponding value of the top benchmark on the indicator to be intervened, the intervention target value of the indicator to be intervened is calculated. From a homogeneous database, historical training data is extracted from a reference group that has the same occupation, gender, age, height, and weight as the target individual. This data is used to improve the current measurement value of the indicator to be intervened to the intervention target value. The historical training data includes weekly training frequency, single training intensity range, and training duration in weeks. The intervention target value, weekly training frequency, single training intensity range, and training duration in weeks are encoded into a structured training scheme.

7. The native digital human that continuously grows alongside a real individual as described in claim 1, characterized in that, The evolution engine is also configured as follows: Generate time-series change curves for various human functional indicators based on the current measurement values ​​of the target individual over multiple consecutive periods; The time series change curves are matched with the typical change curves of the corresponding reference group in a homogeneous database based on similarity. When the similarity exceeds a preset threshold, the extrapolated value of the time series change curve is used to output the predicted value of the human function index of the target individual at a specified time point in the future. If the similarity is less than the preset threshold, the target individual is marked as a developmental shift type, and the time series change curve is stored in a homogeneous database to form a new reference subgroup. At the same time, the output of human function index prediction values ​​is paused.

8. The native digital human that continuously grows alongside a real individual as described in claim 1, characterized in that, The evolution engine is also configured to: construct native digital humans for multiple target individuals and conduct group collaborative simulations in a virtual environment; Group collaborative simulation includes: generating multiple simulation schemes for a preset team task scenario, including at least one of personnel matching schemes, tactical schemes, and collaborative strategy schemes; loading each simulation scheme into the virtual environment, and having each native digital human execute the simulation task according to the simulation scheme; The evolution engine collects group collaboration metrics during the execution of each simulation scheme. These group collaboration metrics include at least one of the following: team task completion efficiency, individual workload distribution, success rate of critical events, and number of collaboration errors. The evolution engine compares the group synergy indicators of each simulation scheme according to the preset optimization rules, selects the optimal scheme, and outputs the team optimization scheme.

9. A method for constructing a native digital human that continuously grows alongside a real individual, characterized in that, include: S1: Establish a mapping relationship between the target activity task and multiple predetermined human functional indicators; S2: Collect the measurement values ​​of human functional indicators of the target individual at multiple time periods and the corresponding actual performance values ​​of the target activity tasks to form an individual functional dataset; S3: Extract high-order statistical values ​​from a reference group that has at least the same occupation, gender, age, height, and weight as the target individual from an external homogeneous database, and use these values ​​as the top-percentile benchmark. S4: Determine the performance prediction value based on the mapping relationship and the current measurement value, and generate the first difference value that characterizes the deviation between the performance prediction value and the actual performance value; S5: Generate a second difference based on individual measurements and the top benchmark; S6: Generate intervention target values ​​for at least one human functional indicator based on the first difference and the second difference, and encode the intervention target values ​​into a structured training scheme, which is used to instruct real individuals to perform corresponding training; S7: The new periodic measurement values ​​generated after the real individual performs the structured training program are continuously injected into the individual functional dataset, and the first difference, second difference and intervention target value are repeatedly generated based on the updated individual functional dataset, so that the native digital human evolves in sync with the training results of the real individual.

Images