Training method based on simulated driving and simulated driving system

Abstract

The invention provides a training method based on simulated driving and a simulated driving system, and relates to the field of simulated driving. And when the terrain picture of the specific type is displayed on the virtual reality equipment, the control operation on the driving control piece is responded. The visual guidance can be provided for the training object, so that the control decision has a clear environment basis. Then, determining vehicle state data at the next moment according to physical characteristics of the terrain, control input and a vehicle dynamics model, and dividing the vehicle state data into two types of perception data, namely data directly perceived by a body and data perceived by vision or space; in this way, the somatosensory device can directly stimulate muscle memory to solidify the correct operation mode, and visual adjustment can deepen the spatial relationship cognition of the environment and the vehicle motion. The process can not only solve the problem of inducing dizziness, but also feed understandable biomechanical signals and spatial visual information back to the training object, so that the driving experience and the training effect of the training object in complex terrain training are improved.

Description

Technical Field

[0001] The present application relates to the field of simulated driving, and more specifically, to a training method and a simulated driving system based on simulated driving in the field of simulated driving. Background Art

[0002] With the improvement of people's living standards, vehicles have become an indispensable means of transportation in people's daily lives, and good driving skills have also become essential skills.

[0003] Currently, traditional on-site training methods face challenges such as low safety and weather and geographic location restrictions. With the maturity of virtual reality (VR) technology and the widespread adoption of vehicle electronic systems, it has become possible to develop a driving simulation system that combines VR equipment with existing vehicle control signal inputs, thereby improving the driving experience and training effectiveness of trainees. Summary of the Invention

[0004] The present application provides a training method and a simulation driving system based on simulated driving, which can improve the driving experience and training effect of the training subject.

[0005] In a first aspect, a training method based on simulated driving is provided, which is applied to a simulated driving system, the simulated driving system including a simulated cockpit and a virtual reality device, the simulated cockpit including driving controls and a seat, the method comprising: in a case where a terrain image is displayed on the virtual reality device, in response to a training subject's control operation of the driving controls on the driving controls, determining first vehicle state data at a next moment based on the terrain type, the control operation, and a preset vehicle dynamics model, the terrain image being used to present a terrain area corresponding to the terrain type in which the vehicle can safely travel; dividing the first vehicle state data into first-category data and second-category data, the first-category data being vehicle state data that can be directly perceived by the body, and the second-category data being vehicle state data that can be perceived visually or spatially; controlling the current states of the driving controls and the seat based on the first-category data; and adjusting the terrain image displayed on the virtual reality device based on the second-category data.

[0006] In the above-mentioned technical solution, when a specific type of terrain suitable for safe vehicle navigation is displayed on a virtual reality device, the training subject's control of the driving controls is responded to. This provides the training subject with intuitive visual guidance of the driving path and obstacle distribution, enabling the training subject to make driving decisions with a clear environmental basis. Subsequently, the vehicle's state data for the next moment is determined in real time based on the terrain's physical characteristics, control input, and the vehicle dynamics model. This vehicle state data is divided into two types of sensory data: data that can be directly perceived by the body and data that can be perceived visually or spatially. In this way, the somatosensory device directly stimulates muscle memory to solidify the correct operating mode, while visual adjustments deepen the understanding of the spatial relationship between the environment and vehicle movement. This process avoids sensory conflict, addressing the potential for nausea and vomiting, while converting abstract vehicle dynamics into biomechanical signals and spatial visual information that can be instinctively understood by the human body. This allows the training subject to simultaneously improve their intuitive control, risk prediction ability, driving experience, and training effectiveness during complex terrain training.

[0007] In combination with the first aspect, in some possible implementations, the method for determining the terrain picture includes: in response to the training subject's triggering operation on the target control, displaying multiple terrain types, and the target control is used to control the simulated driving system to enter a working state after being triggered; in response to the training subject's selection operation of a first terrain type from the multiple terrain types, determining the terrain picture as a terrain picture corresponding to the first terrain type; or, from the multiple terrain types, screening out the second terrain type with the lowest training score or the shortest training time, and determining the terrain picture as a terrain picture corresponding to the second terrain type.

[0008] In the above technical solution, after the training subject triggers the target control, the simulation driving system will display multiple terrain types. This can inform the training subject of the multiple terrain types that can be trained in an intuitive and visual way. When the training subject autonomously selects the first terrain type, the simulation driving system responds to the autonomous selection and displays the corresponding terrain picture on the virtual reality device. This can give the training subject the initiative over the training environment (terrain type), so that it can prioritize training the terrain areas that it is interested in or that it thinks need to be strengthened. When the training subject does not autonomously select from multiple terrain types, the solution can intelligently filter out the second terrain type with the "lowest training score" or "shortest training time" of the training subject, and automatically display the terrain picture accordingly. This "shortcomings-filling" mode based on objective data can prevent the training subject from avoiding weaknesses due to fear of difficulty, thereby significantly improving training efficiency and overall driving skill level.

[0009] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, the method for determining the terrain picture corresponding to each terrain type among the multiple terrain types includes: for any terrain type, inputting the terrain type and random noise into a preset conditional generative adversarial network, and the generator in the conditional generative adversarial network generates an actual terrain picture corresponding to the terrain type; the judge in the conditional generative adversarial network determines the generation deviation of the actual terrain picture, and the generation deviation is the deviation between the elevation distribution of multiple terrain grids divided into the terrain area and the theoretical elevation distribution, as well as the cumulative deviation between the terrain attribute data of the multiple terrain grids and the benchmark attribute data: when the generation deviation is less than the preset deviation, the terrain picture corresponding to the terrain type is determined as the actual terrain picture.

[0010] In the above technical solution, by calculating the deviation in elevation distribution of multiple terrain grids used to divide the terrain area, it is possible to ensure that the geometric structure of the terrain area conforms to natural laws. At the same time, the cumulative deviation of multiple terrain grids in terrain attributes is also considered, which can ensure that key parameters such as friction coefficient and material density conform to the physical characteristics of the real terrain area. The actual terrain picture generated is only adopted when the generated deviation is lower than the preset deviation, which makes the terrain picture have both dynamic diversity and physical rigor. In addition, random noise can make it possible to have rich variants when generating terrain areas of the same terrain type, avoiding terrain repetition. Therefore, this solution can generate terrain pictures that are close to the real terrain and avoid terrain repetition.

[0011] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, the terrain picture displayed on the virtual reality device is adjusted based on the second type of data, including: determining a position change based on the vehicle position in the second type of data and the current position of the vehicle, and adjusting the display content in the terrain picture displayed on the virtual reality device based on the position change; and determining an orientation change based on the vehicle head orientation in the second type of data and the current vehicle head orientation, and adjusting the viewing angle when the terrain picture is displayed on the virtual reality device based on the orientation change.

[0012] In the above technical solution, the displayed content in the terrain image is updated in real time based on the vehicle's position changes. This ensures that environmental displacement is strictly consistent with physical movement. For example, as the vehicle moves forward, reference objects such as trees and rocks in the terrain image move backward in true proportion. Furthermore, the viewing angle of the terrain image is rotated synchronously with the direction of the vehicle's head, ensuring that the training subject's field of view always matches the vehicle's heading angle. For example, when the steering wheel turns right, the horizon rotates left to simulate the changing scenery outside the window. This coordinated adjustment can solve the pain point of the disconnect between the terrain image and the physical sense, significantly optimizing the visual immersion and training effectiveness of simulated driving.

[0013] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, the method also includes: obtaining the vehicle status time series data of the training subject during the simulated driving training process; based on the vehicle status time series data, identifying the abnormal behavior of the training subject and determining the abnormal behavior score; based on the vehicle status time series data, determining the comprehensive skill score of the training subject on multiple preset indicators when driving; based on the abnormal behavior score and the comprehensive skill score, determining the total score; based on the total score and the abnormal behavior, generating a training report.

[0014] In the above technical solution, the scores of the training subjects during the training process are quantified through two dimensions, which can significantly improve the comprehensiveness and guiding value of the training feedback. In the process of determining the total score, the comprehensive skill score and the abnormal behavior score are integrated. The former positively accumulates the operating efficiency based on the preset indicators, and the latter implements negative deductions for abnormal behaviors. This enables the final total score to objectively reveal the compound defect of "technical standards meet but safety awareness is weak", and avoids the blind spot of ignoring dangerous habits in traditional scoring. In the process of generating training reports, the total score and abnormal behavior are deeply combined, which can achieve accurate diagnosis and intervention: for example, if the training subjects with higher total scores only have minor risks such as swerving the steering wheel, they can be prompted to solidify muscle memory in a targeted manner through somatosensory feedback (such as steering wheel shaking simulation). That is, the above solution can form a closed loop with somatosensory feedback, thereby systematically optimizing training behavior.

[0015] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, based on the vehicle state time series data, identifying the abnormal behavior of the training subject and determining the abnormal behavior score include: determining the standard driving data of the training subject, where the standard driving data is the driving operation data that the training subject should present under various road conditions; for any driving type data in the vehicle state time series data, determining the target road condition corresponding to the driving type data, and comparing the driving type data with the target driving operation data that should be presented under the target road condition; when the driving type data is different from the target driving operation data, determining the target driving behavior corresponding to the driving type data as abnormal behavior, and determining the abnormal behavior score of the target driving behavior as the deviation amplitude between the driving type data and the target driving operation data.

[0016] In this technical solution, standard driving data for each driver under various road conditions is pre-established based on the training subject's own abilities. This avoids misjudgments caused by universal standards. When real-time driving data deviates from the individual's optimal benchmark for the same scenario, it is determined to be abnormal behavior. This personalized judgment mechanism truly reflects individual decline or fluctuations in performance rather than absolute ability gaps. When determining the abnormal behavior score, the deviation between the driving type data and the target driving operation data is quantified. This not only objectively measures the severity of the abnormal behavior but also naturally avoids misjudgments caused by differences in driving style.

[0017] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, the multiple preset indicators include stability and fuel economy, the vehicle state time series data includes yaw angular velocity and acceleration, and based on the vehicle state time series data, the comprehensive skill score of the training subject on multiple preset indicators when driving is determined, including: determining the first score of the training subject in stability based on multiple yaw angular velocities and the standard deviation of the multiple yaw angular velocities; determining the second score of the training subject in fuel economy based on the total number of times the training subject performs sudden acceleration behavior during the simulated driving training process, the acceleration of the training subject during each sudden acceleration behavior, and the duration corresponding to each number; and determining the comprehensive skill score based on the first score and the second score.

[0018] In the above technical solution, multiple standard deviations of yaw angular velocity are selected as stability criteria. This is because yaw angular velocity can directly represent the smoothness of vehicle trajectory control. A lower standard deviation means that the training subject maintains continuous and stable steering input in scenarios such as curves or lane changes, which can avoid the risk of skidding due to frequent direction corrections. At the same time, the solution also evaluates fuel economy based on the number of sudden acceleration behaviors, the acceleration when sudden acceleration behaviors occur, and the duration corresponding to each number. This can not only punish the operation of frequently pressing the accelerator pedal, but also distinguish the energy consumption impact of instantaneous deep pressing and continuous high load. This dual-index design can combine multiple indicators to obtain a more accurate comprehensive skill score.

[0019] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, the comprehensive skill score is determined based on the first score and the second score, including: determining the third score of the training subject in completion efficiency based on the training completion time of the simulated driving training process, the shortest completion time of multiple training processes and the standard deviation of the completion time; determining the fourth score of the training subject in terrain adaptability based on the deviation between the actual friction coefficient and the recommended friction coefficient of the terrain type; determining the comprehensive skill score based on the first score, the second score, the third score and the fourth score.

[0020] In the above technical solution, two key indicators, completion efficiency and terrain adaptability, are introduced when determining the comprehensive skill score. The third score is based on the completion time of the training, the shortest completion time in history, and the average time (standard deviation of completion time). This can not only motivate the training subjects to improve their operating speed, but also constrain their performance stability through the standard deviation of completion time. The fourth score quantifies the deviation between the actual friction coefficient and the theoretically recommended friction coefficient, which can directly test the training subjects' perception accuracy of road conditions and their ability to adjust the vehicle. These two key indicators form a four-dimensional evaluation framework with stability and fuel economy. When the above four types of scores are integrated into a comprehensive skill score, the solution can accurately identify the combined shortcomings of the training subjects and promote the training process from the improvement of a single skill to the cultivation of full-dimensional combat capabilities.

[0021] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, a training report is generated based on the total score and the abnormal behavior, including: analyzing the abnormal behavior and generating first suggestion information for improving the abnormal behavior; determining the training rating of the training object based on the total score; formulating new training suggestion information for the training object based on the training rating and the terrain type and corresponding training rating that the training object has trained; and generating the training report by combining the first suggestion information and the training suggestion information.

[0022] In the above technical solution, the cause of each abnormal behavior (such as sudden acceleration or violent steering) is accurately analyzed to generate specific operational suggestions (such as special training for deep control of the accelerator pedal). At the same time, the training rating is determined based on the total score, and new training recommendation information is formulated based on historical training terrain data. If the training subject has a higher training rating in the plain area, training in the mountainous area can be performed. This dual-path suggestion generation solution not only solves the shortcomings of current training, but also plans long-term capability leaps. The training report integrates micro-operation improvements with macro-growth paths, allowing training subjects to clearly perceive current bottlenecks and breakthrough directions, avoiding the blindness and fragmentation of skill improvement in traditional assessments.

[0023] In a second aspect, a driving simulation system is provided, comprising: a simulated cockpit, a virtual reality device, and a driving controller, wherein the simulated cockpit includes driving controls and a seat, and the driving controller is configured to:

[0024] In response to a control operation of the driving control element by the training subject on the virtual reality device while a terrain image is displayed on the virtual reality device, first vehicle state data at a next moment is determined based on the terrain type, the control operation, and a preset vehicle dynamics model, the terrain image being used to present a terrain area corresponding to the terrain type in which the vehicle can safely drive;

[0025] Dividing the first vehicle status data into first-category data and second-category data, the first-category data being vehicle status data that can be directly perceived by the body, and the second-category data being vehicle status data that can be perceived visually or spatially;

[0026] Based on the first type of data, controlling the current state of the driving control component and the seat;

[0027] Based on the second type of data, the terrain image displayed on the virtual reality device is adjusted. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic diagram of a scenario using a simulated driving system provided in an embodiment of the present application;

[0029] Figure 2 This is a modular schematic diagram of a driving simulation system provided by an embodiment of the present application;

[0030] Figure 3 is a schematic flow chart of a training method based on simulated driving provided in an embodiment of the present application;

[0031] Figure 4 This is a schematic diagram of a first-person perspective of a training object provided in an embodiment of the present application;

[0032] Figure 5 This is a schematic diagram of an interface for determining a terrain image provided by an embodiment of the present application;

[0033] Figure 6 This is a schematic diagram of adjusting a terrain image provided by an embodiment of the present application;

[0034] Figure 7 This is a schematic diagram of generating a training report provided in an embodiment of the present application;

[0035] Figure 8 This is a schematic structural diagram of a driving simulation training device provided in an embodiment of the present application;

[0036] Figure 9 It is a structural diagram of a driving simulation system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will clearly and thoroughly describe the technical solutions in this application in conjunction with the accompanying drawings. In the description of the embodiments of this application, unless otherwise specified, " / " means or, for example, A / B can mean A or B: "and / or" in the text is only a description of the association relationship of associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more than two.

[0038] In the following, the terms "first" and "second" are used for descriptive purposes only and should not be understood to imply or suggest relative importance or implicitly indicate the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features.

[0039] Figure 1 This is a schematic diagram of a scenario using a simulated driving system provided in an embodiment of the present application.

[0040] As vehicles become an indispensable means of transportation in people's daily lives, people also need to improve their driving skills.

[0041] Currently, driving skills training is typically conducted through on-site training at driving schools. However, this on-site training method is unsafe and subject to weather and geographical restrictions. However, with the maturity of virtual reality (VR) technology and the widespread adoption of vehicle electronic systems, it is possible to develop a driving simulation system that combines VR equipment with existing vehicle control signal inputs.

[0042] For example, Figure 1 As shown, the training subjects can train their driving skills in the simulated driving system based on the images on the VR device to improve the driving experience and training effect of the training subjects.

[0043] Figure 2 This is a modular schematic diagram of a simulated driving system provided in an embodiment of the present application.

[0044] For example, Figure 2As shown, the driving simulation system 200 includes a hardware interface module 201, a terrain generation module 202, a multi-body dynamics engine module 203, a rendering and physical interaction module 204, and a training evaluation and analysis module 205. Among them, the hardware interface module 201 is used to collect the control operations of the driving control parts, and perform denoising, coordinate conversion and standardization preprocessing on the operation data corresponding to the control operations to ensure the real-time and accuracy of the control operations. The preprocessed operation data is input into the multi-body dynamics engine module 203. In addition, the vehicle state data is processed by a filtering algorithm to reduce the impact of noise and accurately predict the future state of the vehicle. The terrain generation module 202 is used to generate a terrain picture corresponding to the terrain type, and is bound to terrain attribute data (such as friction coefficient and material density), and the terrain picture is displayed on the virtual reality device in the driving simulation system. The multi-body dynamics engine module 203 is used to determine the first vehicle state data at the next moment based on the terrain type, the control operation, and the preset vehicle dynamics model when the training subject controls the driving control components based on the terrain image displayed on the virtual reality device, and divide the first vehicle state data into first-category data and second-category data, and input the first-category data and the second-category data into the VR rendering and physical interaction module 204. In addition, the multi-body dynamics engine module 203 also inputs the first vehicle state data at the next moment into the training evaluation and analysis module 205. The VR rendering and physical interaction module 204 is used to control the current state of the driving control components and the seat based on the first-category data, and adjust the terrain image displayed on the virtual reality device based on the second-category data. The training evaluation and analysis module 205 is used to analyze the training process of the training subject based on the vehicle state data at multiple moments (i.e., the vehicle state time series data during the training process) and generate a training report.

[0045] Figure 3 This is a schematic flow chart of a training method based on simulated driving provided in an embodiment of the present application. The implementation of this training method depends on Figure 2 .

[0046] It should be understood that the training method based on simulated driving provided in the embodiment of the present application can be applied to Figure 2 The simulated driving system shown includes a simulated cockpit and virtual reality equipment, and the simulated cockpit includes driving control parts and seats.

[0047] For example, Figure 3 As shown, the method 300 includes the following steps 301 to 304.

[0048] Step 301, in a case where a terrain image is displayed on the virtual reality device, in response to the training subject's control operation of the driving control component, first vehicle state data at the next moment is determined based on the terrain type, the control operation and a preset vehicle dynamics model. The terrain image is used to present a terrain area corresponding to the terrain type in which the vehicle can safely drive.

[0049] It should be understood that the virtual reality device in the above step 301 includes a display screen, and the virtual reality device is a device that displays a three-dimensional and virtual terrain picture through a display screen. The virtual reality device can be a head-mounted device or a wall-mounted device. When the virtual reality device is a head-mounted device, the training subject can wear the virtual reality device and perform driving training based on the terrain picture on the display screen. In addition, the posture data of the virtual reality device can be measured by the inertial sensor built into the virtual reality device, so that when the posture of the virtual reality device is indicated to change, the viewing angle of the terrain picture displayed by the virtual reality device can be adjusted. For example, when you turn your head, the terrain picture also turns. When the virtual reality device is a wall-mounted device, the terrain picture displayed thereon is the terrain picture from the first perspective of the training subject, specifically Figure 4 As shown, Figure 4 The terrain image of the virtual reality device is used to present a mountainous area in front of the training subject's first perspective, and the terrain area corresponds to the terrain type and does not exist in the real world.

[0050] Terrain types include plains, plateaus, mountains, basins, and hills. A terrain screen can be used to display the terrain area corresponding to a terrain type, such as a plain area corresponding to a plain type, a plateau area corresponding to a plateau type, and a mountain area corresponding to a mountain type. A terrain area corresponding to a terrain type refers to an independent terrain unit dominated by that terrain type. Terrain types can also include deserts, mining areas, ice and snow, wading areas, and muddy areas.

[0051] It should be understood that the aforementioned driving controls include a steering wheel, shift lever, accelerator pedal, parking brake handle, and brake pedal, and may also include a clutch pedal. The training subject can turn the steering wheel, shift gears using the shift lever, press the accelerator pedal and brake pedal or clutch pedal, and control the parking brake handle to raise or lower. In other words, the control operations include turning the steering wheel, moving the shift lever, pressing the accelerator pedal and brake pedal or clutch pedal, and raising or lowering the parking brake handle. After the training subject controls the driving controls, vehicle state data is generated. This vehicle state data is theoretical data determined based on the terrain type, control operations, and a preset vehicle dynamics model, and the simulated cockpit position does not move.

[0052] Changes in vehicle state indicated by the vehicle state data can be fed back to the training subject through changes in the terrain image and seat position. For example, if the accelerator pedal is depressed and the steering wheel angle remains unchanged, the vehicle's position changes, and the content displayed in the terrain image from the training subject's first perspective can be adjusted. For example, if the steering wheel angle is changed, the vehicle's frontal orientation changes, and the perspective of the terrain image displayed on the virtual reality device can be adjusted. In the two examples of terrain image adjustments mentioned above, for the training subject, the first example involves zooming in on a distant, straight-ahead terrain area in the terrain image, and the second example involves switching the terrain area originally from the first perspective to an alternate perspective. This alternate perspective is determined by the steering wheel angle. For example, if the steering wheel is turned left, the alternate perspective represents a leftward turn based on the first perspective. If the steering wheel is turned right, the alternate perspective represents a rightward turn based on the first perspective. For another example, if the brake pedal is pressed hard, the change in vehicle state can be reflected by the seat tilting forward.

[0053] It should be understood that the preset vehicle dynamics model in the above step 301 refers to a mathematical model established based on the physical characteristics and motion laws of the vehicle, which is used to describe the changes in the motion state of the vehicle under force. The vehicle dynamics model realizes the prediction of the future state of the vehicle (such as position, speed, acceleration, heading angle, etc.) by abstracting the mechanical structure, dynamic characteristics and external forces of the vehicle. In some cases, the vehicle dynamics model may ignore the processing of noise, which affects the authenticity of the measurement data of various sensors. The operation data corresponding to the control operation is measured by the sensor. Therefore, based on the terrain type, the control operation and the vehicle dynamics model, the first vehicle state data at the next moment needs to be further processed. The filtering algorithm can be used to process the noise to accurately predict the impact of the future state of the vehicle.

[0054] It should also be understood that the above step 301 is executed by the multi-body dynamics engine module 203 .

[0055] In some embodiments, the vehicle yaw angle in the first vehicle state data at the next moment may be corrected by the following formula (1):

[0056] x k =Fx k-1 +Bu k +w k (1)

[0057] Among them, x k is the corrected vehicle yaw angle, F is the state transfer matrix, x k-1 is the vehicle yaw angle in the first vehicle state data at the next moment, B is the control input matrix, which is used to indicate the influence of the steering wheel angle on the vehicle yaw angle, u kis the steering wheel angle, w k is the process noise.

[0058] It should also be understood that the coordinate system used by the virtual reality device is the world coordinate system, while the simulated cockpit uses its own local coordinate system. Therefore, when adjusting the terrain image displayed on the virtual reality device based on the vehicle status data, a coordinate system conversion is involved. The specific formula (2) is given below:

[0059] T world =R yaw ·R pitch ·R roll ·T local +t offset (2)

[0060] Among them, when taking the front position as an example, T local is the coordinate of the vehicle head in the local coordinate system, R yaw 、R pitch and R roll is the rotation matrix, which represents the rotation transformation around the X axis (roll), Y axis (pitch), and Z axis (yaw), t offset is the coordinate offset, which represents the deviation between the coordinate corresponding to the head of the vehicle in the local coordinate system and the reference point, T world The coordinates of the converted vehicle head in the world coordinate system.

[0061] It should be understood that the sensor measurement data can also be normalized so that the simulated driving system processes the normalized measurement data. For example, the steering wheel angle is normalized based on the following formula (3):

[0062]

[0063] Among them, θ norm-st is the normalized steering wheel angle, θ raw-st is the current steering wheel angle, θ min-st is the minimum steering wheel angle, θ max-st is the maximum steering wheel angle.

[0064] In addition, the target pedal opening can be normalized based on the following formula (4):

[0065]

[0066] Among them, θ norm-pe is the normalized opening, θ raw-pe is the current opening of the target pedal, θ min-pe is the minimum opening of the target pedal, θ max-pe is the maximum opening of the target pedal.

[0067] In some embodiments, the control operation corresponds to operation data, and the preset vehicle dynamics model includes a tire mechanics model. Based on the terrain type, the control operation and the preset vehicle dynamics model, the first vehicle state data at the next moment is determined, including: determining the lateral force generated by the tire based on the tire mechanics model, the steering wheel angle in the operation data and the actual steering angle of the wheel; determining the longitudinal force generated by the tire based on the friction coefficient of the terrain area corresponding to the terrain type, the vehicle speed at the current moment, the tire angular velocity at the current moment and the suspension displacement at the current moment; determining the centripetal force based on the lateral force and the longitudinal force. force; based on the centripetal force, the current position of the vehicle, the vehicle speed and the current acceleration, determine the vehicle position, vehicle speed and current acceleration in the first vehicle state data at the next moment; based on the suspension displacement and the suspension movement speed at the current moment, determine the vertical force of the road surface acting on the tire in the vertical direction; based on the vehicle mass and the change in vehicle speed when the vehicle passes through the sudden terrain change at the target height at the first time, determine the impact force applied to the vehicle; based on the vertical force and the impact force, determine the ground reaction force; based on the ground reaction force and the suspension displacement at the current moment, determine the suspension displacement in the first vehicle state data at the next moment.

[0068] In some embodiments, a method for determining the lateral force generated by a tire includes: determining the lateral force based on the following formula (5);

[0069] F y =Dsin[Carctan(Bα-E(Bα-arctan(Bα)))] (5)

[0070] Among them, F y is the lateral force, D is the maximum lateral force, C is the saturation region linearization coefficient, B is the sideslip angle sensitivity, α is the tire sideslip angle, specifically the deviation between the steering wheel angle and the actual steering angle, and E is the shape parameter used to control the smoothness of the transition region.

[0071] In some embodiments, a method for determining the longitudinal force generated by a tire includes: determining the longitudinal force based on the following formula (6);

[0072] F x =μN·η sr (6)

[0073] Among them, F x is the longitudinal force, N is the vertical force of the road surface acting on the tire in the vertical direction, μ is the friction coefficient, η sr is the tire slip ratio, which is based on the current vehicle speed, tire angular velocity, and tire radius. The vertical force is related to the current suspension displacement. Determine N based on formula (7) as follows:

[0074] N=mg+F sus (7)

[0075] Where m is the mass of the vehicle, g is the acceleration due to gravity, and F sus The support force of the suspension system on the tire can be calculated based on the formula F sus =cv t +ky t Determine, c is the damping coefficient, unit is N (m / s), k is the spring stiffness, unit is N / m, y t is the suspension displacement at the current moment, v t The current speed of the suspension.

[0076] In some embodiments, a method for determining an impact force applied to a vehicle includes: determining the impact force based on the following formula (8);

[0077]

[0078] Among them, F im is the impact force, Δv is the vehicle speed change, Δt is the first time, and when it is assumed that the vertical speed of the tire changes uniformly, Δv=Δh / Δt, where Δh is the displacement change of the tire in the vertical direction.

[0079] In some embodiments, determining the suspension displacement in the first vehicle state data at a next moment based on the ground reaction force and the suspension displacement at a current moment includes: based on the following formulas (9) to (11);

[0080]

[0081] v t+1 =v t +a t ·Δt (10)

[0082] y t+1 =y t +v t ·Δt (11)

[0083] Among them, a t is the acceleration at the current moment, F gr,t is the ground reaction force, Δt is the time difference between the current moment and the next moment, v t+1 is the moving speed of the suspension at the next moment, y t+1 is the suspension displacement at the next moment.

[0084] In one possible implementation, the method for determining the terrain picture in step 301 includes: displaying multiple terrain types in response to the training subject's triggering operation on a target control, where the target control is used to control the simulated driving system to enter a working state after being triggered; in response to the training subject's selection operation of a first terrain type from the multiple terrain types, determining the terrain picture as a terrain picture corresponding to the first terrain type; or, from the multiple terrain types, screening out a second terrain type with the lowest training score or the shortest training time, and determining the terrain picture as a terrain picture corresponding to the second terrain type.

[0085] It should be understood that in the above scheme, each time a terrain image corresponding to a terrain type is trained during the training process, the training process will be scored and the training time will be recorded. The second terrain type mentioned above can specifically be the terrain type with the lowest average training score or the shortest total training time.

[0086] In the above technical solution, after the trainee triggers the target control, the driving simulation system displays multiple terrain types. This intuitively and visually informs the trainee of the multiple terrain types available for training. When the trainee autonomously selects the first terrain type, the driving simulation system responds by displaying the corresponding terrain image on the virtual reality device. This gives the trainee initiative over the training environment (terrain type), allowing them to prioritize terrain areas of interest or areas they believe need improvement. This personalized starting point selection effectively stimulates the trainee's training motivation and engagement, avoids resistance caused by being forced to assign terrain types they are not interested in, and allows training to better suit their personal preferences and learning pace. If the trainee does not autonomously select from the multiple terrain types, the solution intelligently selects the second terrain type with the trainee's "lowest training score" or "shortest training time" and automatically displays the terrain image accordingly. In other words, the solution can leverage the trainee's historical training data to accurately identify the trainee's skill weaknesses and proactively recommend terrain types for training, forcing the trainee to strengthen their skills in a targeted manner. This "shortcomings-filling" model based on objective data can prevent trainees from avoiding weaknesses due to fear of difficulty, or repeatedly training terrain scenes that they have already mastered, thereby significantly improving training efficiency and overall driving skill level.

[0087] Figure 5 This is a schematic diagram of an interface for determining a terrain image provided in an embodiment of the present application.

[0088] For example, Figure 5 As shown in (a), after the training subject clicks the target control on the control platform in the simulated cockpit, the control platform displays the following Figure 5The multiple terrain types shown in (b) include plain, plateau, mountain, basin, and hill. After the training subject clicks on the mountain type among the multiple terrain types, a terrain image corresponding to the mountain type is displayed on the display screen of the virtual reality device.

[0089] In one possible implementation, a method for determining a terrain picture corresponding to each of the multiple terrain types includes: for any terrain type, inputting the terrain type and random noise into a preset conditional generative adversarial network, and having a generator in the conditional generative adversarial network generate an actual terrain picture corresponding to the terrain type; having a judge in the conditional generative adversarial network determine a generation deviation of the actual terrain picture, where the generation deviation is the deviation between the elevation distribution of multiple terrain grids divided into the terrain area and the theoretical elevation distribution, and the cumulative deviation between the terrain attribute data of the multiple terrain grids and the benchmark attribute data: when the generation deviation is less than the preset deviation, the terrain picture corresponding to the terrain type is determined as the actual terrain picture.

[0090] It should be understood that the preset conditional generative adversarial network in the above scheme is specifically CFAN (Conditional Generative Adversarial Networks), and CGAN is a deep learning network that introduces conditional constraints (terrain type) and random noise into the generative adversarial network to achieve directional generation. Its core lies in that in the generation scenario of the terrain picture, when the generator in CGAN receives the terrain type as a conditional constraint and cooperates with random noise, the generator will parse the semantics of the above conditional constraints through a multi-layer neural network structure. For example, the "mountain type" is mapped to a steep and undulating spatial structure, and random noise is used to introduce the randomness of the landform details to obtain a three-dimensional terrain picture. The discriminator in CGAN is used to simultaneously evaluate the visual authenticity of the terrain picture and the degree of realism with the geological features, and through adversarial optimization, it forces the generator to finally output a synthetic terrain picture that is highly consistent with the specified terrain type.

[0091] It should also be understood that the multiple terrain grids used to divide a terrain area refer to a grid array covering the entire surface area formed by discretizing the continuous surface area in the terrain area into regular geometric units (such as squares). Each grid is used to indicate a surface block of a fixed area. The elevation distribution of multiple terrain grids refers to the statistical laws presented by the elevation set of multiple terrain grids within a terrain area (such as a mountainous area), which is used to reflect the overall undulating characteristics of the terrain area. Among them, the statistical laws can be laws on multiple statistical features such as mean and standard deviation. The elevation (height) of each terrain grid can be the elevation of the center point or the mean elevation of the entire terrain grid.

[0092] Among them, different terrain types correspond to different terrain areas, and the elevation distribution of multiple terrain grids divided by the terrain area is also different. The theoretical elevation distribution of multiple terrain grids divided by the terrain area corresponding to a terrain type is pre-set. Optionally, for plain areas, the elevations of most of the multiple terrain grids are distributed in the range of 0 to 200 meters (that is, the elevations are concentrated in the low area), and the standard deviation between the elevations is extremely small; for plateau areas, the elevations of most of the multiple terrain grids are distributed in the range of 1000 to 8000 meters, and the standard deviation between the elevations is medium; for mountainous areas, the elevation range corresponding to multiple terrain grids is extremely wide, which can be in the range of -100 to 9000 meters (that is, the elevations corresponding to multiple terrain grids show multi-peak jumps), and the standard deviation between the elevations is extremely large.

[0093] It should also be understood that the terrain attribute data of each terrain grid in the above scheme include the friction coefficient μ and material density. The friction coefficient is used to measure the adhesion performance between the tire and the road surface. The larger the friction coefficient, the better the adhesion performance between the tire and the road surface, and the more stable the vehicle driving; conversely, the worse the adhesion performance, the easier it is for the vehicle to slip or lose control. Material density refers to the mass of the material forming the terrain area within the unit volume corresponding to each terrain grid. Its core function is to determine the physical response (such as deformation depth and rebound height, etc.) when the vehicle collides with the terrain area by affecting the mass distribution of the terrain grid. For example, in a mountainous area, the actual material density of granite corresponding to a certain terrain grid is 2000kg / m 3 In addition, the terrain area may be formed by more than one substance within the unit volume corresponding to a terrain grid. The cumulative deviation between the terrain attribute data of multiple terrain grids and the baseline attribute data in the above scheme is obtained by summing the deviations between the terrain attribute data of each terrain grid and the baseline attribute data.

[0094] Different terrain types correspond to different terrain areas, and the terrain attribute data of each terrain grid divided by the terrain area is also different. The baseline attribute data of each terrain grid divided by the terrain area corresponding to a terrain type is pre-set. Optionally, for mountainous areas, the baseline friction coefficient corresponding to each terrain grid is 7.5 and the baseline material density of granite is 2650kg / m 3 . Use strict physical constraints to ensure the credibility of the generated terrain.

[0095] It should also be understood that the above technical solution is executed by the terrain generation module 202 .

[0096] In the above technical solution, by calculating the deviation in elevation distribution of multiple terrain grids used to divide the terrain area, it is possible to ensure that the geometric structure of the terrain area conforms to natural laws. At the same time, the cumulative deviation of multiple terrain grids in terrain attributes is also considered, which can ensure that key parameters such as friction coefficient and material density conform to the physical characteristics of the real terrain area. The actual terrain picture generated is only adopted when the generated deviation is lower than the preset deviation, which makes the terrain picture have both dynamic diversity and physical rigor. In addition, random noise can make it possible to have rich variants when generating terrain areas of the same terrain type, avoiding terrain repetition. Therefore, this solution can generate terrain pictures that are close to the real terrain and avoid terrain repetition.

[0097] In some embodiments, determining the generation deviation of the actual terrain image by a judge in the conditional generative adversarial network includes: determining the generation deviation based on the following formula (12);

[0098] L G =-E[logD(G(z|c))]+λ*‖G(z|c)-y real ‖ (12)

[0099] Among them, L G The first term -E[logD(G(z|c))] is used to represent the score of the decision maker for the terrain area G(z|c) corresponding to the actual terrain image. The first term is determined by the deviation between the elevation distribution of multiple terrain grids divided by the terrain area G(z|c) and the theoretical elevation distribution. The second term ‖G(z|c)-y real ‖ is determined by the cumulative deviation of the terrain attribute data of multiple terrain grids and the benchmark attribute data, y real is the baseline attribute data, the terrain region G(z|c) corresponds to the terrain attribute data, and λ is the weight of the second term. The first term can be regarded as the adversarial bias (forcing the generated terrain region to approach the real elevation distribution), and the second term can be regarded as the conditional bias (forcing the generated terrain region to approach the baseline attribute data).

[0100] In step 302 , the first vehicle status data is divided into first category data and second category data. The first category data is vehicle status data that can be directly perceived by the body, and the second category data is vehicle status data that can be perceived visually or spatially.

[0101] It should be understood that the first type of data in step 302 can be directly perceived by the body (i.e., somatosensory), specifically physical stimulation received during training by mechanical receptors such as the skin, muscles, and vestibular system, triggering instinctive physiological reactions without visual conversion. The second type of data in step 302 can be perceived through visual / spatial perception (i.e., visual perception), specifically relying on the visual cortex or spatial cognition ability to determine the relative relationships of objects during training.

[0102] It should also be understood that the above step 302 is executed by the multi-body dynamics engine module 203 .

[0103] In some embodiments, the first vehicle status data is divided into first category data and second category data in step 302, including: determining the data in the first vehicle status data that needs to be directly acted on the human body through a mechanical device as the first category data, and determining the data in the first vehicle status data that needs to be recognized through a visual system or spatial relationship as the second category data.

[0104] For example, the first vehicle state data includes longitudinal / lateral acceleration, steering wheel torque, suspension displacement, vehicle position, vehicle head orientation, pitch angle, and bank angle. Among them, longitudinal / lateral acceleration, steering wheel torque, suspension displacement, and pitch angle can receive physical stimulation during training through mechanical receptors in the body. Therefore, these state data can be classified as first-category data. Vehicle position, vehicle head orientation, pitch angle, and bank angle require the visual cortex or spatial cognition ability in the body to judge the relative relationship of objects during training. Therefore, these state data can be classified as second-category data.

[0105] It should be noted here that the same vehicle status data may be classified as both the first and second categories of data. For example, the pitch angle in the above example is because the pitch angle can be received by mechanical receptors and can also be judged by the visual cortex or spatial cognitive ability. Specifically, when the vehicle goes downhill, the front of the vehicle pitches (there is a pitch angle), and the side wings of the seat can be used to apply pressure to the back of the training subject to simulate the influence of the gravity component. The ratio of the sky to the ground in the terrain picture can also be dynamically adjusted so that the training subject can judge the slope change by vision.

[0106] Step 303: Control the current status of the driving control component and the seat based on the first type of data.

[0107] It should be understood that, for the first type of data, the driving control components may specifically be a steering wheel and a brake pedal.

[0108] It should also be understood that the above step 303 is performed by the VR rendering and physical interaction module 204.

[0109] For example, for the longitudinal / lateral acceleration in the first category of data, reverse pressure is applied to the side wings of the seat to simulate the inertial force; for the steering wheel torque in the first category of data, the torque motor corresponding to the steering wheel generates a reaction force to simulate the change in road grip; for the suspension displacement in the first category of data, the exciter at the bottom of the seat is used to simulate road bumps; for the longitudinal tire force in the first category of data, the change in road grip can be simulated by the recoil force of the brake pedal.

[0110] In some embodiments, based on the first type of data, controlling the current state of the driving control component and the seat includes: controlling the seat to vibrate at a target force and controlling the motor corresponding to the steering wheel in the driving control component to vibrate at a target force based on the following formula (13);

[0111] F tac =Asin(2πf h t)+Bsin(2πf l t) (13)

[0112] Among them, F tac is the target strength, A is the first amplitude, A=k μ μ|v|, v is the vehicle speed in the first type of data, f h is the friction vibration frequency, which is a constant value, f h ≥200Hz, t is the time, B is the second amplitude, B=k s |y sus |,y sus is the suspension displacement in the first type of data, f l is the natural frequency of the suspension system, is a constant value, f l ≤30Hz. Where, k μ Friction-force conversion coefficient, unit is (N·s / m), k s The displacement-force conversion coefficient is expressed in N / m. The motors corresponding to the seat and steering wheel are controlled to vibrate, simulating the vibrations generated by tire-road friction and suspension system deformation during training, which provide tactile feedback to the training subject.

[0113] It should be understood that the first term, the friction component: Asin(2πf h t), its source is: the vibration caused by the friction between the tire and the road, the first amplitude: A = k μ μ|v|, the greater the speed, the greater A, such as when going over a speed bump at high speed, the vibration is stronger; the greater the friction coefficient, the greater A, such as muddy ground vibrates more violently than ice. The second term, suspension component: Bsin(2πf l t), its source is: oscillation caused by the deformation of the suspension system, the second amplitude: B = k s |y sus |, the greater the suspension displacement, the greater the B, for example, the vibration feedback from a deep pit is stronger than that from a shallow pit.

[0114] Step 304: Adjust the terrain image displayed on the virtual reality device based on the second type of data.

[0115] It should be understood that in the above step 304, based on the second type of data, the image content in the terrain image and the viewing angle of the virtual reality device when displaying the terrain image can be adjusted.

[0116] It should also be understood that the above step 303 is performed by the VR rendering and physical interaction module 204.

[0117] For example, the vehicle position in the second type of data is used to adjust the image content in the terrain picture from the first perspective of the training subject; the vehicle head orientation in the second type of data can be used to adjust the perspective when the virtual reality device displays the terrain picture; and the pitch angle in the second type of data is used to dynamically adjust the ratio of the sky to the ground in the terrain picture when the vehicle climbs a slope, so that the training subject can judge the slope visually.

[0118] In some embodiments, the method 300 further includes: for any terrain grid, determining the detail level of the terrain grid based on the actual distance from the position of the training subject's first perspective to the terrain grid and a preset distance, where the preset distance is the distance corresponding to retaining details; and adjusting the display accuracy of the terrain grid on the terrain screen based on the detail level.

[0119] In some embodiments, determining the level of detail of the terrain grid based on the actual distance from the position of the first perspective of the training subject to the terrain grid and the preset distance includes: determining the level of detail based on the following formula (14);

[0120]

[0121] Among them, L lod For this level of detail, d is is the actual distance, d thre for this preset distance.

[0122] It should be understood that the above formula (14) determines the fineness of the terrain grid. The closer the terrain grid is to the first perspective of the training subject, the more details it has; the farther the terrain area is from the first perspective of the training subject, the less details it has. The preset distance can be 50m, L lod The larger the value, the coarser the terrain mesh.

[0123] Exemplarily, the terrain grid is a square, and the preset distance is 50m. When the actual distance is less than 50m, the detail level is 0, and the display accuracy of the terrain grid is adjusted to 0.3m*0.3m / pixel, that is, the size of the surface area represented by each pixel is 0.3m*0.3m; when the actual distance is greater than or equal to 50m and less than 100m, the detail level is 1, and the display accuracy of the terrain grid is adjusted to 0.6m*0.6m / pixel; if the actual distance is farther, the display accuracy can be further reduced.

[0124] In some embodiments, the method 300 further includes: for any terrain mesh, determining the material roughness corresponding to the terrain mesh based on the following formula (15), and binding the material roughness to the terrain mesh;

[0125] deg rough =max(0.1,μ) (15)

[0126] Among them, deg rough is the material roughness corresponding to the terrain mesh, and μ is the friction coefficient corresponding to the terrain mesh. That is, terrain areas with a high friction coefficient correspond to materials with high roughness, appearing rougher and more matte; terrain areas with a low friction coefficient correspond to materials with low roughness, appearing smoother. Furthermore, binding the material roughness to the terrain mesh can be accomplished by loading a texture corresponding to the terrain type and the material roughness onto the terrain mesh.

[0127] In some embodiments, the method 300 further includes: for any terrain grid, determining the brightness of the outgoing light in the outgoing direction of a reference point on the terrain grid based on the following formula (16), and binding the brightness of the outgoing light to the terrain grid;

[0128] L o (w o )=L e (w o )+∫ Ω ρ(w i →w o )f r (w i →w o )L i (w i )dw i (16)

[0129] Among them, L o (w o ) is the emission direction w of any reference point on the terrain grid o The brightness of the outgoing light on e (w o ) is the emission direction o The self-luminous brightness on i →w o ) is the terrain mesh surface at the reference point from the incident direction w i To the outgoing direction w o The reflectivity, f r (w i →w o ) is the incident direction w i To the outgoing direction w o The light reflection ratio, Li (w i ) is the incident direction w of the reference point i The incident light brightness on the i The light energy irradiated on the surface of the terrain mesh. In addition, the brightness of the outgoing light is bound to the terrain mesh, which can be achieved by loading a light map corresponding to the brightness of the outgoing light on the terrain mesh.

[0130] In one possible implementation, step 304 includes: adjusting the terrain image displayed on the virtual reality device based on the second type of data, including: determining a position change based on the vehicle position in the second type of data and the current position of the vehicle, and adjusting the display content of the terrain image displayed on the virtual reality device based on the position change; and determining an orientation change based on the vehicle head orientation in the second type of data and the current vehicle head orientation, and adjusting the viewing angle when displaying the terrain image on the virtual reality device based on the orientation change.

[0131] It should be understood that in the above scheme, adjusting the display content in the terrain picture displayed on the virtual reality device based on the position change refers to bringing the distant terrain area closer; adjusting the perspective when the virtual reality device displays the terrain picture based on the orientation change refers to switching the terrain area originally from the first perspective in the terrain picture to the terrain area from other perspectives.

[0132] In the above technical solution, the displayed content in the terrain image is updated in real time based on the vehicle's position changes. This ensures that environmental displacement is strictly consistent with physical movement. For example, as the vehicle moves forward, reference objects such as trees and rocks in the terrain image move backward in true proportion. Furthermore, the viewing angle of the terrain image is rotated synchronously with the direction of the vehicle's head, ensuring that the training subject's field of view always matches the vehicle's heading angle. For example, when the steering wheel turns right, the horizon rotates left to simulate the changing scenery outside the window. This coordinated adjustment can solve the pain point of the disconnect between the terrain image and the physical sense, significantly optimizing the visual immersion and training effectiveness of simulated driving.

[0133] In some embodiments, based on the vehicle position in the second category of data and the current position of the vehicle, a position change is determined, including: based on the current position, the steering wheel angle at the current position and the vehicle position, inserting multiple positions between the current position and the vehicle position; and determining the position change as the multiple positions.

[0134] In some embodiments, determining the orientation change based on the vehicle head orientation in the second type of data and the current vehicle head orientation of the vehicle includes: determining the orientation change based on the following formula (17);

[0135]

[0136] Among them, k ij is the interpolation time between the current time i and the next time j, q i is the current vehicle head direction at the current time i, q j is the vehicle head direction at the next moment j in the second type of data, and θ is the vehicle head direction q j With the current vehicle heading q i The angular deviation between is the interpolation time k ij The direction of the vehicle's front when .

[0137] It should be understood that when it is necessary to insert the vehicle head direction of multiple moments between the current moment i and the next moment j, the moment (i+j) / 2 can be inserted between the current moment i and the next moment j first, and then the moment (i+((i+j) / 2)) / 2 can be inserted between the current moment i and the moment (i+j) / 2, and the moment (((i+j) / 2)+j) / 2 can be inserted between the moment (i+j) / 2 and the next moment j, and so on, until the interpolated moments meet the quantity requirements.

[0138] It should also be understood that the purpose of the above scheme is to control the adjustment process to smoothly transition when adjusting the viewing angle of the virtual reality device when displaying the terrain image, avoid the terrain image from teleporting or freezing, ensure the continuity of the terrain image, and make the visual changes more in line with human eye expectations.

[0139] Figure 6 This is a schematic diagram of adjusting a terrain image provided in an embodiment of the present application.

[0140] For example, Figure 6 As shown, the VR rendering and physics interaction module 204 receives the second type of data sent by the multi-body dynamics engine module 203, determines a position change based on the vehicle position in the second type of data and the vehicle's current position, and determines an orientation change based on the vehicle's head orientation in the second type of data and the vehicle's current head orientation, ultimately controlling the virtual reality device. Specifically, the display content of the terrain image displayed on the virtual reality device is adjusted based on the position change, and the viewing angle of the terrain image displayed on the virtual reality device is adjusted based on the orientation change.

[0141] In one possible implementation, after step 304, the method 300 further includes: obtaining the vehicle state time series data of the training subject during the simulated driving training process; based on the vehicle state time series data, identifying the abnormal behavior of the training subject and determining the abnormal behavior score; based on the vehicle state time series data, determining the comprehensive skill score of the training subject on multiple preset indicators while driving; based on the abnormal behavior score and the comprehensive skill score, determining the total score; and generating a training report based on the total score and the abnormal behavior.

[0142] It should be understood that in the above scheme, the vehicle state time series data refers to the vehicle state data at each of the multiple moments during the entire training process. Abnormal behavior refers to behavior that deviates from normal driving and has a negative impact on driving safety or training results. Behavior that has a negative impact on training results refers to behavior that interferes with the achievement of training goals, hinders skill improvement, or reduces training efficiency. It can be intentional collision behavior (swerving the steering wheel and crashing into a roadside tree when there are no obstacles in the vehicle on a wide plain area). Abnormal behaviors include sudden acceleration / deceleration (the absolute value of the longitudinal acceleration is large and the rate of change of the opening of the accelerator pedal / brake pedal is large), sudden steering (the steering wheel angular velocity is large and accompanied by a sudden change in the yaw angular velocity), oversteering (the yaw angular velocity is relatively large and the rear wheel slip angle is greater than the front wheel slip angle), understeering (the steering wheel angle continues to increase, the vehicle's lateral acceleration decreases, and the front wheel slip angle is saturated), fatigue driving behavior (the steering wheel steering correction frequency decreases, the brake response delay time is large, the steering wheel angle micro-motion is small and lasts for a long time), and speeding. The abnormal behavior score is used to measure the abnormality of the abnormal behavior.

[0143] The slip angle is the angle between the tire's actual direction of travel and the wheel plane. Front wheel slip angle saturation refers to the phenomenon where, under the influence of lateral force, the front wheel slip angle reaches a certain critical value and no longer increases significantly even with further increase in lateral force.

[0144] It should also be understood that the multiple preset indicators in the above scheme may include stability and fuel economy. Stability refers to the vehicle's ability to maintain its intended driving trajectory while resisting external interference (such as crosswinds, uneven road surfaces, and sharp turns). Fuel economy refers to the distance a vehicle can travel per unit of fuel consumed, typically expressed by fuel consumption (L / 100km). The lower the fuel consumption, the better the fuel economy, which is essentially a comprehensive reflection of engine efficiency, energy consumption, and driving resistance.

[0145] It should also be understood that the above technical solution is executed by the training evaluation and analysis module 205 .

[0146] In the above technical solution, the scores of the training subjects during the training process are quantified through two dimensions, which can significantly improve the comprehensiveness and guiding value of the training feedback. In the process of determining the total score, the comprehensive skill score and the abnormal behavior score are integrated. The former positively accumulates the operating efficiency based on the preset indicators, and the latter implements negative deductions for abnormal behaviors. This enables the final total score to objectively reveal the compound defect of "technical standards meet but safety awareness is weak", and avoids the blind spot of ignoring dangerous habits in traditional scoring. In the process of generating training reports, the total score and abnormal behavior are deeply combined, which can achieve accurate diagnosis and intervention: for example, if the training subjects with higher total scores only have minor risks such as swerving the steering wheel, they can be prompted to solidify muscle memory in a targeted manner through somatosensory feedback (such as steering wheel shaking simulation). That is, the above solution can form a closed loop with somatosensory feedback, thereby systematically optimizing training behavior.

[0147] In one possible implementation, based on the vehicle state time series data, the abnormal behavior of the training subject is identified and an abnormal behavior score is determined, including: determining standard driving data of the training subject, where the standard driving data is the driving operation data that the training subject should present under various road conditions; for any driving type data in the vehicle state time series data, determining a target road condition corresponding to the driving type data, and comparing the driving type data with target driving operation data that should be presented under the target road condition; if the driving type data is different from the target driving operation data, determining the target driving behavior corresponding to the driving type data as abnormal behavior, and determining the abnormal behavior score of the target driving behavior as the deviation amplitude between the driving type data and the target driving operation data.

[0148] It should be understood that in the above scheme, the standard driving data for a training subject can be determined based on the subject's historical training driving data. If no historical training driving data exists for a particular training subject, the standard driving data of another training subject can be used. However, the other training subject must be associated with the first training subject. This association can include the other training subject being of the same gender as the first training subject, or being of similar height to the first training subject. Furthermore, the aforementioned road conditions can include turning conditions, lane changes, and the like.

[0149] Furthermore, the target driving operation data to be presented under the target road conditions in the above scheme refers to the standard operating data that the training subject should demonstrate when driving under the target road conditions. The essence of "presenting" is to provide the training subject with an operating benchmark. Furthermore, the target driving operation data to be presented under the same road conditions may vary for different training subjects.

[0150] In this technical solution, standard driving data for each driver under various road conditions is pre-established based on the training subject's own abilities. This avoids misjudgments caused by universal standards. When real-time driving data deviates from the individual's optimal benchmark for the same scenario, it is determined to be abnormal behavior. This personalized judgment mechanism truly reflects individual decline or fluctuations in performance rather than absolute ability gaps. When determining the abnormal behavior score, the deviation between the driving type data and the target driving operation data is quantified. This not only objectively measures the severity of the abnormal behavior but also naturally avoids misjudgments caused by differences in driving style.

[0151] In some embodiments, the method for determining the abnormal behavior score of the target driving behavior includes: determining the abnormal behavior score based on the following formula (18);

[0152] SC this-abn =sigmoid(w f *tanh(W i X t +b i )+U f h t-1 ) (18)

[0153] Among them, SC this-abn is the abnormal behavior score of the target driving behavior, X t is any driving type data of the target driving behavior at time t, h t-1 is the driving type data at the previous moment t-1, implicitly representing the target driving operation data (i.e., standard driving data) corresponding to any driving type data, W i For X t The weight matrix, W f is the gate weight, which is used to control the influence of any driving type data of the target driving behavior at time t on the abnormal behavior score, U f h t-1 The weight matrix is used to adjust the contribution of standard driving data, b i SC is the bias vector of any driving type data of the target driving behavior at time t. this-abn The value range is (0%, 100%).

[0154] In one possible implementation, the multiple preset indicators include stability and fuel economy, and the vehicle state time series data includes yaw angular velocity and acceleration. Based on the vehicle state time series data, the comprehensive skill score of the training subject on the multiple preset indicators while driving is determined, including: determining a first score of the training subject in stability based on multiple yaw angular velocities and the standard deviation of the multiple yaw angular velocities; determining a second score of the training subject in fuel economy based on the total number of times the training subject performs sudden acceleration behavior during the simulated driving training process, the acceleration of the training subject during each sudden acceleration behavior, and the duration corresponding to each number; and determining the comprehensive skill score based on the first score and the second score.

[0155] It should be understood that the yaw rate in the above scheme describes the angular velocity of the vehicle when rotating around an axis perpendicular to the ground (i.e., the z-axis), and directly reflects the speed of the vehicle's rotation when turning. The first score, second score, and overall skill score are all expressed as percentages.

[0156] In the above technical solution, multiple standard deviations of yaw angular velocity are selected as stability criteria. This is because yaw angular velocity can directly represent the smoothness of vehicle trajectory control. A lower standard deviation means that the training subject maintains continuous and stable steering input in scenarios such as curves or lane changes, which can avoid the risk of skidding due to frequent direction corrections. At the same time, the solution also evaluates fuel economy based on the number of sudden acceleration behaviors, the acceleration when sudden acceleration behaviors occur, and the duration corresponding to each number. This can not only punish the operation of frequently pressing the accelerator pedal, but also distinguish the energy consumption impact of instantaneous deep pressing and continuous high load. This dual-index design can combine multiple indicators to obtain a more accurate comprehensive skill score.

[0157] Furthermore, in the above solution, in the step "determining the comprehensive skill score based on the first score and the second score," the comprehensive skill score can be determined by weightedly combining the first and second scores based on a first weight and a second weight. The first weight indicates the contribution of stability to the comprehensive skill score, and the second weight indicates the contribution of fuel economy to the comprehensive skill score. The sum of the first and second weights is 1.

[0158] In some embodiments, determining a first stability score of the training subject based on a plurality of yaw angular rates and a standard deviation of the plurality of yaw angular rates includes: determining the first score based on the following formula (19);

[0159]

[0160] Among them, SC this-1 is the first score, N is the total number of yaw angular velocities, ψt is the yaw rate at time t among the multiple yaw rates, is the standard deviation of the multiple yaw angular velocities.

[0161] In some embodiments, determining a second fuel economy score of the training subject based on the total number of times the training subject performs sudden acceleration during the simulated driving training, the acceleration of the training subject during each sudden acceleration, and the duration corresponding to each number of times includes: determining the second score based on the following formula (20);

[0162]

[0163] Among them, SC this-2 is the second score, NU is the total number of times, is the number N u The acceleration when a sudden acceleration occurs, is the number N u The corresponding duration.

[0164] In one possible implementation, the comprehensive skill score is determined based on the first score and the second score, including: determining a third score of the training subject in completion efficiency based on the training completion time of the simulated driving training process, the shortest completion time of multiple training processes, and the standard deviation of the completion time; determining a fourth score of the training subject in terrain adaptability based on the deviation between the actual friction coefficient and the recommended friction coefficient of the terrain type; and determining the comprehensive skill score based on the first score, the second score, the third score, and the fourth score.

[0165] It should be understood that in the above scheme, the training completion time, minimum completion time, and completion time standard deviation are all for terrain images of the same terrain type. The recommended friction coefficient is the friction coefficient that the terrain image corresponding to that terrain type should match. The actual friction coefficient of that terrain type may deviate from the recommended friction coefficient. In addition, the third and fourth scores are both expressed as percentages.

[0166] The above technical solution incorporates two key indicators: completion efficiency and terrain adaptability when determining the comprehensive skill score. A third score is based on the training session completion time, the historically best shortest completion time, and the average time (standard deviation of completion time). This not only motivates trainees to improve their operating speed, but also constrains their performance consistency through the standard deviation of completion time. The fourth score quantifies the deviation between the actual friction coefficient and the theoretically recommended friction coefficient, directly testing the trainee's perception of road conditions and their ability to tune the vehicle. These two key indicators, along with stability and fuel economy, form a four-dimensional evaluation framework. Completion efficiency can offset the neglect of timeliness in purely skill-based evaluation, while terrain adaptability complements environmental adaptability, a core element of practical application. When these four scores are combined into a comprehensive skill score, the solution can accurately identify the trainee's multiple weaknesses, shifting the training process from single-skill improvement to comprehensive combat capability development, significantly expanding the practical guidance value of driving skill assessment.

[0167] In some embodiments, determining a third score of the training subject on completion efficiency based on the training completion time of the simulated driving training process, the shortest completion time of multiple training processes, and the standard deviation of the completion time includes: determining the third score based on the following formula (21);

[0168]

[0169] Among them, SC this-3 For the third score, T this is the training completion time, μ T is the shortest completion time, σ T is the standard deviation of the completion time.

[0170] In some embodiments, determining a fourth score of the training subject on terrain adaptability based on a deviation between an actual friction coefficient and a recommended friction coefficient of the terrain type includes: determining the fourth score based on the following formula (22);

[0171]

[0172] Among them, SC this-4 is the fourth score, NG is the total number of terrain grids divided into the terrain area corresponding to the terrain type, is the terrain grid N g The actual friction coefficient, is the terrain grid N g Recommended coefficient of friction.

[0173] In some embodiments, determining the comprehensive skill score based on the first score, the second score, the third score, and the fourth score includes: determining the comprehensive skill score based on the following formula (23);

[0174] SC this-sk =ω3SC this-1 +ω4SC this-2 +ω5SC this-3 +ω6SC this-4 (twenty three)

[0175] Among them, SC this-sk is the comprehensive skill score, ω3 is the third weight, ω4 is the fourth weight, ω5 is the fifth weight, and ω6 is the sixth weight. The third weight indicates the contribution of stability to the comprehensive skill score, the fourth weight indicates the contribution of fuel economy to the comprehensive skill score, the fifth weight indicates the contribution of completion efficiency to the comprehensive skill score, and the sixth weight indicates the contribution of terrain adaptability to the comprehensive skill score. The sum of the third, fourth, fifth, and sixth weights is 1.

[0176] In some embodiments, determining a total score based on the abnormal behavior score and the comprehensive skill score comprises: determining the total score based on the following formula (24);

[0177] SC this =ω7SC this-abn -ω8SC this-sk (twenty four)

[0178] Among them, SC this is the total score, ω7 is the seventh weight, and ω8 is the eighth weight. The sum of the seventh weight and the eighth weight is 1. From the above formula (24), it can be concluded that the abnormal behavior score is a negative score.

[0179] In one possible implementation, a training report is generated based on the total score and the abnormal behavior, including: analyzing the abnormal behavior and generating first suggestion information for improving the abnormal behavior; determining the training rating of the training object based on the total score; formulating new training suggestion information for the training object based on the training rating and the terrain type that the training object has trained and the corresponding training rating; and generating the training report by combining the first suggestion information and the training suggestion information.

[0180] In the above technical solution, the cause of each abnormal behavior (such as sudden acceleration or violent steering) is accurately analyzed to generate concrete operational suggestions (such as special training for deep control of the accelerator pedal). At the same time, the training rating is determined based on the total score, and new training recommendation information is formulated in combination with historical training terrain data. If the training subject has a higher training rating in the plain area, training in the mountainous area can be performed. This dual-path suggestion generation solution not only solves the shortcomings of current training, but also plans long-term capability leaps. The training report integrates micro-operation improvements with macro-growth paths, so that the training subject can clearly perceive the current bottlenecks and breakthrough directions, avoiding the blindness and fragmentation of skill improvement in traditional assessments. Therefore, this solution can significantly improve the improvement efficiency of driving training through intelligent diagnosis and personalized planning mechanisms.

[0181] In some embodiments, based on the total score, the training rating of the training object is determined, including: when the total score is less than or equal to the first preset score, determining the training rating as the first rating; when the total score is greater than the first preset score and less than or equal to the second preset score, determining the training rating as the second rating; when the total score is greater than the second preset score and less than or equal to the third preset score, determining the training rating as the third rating; when the total score is greater than the third preset score, determining the training rating as the fourth rating, the fourth rating being higher than the third rating, the third rating being higher than the second rating, and the second rating being higher than the first rating.

[0182] It should be understood that in the above scheme, the first preset score can be 60%, and the first rating is D, which is used to reflect that the score is not passing and the worst rating; the second preset score can be 75%, and the second rating is C, which is used to reflect that the score is lower than average and the rating is medium; the third preset score can be 85%, and the third rating is B, which is used to reflect that the score is medium and the rating is good; the fourth rating is A, which is used to reflect that the score is above average and the rating is excellent.

[0183] Figure 7 This is a schematic diagram of generating a training report provided in an embodiment of the present application.

[0184] For example, Figure 7As shown, the training evaluation and analysis module 205 receives the vehicle state data (i.e., vehicle state time series data) at multiple moments sent by the multi-body dynamics engine module 203. Based on the vehicle state time series data, the abnormal behavior of the training subject is identified, the abnormal behavior score is determined, and based on the vehicle state time series data, the comprehensive skill score of the training subject on multiple preset indicators when driving is determined. Among them, the comprehensive skill score can be obtained based on the first score and the second score (as well as the third score and the fourth score). Based on the abnormal behavior score and the comprehensive skill score, a total score is determined. Based on the total score and the abnormal behavior, a training report is generated.

[0185] In some embodiments, the method 300 further includes: when the abnormal behavior score is greater than a fourth preset score, outputting a prompt message, wherein the prompt message is used to prompt the training subject to have abnormal behavior and an adjustment method.

[0186] In some embodiments, outputting the prompt information includes any one of the following: broadcasting the prompt information by voice; displaying the prompt information on a control platform; or outputting the prompt information by tactile means.

[0187] Optionally, the training report includes the number of times the same type of abnormal behavior occurs, the corresponding duration and total duration of each time, as well as a radar chart of the training subject's stability, fuel economy, completion efficiency and terrain adaptability determined by the first score, second score, third score and fourth score, and training video clips with abnormal behavior.

[0188] In some embodiments, after generating the training report, the method 300 further includes: sending the training report to the training subject's mobile terminal and / or the trainer's mobile terminal.

[0189] Figure 8 This is a structural diagram of a training device based on simulated driving provided in an embodiment of the present application.

[0190] Exemplarily, the device is built into a simulated driving system, which includes a simulated cockpit and a virtual reality device, wherein the simulated cockpit includes driving control components and seats, such as Figure 8 As shown, the apparatus 800 includes:

[0191] Determination module 801 is configured to determine, in response to a training subject's control operation of the driving control element on the virtual reality device, first vehicle state data at a next moment based on the terrain type, the control operation, and a preset vehicle dynamics model, when a terrain image is displayed on the virtual reality device, wherein the terrain image is used to present a terrain area corresponding to the terrain type in which the vehicle can safely drive;

[0192] a classification module 802 for classifying the first vehicle status data into first category data and second category data, wherein the first category data is vehicle status data that can be directly perceived by the body, and the second category data is vehicle status data that can be perceived visually or spatially;

[0193] The control module 803 is used to:

[0194] Based on the first type of data, controlling the current state of the driving control component and the seat;

[0195] Based on the second type of data, the terrain image displayed on the virtual reality device is adjusted.

[0196] Optionally, the device 800 also includes: a display module, used to display multiple terrain types in response to the training subject's triggering operation on the target control, and the target control is used to control the simulated driving system to enter a working state after being triggered; the determination module 801 is specifically used to: in response to the training subject's selection operation of the first terrain type from the multiple terrain types, determine the terrain picture as the terrain picture corresponding to the first terrain type; or, from the multiple terrain types, screen out the second terrain type with the lowest training score or the shortest training time, and determine the terrain picture as the terrain picture corresponding to the second terrain type.

[0197] Optionally, the determination module 801 is further specifically used to: for any terrain type, input the terrain type and random noise into a preset conditional generative adversarial network, and the generator in the conditional generative adversarial network generates an actual terrain picture corresponding to the terrain type; the judge in the conditional generative adversarial network determines the generation deviation of the actual terrain picture, and the generation deviation is the deviation between the elevation distribution of multiple terrain grids divided into the terrain area and the theoretical elevation distribution, as well as the cumulative deviation between the terrain attribute data of the multiple terrain grids and the benchmark attribute data: when the generation deviation is less than the preset deviation, the terrain picture corresponding to the terrain type is determined as the actual terrain picture.

[0198] Optionally, the control module 803 is specifically used to: determine a position change based on the vehicle position in the second category of data and the current position of the vehicle, and adjust the display content in the terrain picture displayed on the virtual reality device based on the position change; and determine an orientation change based on the vehicle head orientation in the second category of data and the current vehicle head orientation, and adjust the viewing angle when the terrain picture is displayed on the virtual reality device based on the orientation change.

[0199] Optionally, the device 800 also includes: an acquisition module for acquiring the vehicle state time series data of the training subject during the simulated driving training process; the determination module 801 is also used to: identify the abnormal behavior of the training subject based on the vehicle state time series data, and determine the abnormal behavior score; determine the comprehensive skill score of the training subject on multiple preset indicators when driving based on the vehicle state time series data; determine the total score based on the abnormal behavior score and the comprehensive skill score; and a generation module for generating a training report based on the total score and the abnormal behavior.

[0200] Optionally, the determination module 801 is further specifically used to: determine standard driving data of the training subject, where the standard driving data is the driving operation data that the training subject should present under various road conditions; for any driving type data in the vehicle state time series data, determine the target road condition corresponding to the driving type data, and compare the driving type data with the target driving operation data that should be presented under the target road condition; if the driving type data is different from the target driving operation data, determine the target driving behavior corresponding to the driving type data as an abnormal behavior, and determine the abnormal behavior score of the target driving behavior as the deviation amplitude between the driving type data and the target driving operation data.

[0201] Optionally, the multiple preset indicators include stability and fuel economy, the vehicle state time series data includes yaw angular velocity and acceleration, and the determination module 801 is further specifically used to: determine the first score of the training subject in stability based on multiple yaw angular velocities and the standard deviation of the multiple yaw angular velocities; determine the second score of the training subject in fuel economy based on the total number of times the training subject performs sudden acceleration behavior during the simulated driving training process, the acceleration of the training subject during each sudden acceleration behavior and the duration corresponding to each number; and determine the comprehensive skill score based on the first score and the second score.

[0202] Optionally, the determination module 801 is further specifically used to: determine the third score of the training subject in completion efficiency based on the training completion time of the simulated driving training process, the shortest completion time of multiple training processes and the standard deviation of the completion time; determine the fourth score of the training subject in terrain adaptability based on the deviation between the actual friction coefficient and the recommended friction coefficient of the terrain type; and determine the comprehensive skill score based on the first score, the second score, the third score and the fourth score.

[0203] Optionally, the generation module is specifically used to analyze the abnormal behavior and generate first suggestion information for improving the abnormal behavior; based on the total score, the training rating of the training object is determined; the determination module 801 is specifically used to formulate new training suggestion information for the training object based on the training rating and the terrain type and corresponding training rating that the training object has trained; the generation module is specifically used to generate the training report based on the first suggestion information and the training suggestion information.

[0204] Figure 9 It is a structural diagram of a driving simulation system provided in an embodiment of the present application.

[0205] For example, Figure 9 As shown, a simulated driving system 900 includes: a simulated cockpit 901, a virtual reality device 902, and a driving controller 903. The simulated cockpit 901 includes driving controls and a seat. The driving controller 903 is used to:

[0206] When a terrain image is displayed on the virtual reality device 902, in response to the training subject's control operation on the driving control element, first vehicle state data at a next moment is determined based on the terrain type, the control operation, and a preset vehicle dynamics model, wherein the terrain image is used to present a terrain area corresponding to the terrain type in which the vehicle can safely drive;

[0207] Dividing the first vehicle status data into first-category data and second-category data, the first-category data being vehicle status data that can be directly perceived by the body, and the second-category data being vehicle status data that can be perceived visually or spatially;

[0208] Based on the first type of data, controlling the current state of the driving control component and the seat;

[0209] Based on the second type of data, the terrain image displayed on the virtual reality device 902 is adjusted.

[0210] In addition, an embodiment of the present application also protects a device, which may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to execute a training method based on simulated driving provided in an embodiment of the present application.

[0211] In this embodiment, the device can be divided into functional modules based on the above-described method examples. For example, each functional module can be mapped to a specific functional module, or two or more functions can be integrated into a single processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and represents only a logical functional division. In actual implementation, other division methods may be used.

[0212] In the case of dividing the functional modules into corresponding functional modules, the device may further include a determination module, a division module, a control module, an acquisition module, a display module, a generation module, etc. It should be noted that all relevant contents involved in the above method embodiments can be referred to the functional description of the corresponding functional modules and will not be repeated here.

[0213] It should be understood that the device provided in this embodiment is used to execute the above-mentioned training method based on simulated driving, and thus can achieve the same effect as the above-mentioned implementation method.

[0214] In the case of an integrated unit, the device may include a processing module and a storage module. When the device is used in a vehicle, the processing module may be used to control and manage the vehicle's movements. The storage module may be used to support the vehicle's execution of relevant executable program code, etc.

[0215] The processing module may be a processor or controller that implements or executes the various exemplary logic blocks, modules, and circuits described in conjunction with the present disclosure. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processing (DSP) and a microprocessor, and the storage module may be a memory.

[0216] In addition, the device provided in the embodiments of the present application can specifically be a chip, component or module, and the chip may include a connected processor and memory; wherein the memory is used to store instructions, and when the processor calls and executes the instructions, the chip can execute a training method based on simulated driving provided in the above embodiment.

[0217] This embodiment also provides a computer-readable storage medium, which stores executable program code. When the executable program code is run on a computer, the computer executes the above-mentioned related method steps to implement a training method based on simulated driving provided in the above embodiment.

[0218] This embodiment also provides a computer program product. When the computer program product is run on a computer, it enables the computer to execute the above-mentioned related steps to implement a training method based on simulated driving provided in the above embodiment.

[0219] Among them, the device, computer-readable storage medium, computer program product or chip provided in this embodiment are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0220] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0221] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0222] The above content is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A training method based on simulated driving, characterized in that: The method is applied to a simulated driving system, wherein the simulated driving system includes a simulated cockpit and a virtual reality device, wherein the simulated cockpit includes driving control components and a seat, and the method includes: In response to a training subject's control operation of the driving control element when a terrain image is displayed on the virtual reality device, first vehicle state data at a next moment is determined based on the terrain type, the control operation, and a preset vehicle dynamics model, wherein the terrain image is used to present a terrain area corresponding to the terrain type in which the vehicle can safely drive; Dividing the first vehicle status data into a first category of data and a second category of data, the first category of data being vehicle status data that can be directly perceived by the body, and the second category of data being vehicle status data that can be perceived visually or spatially; Based on the first type of data, controlling the current state of the driving control components and the seat; Based on the second type of data, the terrain image displayed on the virtual reality device is adjusted.

2. The method according to claim 1, characterized in that The method for determining the terrain picture includes: In response to the training subject triggering an operation on a target control, displaying a plurality of terrain types, wherein the target control is used to control the driving simulation system to enter an operating state after being triggered; In response to the training subject selecting a first terrain type from the plurality of terrain types, determining the terrain picture as a terrain picture corresponding to the first terrain type; or A second terrain type having the lowest training score or the shortest training time is screened out from the plurality of terrain types, and the terrain picture is determined to be a terrain picture corresponding to the second terrain type.

3. The method according to claim 2, characterized in that The method for determining the terrain picture corresponding to each of the plurality of terrain types includes: For any terrain type, the terrain type and random noise are input into a preset conditional generative adversarial network, and the generator in the conditional generative adversarial network generates an actual terrain image corresponding to the terrain type; The decision maker in the conditional generative adversarial network determines the generation deviation of the actual terrain image. The generation deviation is the deviation between the elevation distribution of the plurality of terrain grids divided into the terrain area and the theoretical elevation distribution, as well as the cumulative deviation between the terrain attribute data of the plurality of terrain grids and the reference attribute data: In a case where the generated deviation is smaller than the preset deviation, the terrain picture corresponding to the terrain type is determined as the actual terrain picture.

4. The method according to claim 1, wherein The adjusting the terrain image displayed on the virtual reality device based on the second type of data includes: Determining a position change based on the vehicle position in the second type of data and the current position of the vehicle, and adjusting display content in a terrain image displayed on the virtual reality device based on the position change; and Based on the vehicle head orientation in the second type of data and the current vehicle head orientation of the vehicle, an orientation change is determined, and based on the orientation change, a viewing angle of the virtual reality device when displaying a terrain image is adjusted.

5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: Acquiring vehicle state time series data of the training subject during the simulated driving training process; identifying abnormal behavior of the training subject based on the vehicle state time series data and determining an abnormal behavior score; Determining, based on the vehicle state time series data, a comprehensive skill score of the training subject on multiple preset indicators while driving; determining a total score based on the abnormal behavior score and the comprehensive skill score; A training report is generated based on the total score and the abnormal behavior.

6. The method according to claim 5, characterized in that The identifying abnormal behavior of the training object based on the vehicle state time series data and determining an abnormal behavior score includes: Determining standard driving data of the training subject, where the standard driving data is driving operation data that the training subject should present under various road conditions; For any driving type data in the vehicle state time series data, determining a target road condition corresponding to the driving type data, and comparing the driving type data with target driving operation data that should be presented under the target road condition; When the driving type data is different from the target driving operation data, the target driving behavior corresponding to the driving type data is determined to be an abnormal behavior, and the abnormal behavior score of the target driving behavior is determined as the deviation amplitude between the driving type data and the target driving operation data.

7. The method according to claim 5, characterized in that The plurality of preset indicators include stability and fuel economy, the vehicle state time series data includes yaw rate and acceleration, and determining the comprehensive skill score of the training subject on the plurality of preset indicators while driving based on the vehicle state time series data includes: determining a first stability score for the training subject based on a plurality of yaw rates and a standard deviation of the plurality of yaw rates; determining a second fuel economy score for the training subject based on a total number of sudden accelerations by the training subject during the simulated driving training, the acceleration of the training subject during each sudden acceleration, and the duration of each sudden acceleration; The comprehensive skill score is determined based on the first score and the second score.

8. The method according to claim 7, characterized in that Determining the comprehensive skill score based on the first score and the second score includes: determining a third score of the training subject in completion efficiency based on a training completion time of the simulated driving training process, a shortest completion time of multiple training processes, and a standard deviation of the completion time; determining a fourth score of the training subject in terrain adaptability based on a deviation between an actual friction coefficient of the terrain type and a recommended friction coefficient; The comprehensive skill score is determined based on the first score, the second score, the third score, and the fourth score.

9. The method according to claim 5, characterized in that The generating of a training report based on the total score and the abnormal behavior includes: Analyzing the abnormal behavior and generating first suggestion information for improving the abnormal behavior; determining a training rating of the training subject based on the total score; formulating new training suggestion information for the training subject based on the training rating and the terrain type on which the training subject has trained and the corresponding training rating; The first suggestion information and the training suggestion information are used to generate the training report.

10. A driving simulation system, characterized in that: The simulated driving system includes: a simulated cockpit, a virtual reality device and a driving controller. The simulated cockpit includes driving control components and a seat. The driving controller is used to: In response to a training subject's control operation of the driving control element when a terrain image is displayed on the virtual reality device, first vehicle state data at a next moment is determined based on the terrain type, the control operation, and a preset vehicle dynamics model, wherein the terrain image is used to present a terrain area corresponding to the terrain type in which the vehicle can safely drive; Dividing the first vehicle status data into a first category of data and a second category of data, the first category of data being vehicle status data that can be directly perceived by the body, and the second category of data being vehicle status data that can be perceived visually or spatially; Based on the first type of data, controlling the current state of the driving control components and the seat; Based on the second type of data, the terrain image displayed on the virtual reality device is adjusted.

Images