Intelligent driving training auxiliary and safety control system and driving training auxiliary teaching method

Abstract

The application discloses a kind of intelligent driving training auxiliary and safety control system and driving training auxiliary teaching method.The system includes: perception layer, for collecting vehicle surrounding environment data, vehicle state data and driver behavior data;Decision planning layer is connected with perception layer, for processing environmental data and state data, constructs dynamic scene model, and generates teaching instruction or safety control instruction based on expert system;Control execution layer is connected with decision planning layer, for receiving safety control instruction and executing the brake, steering or power control operation of vehicle;Interaction layer is connected with decision planning layer, for outputting teaching instruction and receiving the voice or operation feedback of driver.The application can realize centimeter-level trajectory correction based on multi-source perception fusion and expert system decision, humanization real-time teaching guidance and hierarchical active safety intervention, so as to improve the standardization level of teaching and training safety.

Description

Technical Field

[0001] This invention belongs to the field of intelligent driving and driver training technology, and more specifically, relates to an intelligent driving training assistance and safety control system and a driving learning assistance teaching method. Background Technology

[0002] With the continuous growth of motor vehicle ownership, the driver training industry faces both huge market demand and safety challenges. Traditional driver training primarily relies on a mentor-apprentice system using a "training vehicle + human instructor" model. In this model, the instructor provides verbal instructions to the student and intervenes in emergencies by braking from the passenger side or taking control of the steering wheel. However, this traditional model has several significant drawbacks: First, teaching quality is highly dependent on the coach's personal experience and state of mind. Different coaches have vastly different teaching methods, evaluation criteria, and levels of patience, leading to inconsistent learning outcomes for students. Furthermore, coaches are prone to fatigue, emotional fluctuations, and even verbal abuse during prolonged, repetitive instruction, which severely impacts students' mental state and learning efficiency.

[0003] Secondly, there is a lack of quantitative and objective real-time feedback mechanisms. Traditional teaching relies heavily on the instructor's "feeling" to judge whether the student's operation is standardized, making it difficult to conduct centimeter-level quantitative analysis of subtle operations such as vehicle trajectory, steering angle, and throttle control. Students often only know that they "did something wrong," but do not know "where they went wrong" or "how to correct it," leading to the solidification of incorrect actions and the formation of bad driving habits.

[0004] Secondly, safety protection suffers from lag and limitations. Human instructors' reaction speed is limited by physiological constraints. When faced with sudden, extremely dangerous situations such as a student accidentally pressing the accelerator or losing steering control, they often struggle to provide optimal intervention within milliseconds, easily leading to training accidents. Furthermore, traditional secondary braking devices have limited functionality, only capable of emergency braking and unable to provide more precise safety controls such as steering assist or power limiting.

[0005] 1. In recent years, although some driving schools have introduced simple GPS-based auxiliary systems or basic video monitoring equipment, these existing technologies still have significant shortcomings: 2. Low perception accuracy: Most rely solely on ordinary GPS positioning, with errors at the meter level. This cannot meet the centimeter-level accuracy requirements for tasks such as reversing into a parking space and parallel parking. Furthermore, it lacks comprehensive environmental perception capabilities regarding surrounding obstacles, lane lines, and traffic signals.

[0006] 3. Poor interactivity: It mostly consists of post-event playback or simple voice broadcasts, lacking intuitive guidance based on augmented reality (AR), and failing to achieve immersive teaching that is "what you see is what you get".

[0007] 4. Low level of intelligence: It lacks adaptive teaching strategies based on expert systems, and cannot dynamically adjust the teaching difficulty and intervention timing according to the learner's ability profile. It often provides mechanical prompts in a "one-size-fits-all" manner.

[0008] Insufficient safety: The system lacks a robust hierarchical arbitration logic and failure protection mechanism. Existing systems often fail directly in the event of sensor malfunction or communication interruption, failing to guarantee the vehicle's automatic and safe stopping under extreme conditions.

[0009] In summary, how to construct an intelligent driving training assistance and safety control system that integrates high-precision environmental perception, intelligent decision-making and planning, precise execution of drive-by-wire chassis, and human-like interaction to achieve standardization, quantification, intelligence, and inherent safety in the teaching process has become a key technical problem that urgently needs to be solved in the current driving training technology field. Summary of the Invention

[0010] The purpose of this invention is to propose an intelligent driving training assistance and safety control system and a driving assistance teaching method to solve the technical problems of inconsistent teaching quality, lack of objective quantitative feedback, and lagging safety protection caused by reliance on human experience in the traditional driver training model; to achieve centimeter-level trajectory correction, anthropomorphic real-time teaching guidance, and graded active safety intervention based on multi-source perception fusion and expert system decision-making, thereby improving the standardization of teaching and training safety.

[0011] To achieve the above objectives, in a first aspect, the present invention proposes an intelligent driving training assistance and safety control system, comprising: The perception layer is used to collect data on the vehicle's surrounding environment, the vehicle's own status, and the driver's behavior. The decision planning layer, connected to the perception layer, is used to process the environmental data and its own state data, construct a dynamic scene model, and generate teaching instructions or safety control instructions based on the expert system. The control execution layer, connected to the decision planning layer, is used to receive the safety control commands and execute the vehicle's braking, steering, or power control operations. The interaction layer, connected to the decision planning layer, is used to output the teaching instructions and receive voice or operation feedback from the driver.

[0012] Optionally, the sensing layer includes: LiDAR is used to acquire three-dimensional point cloud data and distance information of surrounding obstacles; The visual sensor array, including surround-view and forward-view cameras, is used to identify lane lines, traffic signs, and traffic lights. Millimeter-wave radar, positioned at the four corners of the vehicle, is used for blind spot monitoring and lane change assistance; Ultrasonic radar, installed on the front and rear bumpers of the vehicle, is used for near-field collision protection; The positioning module, including an RTK-GNSS receiver, a dual-antenna direction-finding unit, and an inertial measurement unit, is used to acquire the vehicle's centimeter-level absolute position, heading angle, and attitude information in a stationary state, and automatically switch to SLAM positioning mode based on lidar point cloud matching or visual odometry in areas where GNSS signals are blocked. The driver monitoring module is used to capture the driver's facial expressions, gaze direction, and fatigue level.

[0013] Optionally, the control execution layer includes a drive-by-wire chassis modification mechanism: The brake-by-wire unit, using a co-pilot auxiliary brake motor or an electro-hydraulic braking system, is used to receive emergency braking commands and perform active braking. The steer-by-wire unit obtains the electric power steering protocol by parsing the electric power steering protocol or establishing a direct communication connection, and directly sends steering angle or torque commands to achieve automatic steering or force feedback teaching. The drive-by-wire power unit includes an electronic throttle signal interceptor and a clutch travel sensor, used to limit maximum throttle output, block accidental throttle input signals, or automatically replenish throttle when engine stall is anticipated.

[0014] Optionally, the decision planning layer has a built-in expert system engine, which includes: The ideal trajectory generation module stores demonstration trajectory data from experienced coaches; The deviation analysis module uses dynamic time warping or Frescher distance algorithm to calculate the trajectory deviation between the student's actual trajectory and the ideal trajectory; The tiered arbitration logic module is used to trigger different levels of intervention strategies based on the level of deviation value, and is defined as follows: Level 0: When the trajectory deviation value is less than the monitoring threshold, only monitoring is performed, without intervention; Level 1: When the trajectory deviation value is greater than or equal to the monitoring threshold and less than the first preset threshold, and there is no risk of collision, a visual or auditory prompt is issued through the interaction layer. Level 2: When the trajectory deviation value is greater than or equal to the first preset threshold and less than the second preset threshold, or when overspeeding or crossing the line is detected, the throttle output is limited or a reverse torque is applied to the steering wheel. Level 3: When the trajectory deviation value is greater than or equal to the second preset threshold, or the collision time is less than the safety threshold, or the student is detected to have accidentally stepped on the accelerator, the vehicle control is forcibly taken over and emergency braking or steering is performed to avoid the collision. Among them, the monitoring threshold is less than the first preset threshold and less than the second preset threshold.

[0015] Optionally, the expert system engine also includes an adaptive teaching module: This module builds a competency profile based on the trainee's historical training data and dynamically adjusts the level of detail in teaching instructions and the timing of intervention. In beginner mode, frequent and detailed step-by-step instructions and early intervention protection are provided; In skilled mode, streamlined error correction instructions are provided and control transfer conditions are relaxed.

[0016] Optionally, the interaction layer includes an augmented reality head-up display and an intelligent voice assistant: The augmented reality head-up display is used to project virtual guide lines, hazard markings, and visual animations of wheel corners onto the windshield, achieving a fusion display of information and the real road surface; The intelligent voice assistant supports multi-turn dialogue and can explain the reasons for deductions or answer students' questions about the timing of operations based on current environmental data.

[0017] Optionally, it also includes a cloud-based monitoring service platform: The vehicle video stream, dashboard data and location information are uploaded to the cloud in real time via the 5G / V2X communication module. It supports remote safety operators to take over control when the in-vehicle system is unable to handle complex scenarios. The cloud-based monitoring service platform is used to generate learning reports that include trajectory playback, deduction point statistics, and operation smoothness analysis, and uses the collected training data to iteratively optimize the expert system model.

[0018] Optionally, it also includes a high-precision map module and an electronic fence module: The high-precision map module pre-stores a semantically annotated digital map of the training ground, including lane lines, stop lines, project areas, and virtual walls, which is used to compare with real-time perception data to determine the current teaching stage of the vehicle. The electronic fence module uses RTK positioning to determine whether a vehicle has driven out of the preset training area boundary. If it does, it automatically cuts off the throttle and triggers a slow braking stop. It also sets dynamic speed thresholds for different training projects to intervene in speeding.

[0019] Optionally, a failure protection mechanism may also be included: It features hardware redundancy design, and key sensor signals are input via dual channels; A heartbeat monitoring mechanism is provided. When the communication between the central computing unit and the underlying controller is interrupted, the underlying controller automatically executes the safety parking logic of cutting off power, applying maximum braking and activating hazard lights. Physical emergency stop buttons are provided inside the vehicle, outside the vehicle, and remotely. Pressing the button will directly cut off the power to the actuator and trigger the mechanical brake.

[0020] Secondly, this invention proposes a driving training assistance method based on the intelligent driving training assistance and safety control system described in any one of the first aspects, comprising the following steps: Load the pre-built digital twin model of the training site based on multi-source sensor fusion, and load the rule engine for the corresponding examination item; Real-time collection of driving operation data and environmental data from trainees, and calculation of the deviation between the vehicle's current position and the ideal trajectory; Based on the level of deviation and potential risks, provide human-like teaching guidance through AR-HUD and voice assistant; When the system detects that a student's operation is about to cause a safety accident or violate the examination rules, it will automatically intervene to control the vehicle according to the hierarchical arbitration logic until the danger is eliminated. After the training is completed, the data is uploaded to the cloud to generate a personalized analysis report and update the trainee's competency profile.

[0021] The beneficial effects of this invention are as follows: Through comprehensive and high-precision real-time acquisition of data on the vehicle's surrounding environment, vehicle status, and driver behavior by the perception layer, a comprehensive and objective quantitative data foundation is provided for the system, fundamentally solving the problem of traditional manual teaching relying on the instructor's subjective observation and experience-based judgment, which lacks accurate data support. Through deep fusion processing of the perception data by the decision-making and planning layer, a dynamic scene model including the vehicle, obstacles, and training ground is constructed. Based on the built-in expert system, the actual operation trajectory of the trainee is compared and deviation analyzed in real time with the ideal trajectory of the experienced instructor. According to the hierarchical arbitration logic, different levels of teaching instructions or safety control instructions are intelligently generated, achieving a seamless switch from merely monitoring to proactive safety intervention. This ensures the trainee's autonomous practice space within a safe range and enables precise responses to impending safety incidents within milliseconds. It effectively overcomes the inherent shortcomings of human instructors, such as reaction speed being limited by physiological limits and susceptibility to errors under fatigue. By controlling the execution layer to precisely execute steer-by-wire and power-by-wire units, it transforms safety control commands issued by the decision-making and planning layer into smooth and reliable vehicle braking, steering, or power control actions, while retaining the mechanical connection and highest priority of the physical pedals. This ensures that active safety intervention is both highly efficient and precise, and has multiple redundancy guarantees. Through the connection between the interaction layer and the decision-making and planning layer, it outputs teaching instructions generated by the expert system to the driver in the form of visual guidance via augmented reality head-up display and natural language dialogue via intelligent voice assistant. At the same time, it receives voice or operation feedback from the driver to form a closed loop of human-computer interaction. This transforms abstract teaching information into intuitive visual symbols and warm, anthropomorphic dialogue, significantly reducing the cognitive load of learners and enhancing learning immersion and teaching effectiveness. Therefore, this invention constructs a complete technical solution integrating high-precision environmental perception, intelligent decision-making teaching, drive-by-wire active safety, and human-like interaction. It effectively solves the technical problems of inconsistent teaching quality, lagging safety protection, lack of objective quantitative feedback, and rigid human-computer interaction in traditional driver training models, and realizes the standardization, datafication, intelligence, and inherent safety of the driver training process.

[0022] The system of the present invention has other features and advantages that will be apparent from or will be set forth in detail in the accompanying drawings and following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description

[0023] The above and other objects, features and advantages of the present invention will become more apparent from the accompanying drawings, in which like reference numerals generally denote like parts.

[0024] Figure 1 A schematic diagram of an intelligent driving training assistance and safety control system according to Embodiment 1 of the present invention is shown.

[0025] Figure 2 A flowchart illustrating the steps of a driving assistance teaching method according to Embodiment 2 of the present invention is shown. Detailed Implementation

[0026] The invention will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0027] Example 1

[0028] like Figure 1 As shown, this embodiment provides an intelligent driving training assistance and safety control system, including: The perception layer is used to collect data on the vehicle's surrounding environment, the vehicle's own status, and the driver's behavior. The decision-making and planning layer, connected to the perception layer, is used to process environmental data and its own state data, build dynamic scene models, and generate teaching instructions or safety control instructions based on the expert system. The control execution layer, connected to the decision planning layer, is used to receive safety control commands and execute vehicle braking, steering, or power control operations. The interaction layer, connected to the decision-making and planning layer, is used to output teaching instructions and receive voice or operational feedback from the driver.

[0029] Specifically, the system adopts a layered architecture design, constructing a complete closed loop for intelligent driving training assistance and active safety control through the organic collaboration of the perception layer, decision planning layer, control execution layer, and interaction layer. The perception layer, as the foundation for the system's data acquisition, integrates multi-source sensing devices such as LiDAR, visual sensor arrays, millimeter-wave radar, ultrasonic radar, high-precision positioning modules, and driver monitoring modules. It is used to acquire in real time the 3D point cloud and semantic information of the vehicle's surrounding environment, the vehicle's centimeter-level position and attitude data, and behavioral characteristics such as the driver's facial expressions and eye movements. The system integrates LiDAR and visual sensors via a post-fusion strategy, combining visually recognized object semantics with precise spatial coordinates measured by LiDAR to generate semantically tagged 3D obstacle data. Millimeter-wave radar, positioned at the four corners of the vehicle, monitors blind spots and assists lane changes, calculating the collision time of vehicles approaching from behind in real time. Ultrasonic radar covers the near-field area of ​​the front and rear bumpers, providing collision protection within the last 30 centimeters. The positioning module integrates RTK-GNSS, a dual-antenna direction-finding unit, and an inertial measurement unit, providing centimeter-level absolute position and stationary heading angle when satellite signals are good. In areas with signal obstruction, it automatically switches to SLAM positioning mode based on LiDAR point cloud matching or visual odometry to ensure continuous and reliable positioning information. The driver monitoring module captures the trainee's facial features through an in-vehicle camera for fatigue detection, eye tracking, and emotion analysis. These multi-source data, synchronized spatiotemporally, collectively construct a digital teaching environment encompassing the vehicle itself, surrounding obstacles, and the training ground, providing comprehensive and accurate input information for the decision-making and planning layer.

[0030] The decision-making and planning layer connects with the perception layer, serving as the core processing unit of the system and responsible for the fusion processing and intelligent decision-making of massive amounts of perception data. This layer first performs spatiotemporal alignment and fusion of the received environmental data and its own state data to construct a dynamic scene model including lane lines, obstacles, and virtual electronic fences. Based on this, the built-in expert system engine initiates the core analysis logic: the ideal trajectory generation module pre-stores demonstration trajectory data from experienced instructors, covering standard driving routes and timing for each test item; the deviation analysis module uses dynamic time warping or Friesian distance algorithms to compare the real-time trajectory generated by the student's actual operation with the ideal trajectory frame by frame, calculating a quantified trajectory deviation value; and the tiered arbitration logic module triggers different levels of intervention strategies based on the threshold range of the deviation value, combined with safety indicators such as collision time and speeding trends. When the trajectory deviation is less than the monitoring threshold, the system is in Level 0, monitoring only, recording data but not intervening. When the deviation is greater than or equal to the monitoring threshold but less than the first preset threshold, and there is no risk of collision, the system enters Level 1, issuing visual or auditory prompts through the interaction layer. When the deviation exceeds the first preset threshold or speeding or lane-crossing trends are detected, the system enters Level 2, limiting throttle output or applying steering wheel counter-torque. When the deviation reaches the second preset threshold, the collision time is less than the safety threshold, or the student accidentally presses the accelerator, the system enters Level 3, forcibly taking over vehicle control and performing emergency braking or steering avoidance. Through this hierarchical decision-making logic, the decision-making and planning layer maximizes the student's learning initiative while ensuring safety, achieving a dynamic balance between teaching tolerance and safety bottom line.

[0031] The control execution layer connects with the decision-making and planning layer, responsible for translating safety control commands into specific vehicle actions. This layer constructs three major execution subsystems—braking, steering, and power—through non-destructive steer-by-wire modifications to traditional training vehicles. The steer-by-wire unit uses a co-driver auxiliary brake motor or an electro-hydraulic braking system, capable of responding to emergency braking commands within milliseconds while preserving the mechanical connection of the physical pedal and assigning it the highest priority, ensuring manual braking intervention is possible in any situation. The steer-by-wire unit analyzes the original vehicle's electric power steering protocol or adds an external steering motor to directly send angle or torque commands, achieving automatic steering correction and force feedback teaching functions. When the student deviates from the lane, the steering wheel can provide a reverse torque prompt to return to the correct direction. The power-by-wire unit includes an electronic throttle signal interceptor and a clutch travel sensor. The former limits maximum throttle output and blocks accidental throttle input signals, while the latter monitors the clutch pedal position and release speed in real time, automatically instructing the steer-by-wire to replenish throttle or issuing voice prompts through the interaction layer when the system anticipates an impending engine stall. Through the coordinated work of these three execution units, the control execution layer transforms abstract safety commands into precise vehicle dynamics responses, constructing a comprehensive active safety protection system from the three dimensions of braking, steering, and power.

[0032] The interaction layer connects with the decision-making and planning layer, serving as a bridge for human-computer interaction. It transforms the system's intelligent decisions into teaching information that students can intuitively perceive and understand, while simultaneously receiving student feedback to form a closed loop. This layer's core components include an augmented reality head-up display (ARH) and an intelligent voice assistant. The ARH projects virtual teaching information precisely onto the driver's head-up view through the windshield, seamlessly integrating it with the real road scene: dynamically laid green virtual guide lines indicate standard driving trajectories, red hazard markers superimposed on obstacle locations enhance the student's risk perception, and real-time animations of steering wheel and wheel angles help students develop a concrete understanding of the relationship between operation and vehicle response. The intelligent voice assistant, based on a natural language processing engine, supports multi-turn continuous dialogue. It can answer students' questions about the timing of operations based on current environmental data, such as "Can I change lanes now?", and proactively explain the reasons after a penalty is incurred, such as "You crossed the line because you turned the steering wheel 0.5 seconds too late. Next time, please start turning the steering wheel when your shoulder is level with the marker." By deeply integrating visual guidance and voice interaction, the interaction layer transforms abstract driving knowledge into intuitive visual symbols and natural language dialogue, significantly reducing the cognitive load on learners and improving the learning experience and teaching effectiveness. The four interconnected layers—data acquisition at the perception layer, intelligent analysis at the decision-making and planning layer, precise operation at the control and execution layer, and human-machine communication at the interaction layer—together constitute the complete technical solution for this intelligent driving training assistance and safety control system.

[0033] In this embodiment, the sensing layer includes: LiDAR is used to acquire three-dimensional point cloud data and distance information of surrounding obstacles; The visual sensor array, including surround-view and forward-view cameras, is used to identify lane lines, traffic signs, and traffic lights. Millimeter-wave radar, positioned at the four corners of the vehicle, is used for blind spot monitoring and lane change assistance; Ultrasonic radar, installed on the front and rear bumpers of the vehicle, is used for near-field collision protection; The positioning module, including an RTK-GNSS receiver, a dual-antenna direction-finding unit, and an inertial measurement unit, is used to acquire the vehicle's centimeter-level absolute position, heading angle, and attitude information in a stationary state, and automatically switch to SLAM positioning mode based on lidar point cloud matching or visual odometry in areas where GNSS signals are blocked. The driver monitoring module is used to capture the driver's facial expressions, gaze direction, and fatigue level.

[0034] Specifically, the perception layer, as the data acquisition foundation of the system, integrates multi-source sensing devices such as lidar, vision sensor group, millimeter-wave radar, ultrasonic radar, high-precision positioning module and driver monitoring module. It is used to acquire in real time the three-dimensional structural information, semantic information, dynamic target information, near-field obstacle information, high-precision position and attitude information of the vehicle itself, as well as the physiological and behavioral characteristics of the driver, providing accurate, redundant and reliable data support for the decision planning layer to build a comprehensive dynamic scene model.

[0035] LiDAR, employing 16- or 32-line mechanical or solid-state radar, is typically mounted on the roof of a vehicle to achieve 360-degree scanning without blind spots. Its core function is to acquire 3D point cloud data and precise distance information of surrounding obstacles. Unlike high-speed autonomous driving scenarios, driver training scenarios, especially the second stage of driving test training, require extremely high near-field perception accuracy, needing to determine the centimeter-level relative positions of wheels with edges, corners, and other targets. LiDAR measures the reflection time of objects by emitting laser beams, providing precise depth information unaffected by changes in lighting conditions, effectively solving the problem of traditional monocular cameras failing in scenarios with weak texture, such as strong backlighting, entering or exiting tunnels, or concrete surfaces. In specific applications, such as reversing into a parking space, LiDAR can accurately detect the 3D position of the corners of the parking space, providing a high-precision spatial coordinate reference for subsequent fusion with visual data.

[0036] The visual sensor array includes surround-view cameras and forward-view cameras. The surround-view cameras typically consist of four fisheye cameras distributed around the vehicle, front, rear, left, and right, used to construct a 360-degree panoramic image and monitor near-field blind spots, providing trainees with a comprehensive visual reference of the vehicle's surroundings. The forward-view cameras are responsible for recognizing semantic information such as lane lines, traffic signs, and traffic lights, providing the necessary traffic rule understanding for driving test road training. At the algorithm level, the system employs a post-fusion strategy, matching and fusing the semantic information of visually recognized objects (such as "cones," "cabin lines," and "stop lines") with precise spatial coordinate information measured by LiDAR, generating 3D obstacle data with semantic labels. This allows the system to not only know "what is there," but also "where it is" and "what it is."

[0037] Millimeter-wave radar is positioned at the four corners of the vehicle, primarily used for blind spot monitoring and lane change assistance during road training for driving test part three. Millimeter-wave radar boasts the advantage of all-weather operation, unaffected by rain, fog, dust, or other adverse weather conditions. It can calculate the distance and speed of vehicles approaching from behind in real time and determine the collision time through a fusion algorithm. This function is particularly important when trainees are changing lanes: when a trainee activates their turn signal to prepare for a lane change, the system integrates millimeter-wave radar data. If the calculated collision time is less than a safe threshold (e.g., 5 seconds), the system will lock the steering wheel and issue a voice warning, "Vehicle approaching from behind, lane change prohibited," preventing reckless actions by the trainee from causing a traffic accident.

[0038] Ultrasonic radar is installed on the front and rear bumpers of vehicles, typically with 8 to 12 sensors, for near-field collision protection within the last 30 centimeters. In low-speed maneuvers such as reversing into a parking space or parallel parking, ultrasonic radar can accurately detect the extremely close distance between the vehicle and curbs, posts, and low obstacles, compensating for the blind spots of lidar and vision sensors in the ultra-close-field area. When the distance is less than a safe threshold, it triggers emergency braking to prevent low-speed collisions.

[0039] The positioning module is the core of the perception layer for achieving high-precision spatial positioning, comprising an RTK-GNSS receiver, a dual-antenna direction-finding unit, and an inertial measurement unit. RTK technology employs carrier phase differential technology, transmitting correction signals from ground base stations to correct GPS / BeiDou positioning errors from meter-level to centimeter-level (1-2 cm), providing a high-precision position reference for electronic fences and automatic assessment. The dual-antenna direction-finding unit utilizes the baseline vector between two antennas mounted on the vehicle roof to directly calculate the vehicle's heading angle when stationary, solving the problem of traditional single-antenna GPS's inability to obtain the vehicle's orientation when stationary, ensuring that the vehicle's direction is correct before starting. The inertial measurement unit includes accelerometers and gyroscopes, outputting the vehicle's acceleration and angular velocity information at high frequencies (above 100Hz), which is then fused with the low-frequency (10Hz) output of the RTK-GNSS using Kalman filtering to output smooth, continuous, and highly dynamic attitude and position information. More importantly, in areas where GNSS signals are blocked, such as tree-lined roads, under overpasses, or inside examination room canopies, the system can automatically switch to NDT / ICP algorithm based on lidar point cloud matching or SLAM positioning mode based on visual-inertial odometry. By matching with pre-collected high-precision map point clouds, it can achieve continuous and reliable positioning in environments without satellite signals, ensuring the uninterrupted operation of electronic fences and teaching evaluation functions around the clock.

[0040] The driver monitoring module captures real-time facial images of the driver using infrared cameras installed near the rearview mirror or on the dashboard. Deep learning algorithms are then used to analyze facial expressions, gaze direction, and fatigue levels. This module has a dual function: at the teaching level, it rigorously detects the "looking back" action required in the driving test (Part 3) through eye-tracking technology. If the student fails to scan the left and right rearview mirrors when passing intersections or changing lanes, the system will deduct points according to the test standards. At the safety and psychological support level, it analyzes facial expressions to identify whether the student is tense, flustered, or fatigued. If extreme tension is detected, the system automatically slows down its speech, plays reassuring messages, and may even suggest stopping for a rest, achieving human-like emotional perception and teaching support. Through the organic integration and collaborative work of these multi-source sensors, the perception layer constructs a digital teaching environment containing semantic information about the vehicle itself, surrounding obstacles, and the training ground, providing a comprehensive, accurate, and real-time input data foundation for the decision-making and planning layer.

[0041] In this embodiment, the control execution layer includes a drive-by-wire chassis modification mechanism: The brake-by-wire unit, using a co-pilot auxiliary brake motor or an electro-hydraulic braking system, is used to receive emergency braking commands and perform active braking. The steer-by-wire unit obtains the electric power steering protocol by parsing the electric power steering protocol or establishing a direct communication connection, and directly sends steering angle or torque commands to achieve automatic steering or force feedback teaching. The drive-by-wire power unit includes an electronic throttle signal interceptor and a clutch travel sensor, used to limit maximum throttle output, block accidental throttle input signals, or automatically replenish throttle when engine stall is anticipated.

[0042] Specifically, the control execution layer, through the modification of traditional training vehicles to be steerable by wire, establishes an execution link connecting the decision-making and planning layer with the vehicle's physical actions, forming the physical foundation for active safety control. For common manual transmission vehicles in driving schools, such as the Santana and Jetta, the system employs non-destructive or minimally invasive modification schemes to transform the braking, steering, and power systems, which originally relied on mechanical transmission, into steerable actuators that can be directly controlled by electronic signals, thereby accurately responding to safety control commands issued by the decision-making and planning layer.

[0043] The brake-by-wire unit, as the most critical safety actuator, employs two implementation schemes: a co-driver auxiliary brake motor or an electro-hydraulic braking system. In the low-cost scheme, a cable-operated brake robot or lever-type brake mechanism is installed in the co-driver's position, using a motor to pull the original vehicle's brake pedal. This scheme is simple to install, requires no modification to the original vehicle's hydraulic system, and fully preserves the physical form of the auxiliary brake, facilitating manual emergency takeover. For newer electric training vehicles, an electro-hydraulic braking system can be used, directly controlling the electronic vacuum booster via the CAN bus. Its response speed is extremely fast, establishing maximum braking force within 150 milliseconds, and it offers excellent braking linearity, enabling smooth deceleration and effectively avoiding the "nose-diving" phenomenon of traditional emergency braking. Crucially, this unit employs a "dual-control" design logic: when the system determines the danger level reaches the strong takeover threshold through graded arbitration, electronic braking is triggered first; however, if the electronic control system fails, the instructor or student can still brake by physically pressing the brake pedal. The mechanical connection always has the highest priority, ensuring that the braking function does not fail under any extreme circumstances.

[0044] The steer-by-wire unit achieves protocol cracking and adaptation to the original vehicle's EPS system by parsing the electric power steering protocol or establishing a direct communication connection. Specifically, after accessing the vehicle's CAN bus and obtaining the communication protocol of the steering control unit, the system can directly send steering angle or torque commands to achieve automatic steering. This design not only enables the vehicle to automatically correct its direction in dangerous situations but also applies a reverse torque through the motor to achieve a "force feedback teaching" effect: when the learner deviates from the lane or is about to cross the line, the steering wheel automatically becomes heavier, providing a tactile prompt to the learner to correct the direction in time, transforming abstract driving perception into concrete operational feel. For some older hydraulic power steering models that are not equipped with EPS, the system adopts an external steering motor solution, adding a gear-driven motor below the steering column. Although the accuracy is slightly lower than the original EPS protocol control, it can still meet the needs of low-speed on-site teaching.

[0045] The drive-by-wire power unit includes an electronic throttle signal interceptor and a clutch travel sensor, primarily designed to address the driving difficulties of manual transmission vehicles. The electronic throttle signal interceptor, installed between the accelerator pedal sensor and the engine ECU, modifies the voltage signal emitted by the pedal sensor in real time, enabling various safety assistance functions: In speed-limited mode, it limits the maximum throttle output voltage to ensure the learner does not exceed the speed limit within designated areas; when the system detects an obstacle ahead and the learner continues to press the accelerator, it automatically triggers a false throttle blocking function, cutting off the throttle signal input; during start-up or hill start, the system anticipates the critical state of impending engine stall and automatically instructs the drive-by-wire throttle to appropriately supplement the throttle, assisting the learner in a smooth start. The clutch travel sensor monitors the clutch pedal position and release speed in real time. Combined with engine speed data, when the system predicts the engine is about to stall based on the clutch release speed and the rate of change in engine speed, in addition to automatically supplementing the throttle, it can also issue a voice prompt of "slowly release the clutch" through the interaction layer, helping the learner gradually develop muscle memory for clutch and throttle coordination. Through the coordinated work of the three execution units mentioned above, the control execution layer transforms the abstract safety commands issued by the decision-making and planning layer into precise vehicle dynamics responses, constructing a comprehensive active safety protection system from the three dimensions of braking, steering, and power.

[0046] In this embodiment, the decision planning layer has a built-in expert system engine, which includes: The ideal trajectory generation module stores demonstration trajectory data from experienced coaches; The deviation analysis module uses dynamic time warping or Frescher distance algorithm to calculate the trajectory deviation between the student's actual trajectory and the ideal trajectory; The tiered arbitration logic module is used to trigger different levels of intervention strategies based on the level of deviation value, and is defined as follows: Level 0: When the trajectory deviation value is less than the monitoring threshold, only monitor, do not intervene; Level 1: When the trajectory deviation value is greater than or equal to the monitoring threshold and less than the first preset threshold, and there is no risk of collision, a visual or auditory prompt is issued through the interaction layer. Level 2: When the trajectory deviation value is greater than or equal to the first preset threshold and less than the second preset threshold, or when overspeeding or crossing the line is detected, the throttle output is limited or a reverse torque is applied to the steering wheel. Level 3: When the trajectory deviation value is greater than or equal to the second preset threshold, or the collision time is less than the safety threshold, or the student is detected to have accidentally stepped on the accelerator, the vehicle control is forcibly taken over and emergency braking or steering is performed to avoid the collision. Among them, the monitoring threshold is less than the first preset threshold and less than the second preset threshold.

[0047] Specifically, the expert system engine built into the decision-making and planning layer is the core decision-making unit of the entire intelligent driving training assistance and safety control system, undertaking the crucial function of transforming the massive amounts of data acquired by the perception layer into teaching guidance and safety control commands. This engine, by simulating the thought process of experienced instructors, constructs a hybrid intelligent decision-making system based on rules and data-driven principles. Internally, it mainly comprises three core components: an ideal trajectory generation module, a deviation analysis module, and a hierarchical arbitration logic module, which together achieve real-time evaluation and precise intervention of the learner's driving behavior.

[0048] The ideal trajectory generation module serves as the evaluation benchmark for the expert system, pre-storing a large amount of perfect demonstration trajectory data completed by experienced instructors. These demonstration trajectories are not simply records of geometric paths, but rather include diverse information such as the vehicle's standard driving routes in different training exercises, the timing of key operational points, and corresponding speeds and steering angles. In practical applications, the system has established corresponding ideal trajectory models for specific subjects such as reversing into a parking space, hill start parking, and curved driving, providing an objective and quantifiable reference standard for subsequent student performance evaluation. The sources of this demonstration data include both data collected from top-performing instructors' actual operations and the standardized requirements for completing the tasks in the driving test syllabus, ensuring the scientific rigor and authority of the evaluation benchmark.

[0049] The deviation analysis module is the core technology for achieving precise teaching feedback. This module dynamically compares the actual operation trajectory of the student, collected in real time by the perception layer, with the demonstration trajectory stored in the ideal trajectory generation module. At the algorithm implementation level, the system employs advanced similarity calculation methods such as dynamic time warping or Friesian distance. Both algorithms effectively handle the nonlinear scaling problem of two trajectories on the time axis, allowing for differences in the speed of operation when students perform the same task, focusing on evaluating the degree of fit of their spatial paths. Through this analysis process, the system can not only calculate the quantitative deviation value between the student's actual trajectory and the ideal trajectory, but also further pinpoint the specific location and type of deviation. For example, in the right-angle turn task, it can determine whether the steering wheel was turned "too early" or "too late," and in the reversing parking task, it can identify whether it is "left-side line violation" or "right-side parking violation," providing precise basis for subsequent targeted teaching.

[0050] The tiered arbitration logic module serves as a bridge between analysis and decision-making within the expert system engine. Based on the trajectory deviation value output by the deviation analysis module and combined with real-time environmental risk assessment, this module employs a four-level progressive intervention strategy to achieve a dynamic balance between teaching tolerance and safety baseline. Specifically, the system sets three key critical points: a monitoring threshold, a first preset threshold, and a second preset threshold. These three points maintain an increasing relationship: the monitoring threshold is less than the first preset threshold, and the first preset threshold is less than the second preset threshold, thus classifying different risk levels.

[0051] At Level 0, when the trajectory deviation is less than the monitoring threshold, it indicates that the student's operation is within an acceptable range of fluctuation. The system remains in a background monitoring state, recording all data but not interfering with the student, giving them ample space for independent practice. This design reflects the concept of teaching tolerance, avoiding interference with the student's normal learning process due to excessive intervention.

[0052] At Level 1, when the trajectory deviation value has reached the monitoring threshold but has not yet exceeded the first preset threshold, and the system determines that there is currently no risk of collision, it indicates that the student's operation has slightly deviated but has not yet constituted a safety hazard. At this time, the system issues visual prompts (such as a color-changing warning on the trajectory on the AR-HUD) or auditory prompts (such as "Please correct your direction slightly to the left") through the interactive layer to gently remind the student to pay attention to the operational deviation and help them develop self-correction awareness.

[0053] At Level 2, when the trajectory deviation exceeds the first preset threshold but remains within the second preset threshold range, or when the system detects a clear tendency for the vehicle to violate regulations such as speeding or about to cross the line, it indicates that the student's operation has deviated significantly and requires more proactive intervention. In this case, the system controls the vehicle speed by limiting the maximum throttle output, or applies a counter-torque to the steering wheel through the steer-by-wire unit, providing a physical feedback to prompt the student to correct the direction. This soft-limiting strategy provides substantial assistance while preserving the student's primary control.

[0054] At Level 3, when the trajectory deviation reaches or exceeds the second preset threshold, or when the system calculates, using millimeter-wave radar data, that the collision time with an obstacle is less than a safe threshold, or when the system detects that the learner has mistakenly pressed the accelerator in a dangerous situation, it indicates an emergency danger state. At this point, the tiered arbitration logic module immediately triggers a strong takeover mechanism, forcibly seizing control of the vehicle. Automatic emergency braking is executed via the brake-by-wire unit, and if necessary, active steering avoidance is performed by the steering-by-wire unit until the danger is completely eliminated. Through this four-level progressive tiered arbitration logic, the system maximizes the learner's learning initiative while ensuring absolute safety, achieving a seamless transition from prompting and guidance to active protection.

[0055] In this embodiment, the expert system engine also includes an adaptive teaching module: This module builds a competency profile based on the trainee's historical training data and dynamically adjusts the level of detail in teaching instructions and the timing of intervention. In beginner mode, frequent and detailed step-by-step instructions and early intervention protection are provided; In skilled mode, streamlined error correction instructions are provided and control transfer conditions are relaxed.

[0056] Specifically, the expert system engine also includes an adaptive teaching module, a core component for achieving human-like and personalized instruction. This module aims to break away from the rigid, one-size-fits-all approach of traditional machine learning, enabling the system to dynamically adjust teaching strategies based on individual student differences, much like a seasoned coach. By continuously collecting and analyzing students' historical training data, this module constructs a multi-dimensional profile of student abilities. Based on this profile, it dynamically adjusts the detail of teaching instructions, output frequency, and the timing of safety interventions, thus achieving a seamless transition and adaptive evolution from "novice mode" to "proficient mode."

[0057] The construction of a learner competency profile forms the data foundation for the adaptive learning module. During each training session, the system not only records operational data such as vehicle trajectory deviation and violation frequency, but also captures the learner's facial expressions, gaze focus, and physiological reactions during operation through the driver monitoring module. For example, the system can identify whether the learner frequently blinks nervously when starting on a slope, or shows hesitation or panic when crossing intersections. This multi-dimensional data, after fusion and analysis, forms a comprehensive profile of the learner's skill level, psychological state, and learning habits. The profile includes not only objective driving skill evaluations (such as clutch control smoothness and steering correction accuracy) but also subjective psychological characteristic labels (such as "easily stressed," "careless," and "quick learner"), providing precise decision-making basis for subsequent personalized teaching.

[0058] The dynamic adjustment mechanism based on competency profiles is the core of adaptive teaching. During training, the system matches the current learner's profile characteristics with a pre-set teaching strategy library in real time, dynamically adjusting three major teaching parameters: the level of detail in teaching instructions, the frequency of output, and the timing of safety interventions. For learners who are new to the vehicle or whose competency profiles indicate "easily stressed," the system automatically switches to novice mode. In this mode, teaching instructions are presented as high-frequency, low-granularity detailed guidance: for example, in the starting phase, the system breaks down the action into step-by-step voice prompts such as "Release the clutch a little... a little more... okay, hold," and "Now slowly accelerate," transforming abstract operational sensations into executable, step-by-step instructions. Regarding safety interventions, the system appropriately relaxes the tolerance for deviations, but the intervention timing is significantly earlier. Even in the low-risk, minor deviation phase, it proactively provides steering wheel vibrations or voice reminders, giving learners a strong sense of psychological security.

[0059] As training sessions increase and the learner's skill profile is updated, the system automatically switches to proficiency mode when it determines that the learner has mastered the basic operating techniques and their mental state is stable. In this mode, the teaching style changes significantly: instruction output is greatly simplified, no longer repeating basic operating steps, but focusing on error-correcting prompts such as "Pay attention to speed" and "Straighten the steering wheel," returning more control to the learner; at the same time, the threshold for safety intervention is correspondingly relaxed, only actively intervening when there is a large deviation or a clear safety risk, giving learners ample room for trial and error and opportunities for self-correction, helping them transition from mechanical imitation to autonomous driving.

[0060] Through the aforementioned adaptive teaching mechanism, the expert system engine is no longer a cold, impersonal judgment machine, but a virtual coach that can sense the student's state, understand their needs, and accompany them on their journey of growth. This fundamentally solves the rigidity of the traditional "one-size-fits-all" teaching method of electronic assistance systems, and significantly improves the student's learning experience and training efficiency.

[0061] In this embodiment, the interaction layer includes an augmented reality head-up display and an intelligent voice assistant: Augmented reality head-up displays are used to project virtual guide lines, hazard markings, and visual animations of wheel corners onto the windshield, achieving a fusion display of information and the real road surface; The intelligent voice assistant supports multi-turn dialogue and can explain the reasons for deductions or answer students' questions about the timing of actions based on current environmental data.

[0062] Specifically, the interaction layer, acting as a bridge between the system and the driver, bears the core function of transforming the teaching instructions generated by the decision-making and planning layer into an intuitive and natural human-computer interaction experience. This layer abandons the single beeping alarm or mechanical voice broadcast mode of traditional electronic assistance systems, and constructs an immersive, anthropomorphic teaching environment through the deep integration of augmented reality head-up display and intelligent voice assistant, enabling learners to obtain intuitive, timely, and explanatory learning feedback in real driving scenarios.

[0063] Augmented reality head-up display (HUD) is a core technology for achieving a "what you see is what you get" learning experience in the interactive layer. This system precisely projects virtual information onto the driver's head-up field of vision through the windshield and merges it pixel-by-pixel with the real road scene, allowing learners to obtain crucial driving information without looking down or shifting their gaze, greatly improving safety during the learning process. Specifically, AR-HUD primarily presents three types of visual teaching information: First, virtual guide lines. The system dynamically projects a green light strip or "ghost car" trajectory onto the road surface. Students only need to control the vehicle to follow the guide line to reproduce a perfect standard driving route. This function is particularly effective in projects with strict trajectory requirements, such as curved driving and right-angle turns. Second, hazard markers. When the perception layer detects pedestrians, cones, or other obstacles, AR-HUD overlays a red highlighted border at their actual spatial location and dynamically displays real-time distance values, enhancing students' risk perception capabilities in a visually impactful way. Third, operational visualization animations. The system displays the corresponding animation relationship between the steering wheel angle and the wheel angle in real time at the edge of the field of vision, helping students establish a concrete understanding of "how much steering wheel to turn" and "how much wheel to turn," which is especially suitable for students who lack sufficient understanding of the steering mechanism in the early stages of learning.

[0064] The intelligent voice assistant forms the language interaction hub of the interaction layer. Its functionality far surpasses that of a traditional command player; it acts as a virtual teaching partner with comprehension and explanation capabilities. Based on a natural language processing engine, this assistant supports multi-turn continuous dialogue. Students can ask questions in a conversational manner at any time during driving, such as "Am I in the right position?", "Can I change lanes now?", and "Why did I get penalized?". Upon receiving voice commands, the system integrates environmental data from the perception layer and deviation analysis results from the decision-making and planning layer in real time to generate context-aware answers. For example, after the system announces "You crossed the line," if the student asks "Why?", the voice assistant can further explain: "During the reversing process, you turned the steering wheel to the right about 0.5 seconds later than ideal, resulting in insufficient distance between the left rear wheel and the corner of the parking space, causing you to cross the left edge line. Next time, please start turning the steering wheel when your shoulder is aligned with the marker." This feedback with causal explanation capabilities allows students to not only know what happened but also why, significantly improving the depth and effectiveness of teaching guidance. Through the visual guidance of AR-HUD and the semantic interaction of intelligent voice, the interaction layer transforms abstract driving knowledge into concrete visual symbols and natural language dialogue, fundamentally eliminating the cognitive gap problem in traditional teaching.

[0065] This embodiment also includes a cloud-based monitoring service platform: The vehicle video stream, dashboard data and location information are uploaded to the cloud in real time via the 5G / V2X communication module. It supports remote safety operators to take over control when the in-vehicle system is unable to handle complex scenarios. The cloud-based monitoring service platform is used to generate learning reports that include trajectory playback, deduction point statistics, and operation smoothness analysis, and uses the collected training data to iteratively optimize the expert system model.

[0066] Specifically, the cloud-based monitoring service platform, serving as the system's data aggregation and capability evolution hub, establishes a high-speed, low-latency two-way data link between the vehicle-mounted terminal and the cloud server through a 5G / V2X communication module, achieving a leapfrog upgrade from single-vehicle intelligence to system intelligence. This platform not only solves the "data silo" problem in the traditional driver training model but also constructs a complete "vehicle-road-cloud" collaborative ecosystem, providing driving school administrators, remote safety officers, and students with multi-dimensional monitoring, intervention, and analysis capabilities.

[0067] At the data uplink level, the system uses the vehicle-mounted 5G / V2X communication module to upload diverse data from the training process to the cloud service platform in real time. The uploaded data includes three core dimensions: first, video stream data, including images of the road ahead, the trainee's operating posture inside the vehicle, and real-time dashboard footage, providing a visual monitoring foundation for remote supervision; second, vehicle dynamic data, including CAN bus data such as vehicle speed, engine speed, throttle opening, braking status, and steering wheel angle, providing raw parameters for subsequent quantitative analysis; and third, high-precision position data, namely centimeter-level trajectory information obtained through the perception layer positioning module, enabling the cloud to accurately reproduce the vehicle's movement status every second within the training area. This massive amount of data is uploaded in real time thanks to the high bandwidth of the 5G network, with latency controlled to the millisecond level, ensuring that the remote monitoring end can obtain real-time images and data almost synchronously with the driving inside the vehicle.

[0068] At the remote monitoring and intervention level, the cloud service platform has constructed a remote safety operator cockpit system that allows one person to manage multiple vehicles. The backend monitoring screen simultaneously displays real-time forward video, in-vehicle video, dashboard data, and the precise location of the vehicles on an electronic map for multiple training vehicles in a multi-window format. When the onboard expert system encounters complex scenarios that it cannot handle independently, such as uncertainties in the system algorithm's judgment of special traffic conditions, or when a trainee is unresponsive for an extended period due to extreme tension, the system can automatically send an assistance request to the cloud. Upon receiving the request, the remote safety operator can take over vehicle control remotely via a driving simulator or remote control terminal, safely removing the vehicle from the complex scenario. Control is then returned to the onboard system after the danger has passed. This design provides the highest level of safety redundancy for the system, ensuring a last line of defense for human intervention in any extreme situation.

[0069] At the data post-processing and application level, the core value of the cloud-based monitoring service platform lies in the automatic generation of learning reports and the continuous iteration of expert system models. After each training session, the cloud server automatically retrieves the training data stored in the database to generate a personalized learning report containing multi-dimensional analysis results, which is then pushed to the trainee's mobile phone via a mobile app. This report not only includes a complete replay animation of the training trajectory but also detailed statistics on the distribution of deduction points for each exam item, operational smoothness analysis curves (such as clutch engagement smoothness and direction correction frequency), and trend charts comparing it with historical training data. This allows trainees to intuitively understand their skill progress trajectory and identify areas for improvement. More importantly, the cloud-based monitoring service platform aggregates massive amounts of error operation data and success case data from hundreds or thousands of trainees and tens of thousands of training hours. These valuable data resources are deeply mined using big data analytics to train smarter predictive models. For example, they can identify specific locations where trainees are most likely to stall or under what psychological states they are most prone to making mistakes. The analysis results are then fed back into the in-vehicle expert system to continuously optimize its ideal trajectory model, deviation analysis algorithm, and adaptive teaching strategies. This enables the entire system to self-evolve, becoming increasingly intelligent and understanding of trainees as it accumulates data. Through the organic integration of these functions, the cloud-based monitoring service platform maximizes the value of each training session, achieving a complete closed loop from data collection and real-time monitoring to in-depth analysis and model iteration.

[0070] This embodiment also includes a high-precision map module and an electronic fence module: The high-precision map module has a pre-stored semantically annotated digital map of the training ground, including lane lines, stop lines, project areas and virtual walls, which is used to compare with real-time perception data to determine the current teaching stage of the vehicle; The electronic fence module uses RTK positioning to determine whether a vehicle has driven out of the preset training area boundary. If it does, it automatically cuts off the throttle and triggers a slow braking stop. It also sets dynamic speed thresholds for different training programs to intervene in speeding.

[0071] Specifically, the system also includes a high-precision map module and an electronic fence module. These two modules together form the digital boundary and spatial semantic understanding foundation of the training ground, providing geospatial support for the expert system's precise teaching and proactive safety control. The high-precision map module pre-stores a digital map of the training ground. This map is not a traditional navigation electronic map, but a high-density point cloud map generated by a data acquisition vehicle equipped with a high-beam laser radar after pre-scanning the ground, and is further annotated with fine semantic details. The annotations cover all geometric and logical information within the training ground, including the precise location and type of lane lines, the spatial coordinates of stop lines, the start and end points of each test item area (such as reversing into a parking space, parallel parking, and curved driving), and virtual walls set up to restrict vehicles from crossing boundaries. These virtual walls are invisible but perceptible spatial boundaries, such as electronic no-entry zones set at the edge of the training area, near ditches, or fixed obstacles. During vehicle operation, the high-precision map module continuously compares the vehicle's real-time position, LiDAR point cloud, and visual recognition results acquired by the perception layer with the pre-stored high-precision map. Through a matching algorithm, it determines the vehicle's precise location within the high-precision map, thereby identifying the specific stage of the training exercise the vehicle is in. For example, the system can identify in real-time that the vehicle has entered the "parking" stage of the "parallel parking" exercise, thus invoking the corresponding teaching logic and evaluation criteria, achieving precise linkage between teaching guidance and site location.

[0072] The electronic fence module is a dynamic safety protection system built on high-precision maps. This module continuously monitors the vehicle's real-time location using RTK centimeter-level positioning technology and compares it with the preset training area boundaries. When the system determines that the vehicle is about to leave the permitted training area, enter public roads, or approach a dangerous area, it immediately activates the boundary protection mechanism: first, it automatically cuts off the power output of the drive-by-wire throttle to prevent the vehicle from accelerating outwards; then, it triggers the drive-by-wire braking unit to perform a gentle braking operation, smoothly stopping the vehicle in a safe position, and issuing a voice warning through the interaction layer. This function effectively prevents the risk of serious safety accidents caused by trainees driving the vehicle off the training ground or into public roads due to operational errors. At the same time, the electronic fence module also has a built-in dynamic speed limit function, setting differentiated speed thresholds for different areas according to the safety requirements of different training programs and the regulations of the driving test syllabus. For example, in complex operating areas such as right-angle turns and reversing into a parking space, the system limits the maximum speed to within 5 km / h; in the straight driving sections of the third subject, the speed limit can be relaxed to 40 km / h. The system monitors vehicle speed in real time via a drive-by-wire throttle unit. If the current speed exceeds the dynamic speed limit threshold of the area, it prioritizes reducing power output through an electronic throttle interceptor to limit the speed. If the speed is still not effectively controlled, it triggers active braking intervention to ensure the training process remains within a safe speed range. Through spatial semantic understanding of high-precision maps and dynamic boundary protection via electronic fences, the system constructs an invisible digital safety barrier, fundamentally eliminating the risk of spatial boundary violations and speeding loss of control due to student misoperation.

[0073] This embodiment also includes a failure protection mechanism: It features hardware redundancy design, and key sensor signals are input via dual channels; A heartbeat monitoring mechanism is provided. When the communication between the central computing unit and the underlying controller is interrupted, the underlying controller automatically executes the safety parking logic of cutting off power, applying maximum braking and activating hazard lights. Physical emergency stop buttons are provided inside the vehicle, outside the vehicle, and remotely. Pressing the button will directly cut off the power to the actuator and trigger the mechanical brake.

[0074] Specifically, the system also features multiple redundant fail-safe mechanisms, which serve as the last line of defense to ensure the "robot coach" can land safely in extreme and abnormal situations. Since the system directly controls the vehicle's braking, steering, and power actuators, any software malfunction, communication interruption, or hardware failure could lead to loss of vehicle control. Therefore, the fail-safe mechanism is designed according to the fundamental principle of "fail-safety," employing hardware redundancy, heartbeat monitoring, and physical emergency stop as triple safeguards to ensure that the vehicle can automatically enter a safe mode or be manually stopped in any abnormal state.

[0075] At the hardware design level, the system employs a dual-input redundancy design for critical sensor signals. Specifically, two independent physical channels are set up for braking command signals, vehicle speed signals, and critical communication links between the central computing unit and the lower-level controller. When the primary channel fails due to line faults, signal interference, or component damage, the system can seamlessly switch to the backup channel within milliseconds, ensuring uninterrupted transmission of critical signals. This hardware redundancy design effectively overcomes the risk of complete system failure due to a single point of failure, significantly improving the system's reliability and fault tolerance.

[0076] At the communication monitoring level, the system establishes a heartbeat monitoring mechanism. The central computing unit and the various actuators of the drive-by-wire chassis maintain high-frequency "heartbeat signal" interaction; that is, both sides send handshake data packets of a preset format at a fixed frequency to confirm that the other is in normal working condition. When the lower-level controllers fail to receive heartbeat signals from the central computing unit multiple times consecutively, it determines that the communication link is interrupted or the central computing unit has crashed. At this time, the lower-level controllers no longer wait for instructions from higher levels but automatically trigger preset safety stopping logic: immediately cutting off engine power output through the drive-by-wire power unit, simultaneously instructing the drive-by-wire braking unit to apply maximum braking force to smoothly bring the vehicle to a stop, and automatically activating hazard warning lights to warn surrounding vehicles and pedestrians. This design ensures that even if the central computing unit, which is the "brain" of the system, completely fails, the lower-level actuators, which are the "cerebellum," can still independently complete an emergency safety stop, avoiding the risk of vehicle loss of control due to system crashes.

[0077] At the manual intervention level, the system features physical emergency stop buttons inside the vehicle, outside the vehicle, and remotely, forming a multi-dimensional network for manual emergency intervention. The in-vehicle emergency stop button is prominently located on the passenger side and center console for easy access by instructors or students in emergencies. The external emergency stop button is located on the vehicle's exterior, such as the trunk lid or side of the vehicle, allowing off-site safety personnel to operate it directly when approaching the vehicle. The remote emergency stop button is integrated into the cloud-based monitoring service platform's backend interface; remote safety personnel can trigger an emergency stop by clicking a virtual button when an anomaly is detected on the monitoring screen. Pressing any of these physical emergency stop buttons directly cuts off the power supply to all actuators and triggers the mechanical braking mechanism, forcing the vehicle to stop in the most direct physical way. This purely hardware-based emergency shutdown pathway, independent of the software system, provides the highest level of security for the system, ensuring that manual intervention can be achieved in the event of any software logic anomalies, communication blockages, or algorithmic misjudgments. Through a triple failure protection design of hardware redundancy, heartbeat monitoring, and physical emergency stop, the system constructs a complete safety net from proactive prevention and automatic detection to manual intervention, completely eliminating safety concerns when applying the intelligent teaching system in real training scenarios.

[0078] Example 2

[0079] like Figure 2 As shown, this embodiment provides a driving training assistance method based on the intelligent driving training assistance and safety control system described in Embodiment 1, including the following steps: S1. Load the pre-built digital twin model of the training site based on multi-source sensor fusion, and load the rule engine for the corresponding examination item; Specifically, before the system starts and enters the teaching and training mode, the first step is to load a pre-built digital twin model of the training site based on multi-source sensor fusion technology, and simultaneously load the rule engine for the examination items corresponding to this training. The essence of this step is to establish a high-precision spatial reference benchmark and behavioral evaluation standard for the entire intelligent assisted teaching system, which is the fundamental prerequisite for the effective functioning of all subsequent real-time perception, deviation analysis and teaching intervention functions.

[0080] The digital twin model is not built in real-time online. Instead, it is generated in the early stages of system deployment by using a specialized data acquisition vehicle equipped with a high-beam laser radar, a high-precision inertial measurement unit, and an RTK-GNSS receiver to conduct a full-coverage scan and measurement of the training area. The data is then generated offline through data processing and semantic annotation. During the data acquisition process, the laser radar obtains 3D point cloud data of all stationary elements within the area, including the precise location and geometry of lane lines, parking space lines, curbs, markers, buildings, and fixed obstacles. The high-precision positioning system ensures that all point cloud data are assigned absolute geographic coordinates with centimeter-level accuracy. The collected raw point cloud data is then stitched, denoised, and optimized to form a high-precision base map. Based on this, manual or automated semantic annotation is performed, assigning each element in the base map a logical meaning related to teaching and safety: for example, a specific line is labeled "left parking space line for reversing into a parking space," a specific area is labeled "parallel parking area," and the physical boundaries or hazardous areas of the training area are labeled "virtual electronic fence." The resulting digital twin model is not only a digital replica of the physical world of the site, but also an intelligent spatial database rich in pedagogical semantics, enabling the system to understand the precise location of the vehicle and the corresponding training items and rule requirements.

[0081] Loading the rule engine for the corresponding exam item injects dynamic behavioral evaluation logic into the system based on the spatial context provided by the digital twin model. The rule engine encapsulates the evaluation criteria, operating procedures, and safety thresholds for different training items (such as reversing into a parking space and hill start in Driving Test 2, or straight driving and lane changing in Driving Test 3). For example, the "hill start" rule includes the required distance error range between the front bumper and the stop line, the distance requirement between the vehicle body and the edge line, and the trigger threshold for detecting vehicle rollback. The "crossing an intersection" rule includes the logical judgment that trainees must complete left and right observation actions within a specified distance and correctly use the turn signals. After loading the rule engine, the system combines the static spatial information from the digital twin model with dynamic behavioral norms, enabling it to judge in real time whether each of the trainee's operations conforms to the standards during subsequent training, and accurately identify the specific violation type and reason for deduction when deviations occur. Through this pre-built model and pre-loaded engine, the system achieves a leap from "perceiving spatial location" to "understanding scene rules," laying a solid foundation for providing accurate and human-like teaching guidance.

[0082] S2. Real-time collection of driving operation data and environmental data from trainees, and calculation of the deviation between the vehicle's current position and the ideal trajectory; Specifically, after the system completes the loading of the digital twin model and rule engine, it immediately enters the real-time training and monitoring phase. Its core task is to continuously collect driving operation data and environmental data from the trainee through the perception layer, and then have the deviation analysis module of the decision-making and planning layer calculate the quantitative deviation between the vehicle's current position and the ideal trajectory in real time. This process forms the data foundation for achieving precise teaching feedback and tiered safety intervention. Essentially, it dynamically compares the trainee's real-time driving performance with the "gold medal coach" standard built into the expert system, thereby identifying subtle deviations and potential risks in the operation.

[0083] At the data acquisition level, the multi-source sensors in the perception layer continuously output data streams at high frequency and low latency. LiDAR outputs hundreds of thousands of 3D point cloud data points per second, accurately depicting the distribution and distance of obstacles around the vehicle; the vision sensor group acquires image data at a rate of over 30 frames per second, using a deep learning model to identify lane line positions, traffic sign status, and traffic light colors in real time; millimeter-wave radar continuously scans the four corners of the vehicle, outputting the relative speed and distance of vehicles approaching from behind; ultrasonic radar detects the distance between the front and rear bumpers and near-field obstacles at millisecond intervals; the positioning module integrates data from RTK-GNSS and inertial measurement units, outputting the vehicle's current centimeter-level precise coordinates, heading angle, pitch angle, and yaw rate at a frequency of over 100 Hz; simultaneously, it reads the student's input in real time via the CAN bus, including steering wheel angle, accelerator pedal opening, brake pedal status, clutch position, and gear information. These heterogeneous data streams, after spatiotemporal synchronization and coordinate transformation within the perception layer, form a dynamic scene description in a unified coordinate system, containing three core pieces of information: the vehicle's state vector, a list of environmental targets, and the student's operation commands.

[0084] At the deviation calculation level, the deviation analysis module of the decision-making and planning layer compares the real-time received vehicle position information with the pre-stored ideal trajectory in the ideal trajectory generation module frame by frame. The ideal trajectory is not a single spatial curve, but a four-dimensional trajectory set containing a time dimension, recording the standard speed, heading angle, and timing of the operation at each spatial point when an experienced instructor perfectly completes the project. The deviation analysis module first finds the closest matching point to the current vehicle position on the ideal trajectory using nearest neighbor search or interpolation algorithms, and then calculates the Euclidean distance between the two as the lateral position deviation; simultaneously, it compares the current vehicle speed with the standard vehicle speed at the matching point to calculate the speed deviation; and compares the current heading angle with the standard heading angle to calculate the angle deviation. For time-sensitive operations, such as the timing of starting on a slope or turning the steering wheel at a right angle, the system also uses a dynamic time warping algorithm to flexibly align the student's actual operation sequence with the ideal operation sequence, thereby accurately judging whether the operation is "too early" or "too late". For example, in the reverse parking exercise, the system not only calculates the real-time distances between the four corners of the vehicle and the edge of the parking space, but also uses a predictive algorithm to estimate whether the trajectory will collide with the parking space edge in the next few seconds if the current operating trend continues, thus achieving proactive deviation warnings. Finally, the deviation analysis module outputs a multi-dimensional deviation vector, including position deviation, speed deviation, angle deviation, and operation timing deviation, and sends these quantitative results to the tiered arbitration logic module as the core input for subsequent intervention level determination. Through this real-time data collection and dynamic comparison mechanism, the system analyzes every operation performed by the trainee under a digital microscope, transforming the originally vague "feeling" into precise "data," providing solid data support for subsequent humanized teaching and proactive safety intervention.

[0085] S3. Provide human-like teaching guidance through AR-HUD and voice assistant based on deviation level and potential risk; Specifically, after the decision-making and planning layer completes the deviation analysis and determines the deviation level and potential risks, the interaction layer undertakes the core function of transforming the abstract quantitative results into teaching information that trainees can intuitively perceive and understand. This transformation process is not a simple one-way instruction broadcast, but rather, through the deep integration of augmented reality head-up displays and intelligent voice assistants, it constructs a multimodal teaching feedback system with anthropomorphic characteristics. This allows trainees to receive both visual guidance and verbal explanations in real-time in real driving scenarios, thereby achieving timely perception and accurate correction of operational deviations.

[0086] When the deviation level output by the tiered arbitration logic module is Level 1, the system determines that the student's operation has a slight deviation but poses no safety risk. At this point, the interaction layer primarily provides suggestive feedback. The augmented reality head-up display presents guidance information on the windshield using soft visual elements: for example, when the student slightly deviates from the ideal trajectory while driving on a curve, the originally static green virtual guide line begins to highlight in a slow, pulsating manner, with the color of the guide line gradually changing from green in the center to yellow at the edge, intuitively indicating the direction the student needs to move closer to the center line; simultaneously, a faint yellow halo is projected onto the edge of the windshield on the side where the vehicle is about to approach the lane edge, triggering the student's instinctive attention through edge vision without requiring the student to look down at the instrument panel. The intelligent voice assistant issues concise prompts in a gentle and natural voice, such as "correct the direction slightly to the left," with a steady tone and clear instructions, providing necessary guidance while avoiding excessive interference with the student's independent operation.

[0087] When the deviation level rises to Level 2, the system detects a significant deviation or violation trend in the student's operation, such as continuous driving over the line or significant speeding. The interaction layer activates a more enhanced guidance and warning mechanism. The virtual guide line in the augmented reality head-up display turns a striking orange-yellow, its pulsation frequency increases, and a dynamic arrow is superimposed on the deviation side to indicate the correct direction. Simultaneously, a magnified digital warning is projected in the speedometer area to remind the student that they are currently speeding. More importantly, the system applies a slight counter-torque through the steer-by-wire unit, creating gentle resistance on the steering wheel, providing a tactile prompt for the student to straighten the wheel, achieving coordinated feedback from three modalities: visual, auditory, and tactile. The intelligent voice assistant's tone shifts to a more serious yet restrained warning, such as "Speed ​​too high, please slow down" or "About to cross the line, please correct to the right," with concise and powerful instructions designed to draw the student's attention.

[0088] When the deviation level reaches Level 3, the system determines that there is an emergency safety risk, such as an impending collision with an obstacle or the student accidentally pressing the accelerator. At this point, the interaction layer enters a strong takeover warning mode. The augmented reality head-up display projects a striking red flashing border onto the windshield, covering the entire edge of the field of vision, and overlays a pulsating red 3D mark at the actual location of the obstacle, while displaying text warnings such as "Danger! Emergency Braking," visually informing the student that the system is about to take forceful intervention. The intelligent voice assistant issues a final warning in a rapid but clear voice, such as "Attention! The system will take over braking," and then the system executes emergency braking. Even during the strong takeover, the interaction layer maintains communication with the student. After the danger has passed, the voice assistant will proactively explain the reason for the intervention, such as "The obstacle ahead was too close, and the system has automatically braked," helping the student understand the event and learn from it.

[0089] In another dimension of anthropomorphic teaching, the interaction method of the intelligent voice assistant itself embodies deep anthropomorphic design. Based on a natural language processing engine, this assistant allows learners to initiate multi-round conversational questions at any time during training. For example, after failing to reverse into a parking space, a learner can ask, "Where did I go wrong?" The system can not only retrieve the deviation analysis results but also answer with explanatory language: "You turned the steering wheel to the right about 0.5 seconds too late, resulting in insufficient distance between the left rear wheel and the corner of the parking space. Next time, please start turning the steering wheel when your shoulder is level with the marker." This feedback, with its causal explanation, allows learners not only to know they "made a mistake" but also to understand "where the mistake was" and "how to correct it." When the driver monitoring module detects that the learner's facial expression shows tension or panic, the system proactively adjusts the voice assistant's speech rate and tone, slowing down the speech, lowering the tone, and playing reassuring phrases such as "Don't worry, we'll take it slowly," while reducing unnecessary prompts and giving learners more space for independent thinking. Through visual guidance from AR-HUD and language interaction from intelligent voice, the system transforms cold, quantitative biases into warm, engaging teaching dialogues. This allows learners to experience the patient guidance and emotional care of a seasoned coach while interacting with the machine, fundamentally eliminating the rigidity and mechanical feel of traditional electronic teaching systems and significantly improving the learning experience and teaching effectiveness.

[0090] S4. When it is detected that the student's operation is about to cause a safety accident or violate the examination rules, the vehicle will be automatically controlled according to the hierarchical arbitration logic until the danger is eliminated. Specifically, when the hierarchical arbitration logic module of the decision-making and planning layer determines, through real-time deviation analysis and safety monitoring, that the student's current operation is about to cause a safety accident or violate examination rules, the system will automatically intervene to control the vehicle according to the preset intervention level until the danger is completely eliminated. This process is the core link in achieving active safety protection. Its essence is that at the critical moment when the student's operation fails, the system takes over vehicle control with a millisecond-level response speed, and rescues the vehicle from the dangerous state through precise braking, steering, or power intervention, while ensuring that the intervention process is smooth and controllable, avoiding secondary risks caused by excessive system intervention.

[0091] The triggering conditions for intervention control cover two core scenarios: the first is scenarios where a safety accident is about to occur, including situations where the collision time with an obstacle in front or to the side, calculated by the fusion of millimeter-wave radar and vision, is less than a safety threshold (e.g., 1.5 seconds); a trainee accidentally pressing the accelerator in a dangerous situation, causing the vehicle to accelerate towards an obstacle; the vehicle is about to collide with a corner of a parking space or an obstacle behind it while reversing; and the vehicle deviates from the road surface or goes beyond the boundary of the training area while driving. The second is scenarios involving serious violations of examination rules, including speeding to a dangerous level, forcibly changing lanes in a prohibited lane-changing area, running a red light, or ignoring a stop line. When any of the above conditions are triggered, the tiered arbitration logic module immediately raises the intervention level to Level 3 (strong takeover level) and generates corresponding safety control commands.

[0092] In an emergency braking intervention scenario, taking the example of a student accidentally accelerating towards a stationary vehicle during road training in Driving Test Part 3, the system uses forward-facing millimeter-wave radar and visual sensors to monitor the rapidly decreasing collision time. When the Time To Chance (TTC) is less than the safe threshold of 1.5 seconds and the student continues to press the accelerator pedal deeply, the tiered arbitration logic module instantly triggers a strong takeover command. This command is simultaneously sent to the brake-by-wire unit and the power-by-wire unit via the high-speed CAN bus: the electronic throttle signal interceptor in the power-by-wire unit immediately cuts off the accelerator pedal signal output, so the engine no longer responds to acceleration commands regardless of how the student presses the accelerator; the electro-hydraulic braking system or the passenger brake motor in the brake-by-wire unit establishes maximum braking force within 150 milliseconds, smoothly stopping the vehicle with a preset optimal deceleration curve, avoiding the student lurching forward or a rear-end collision due to sudden braking. During braking, the AR-HUD in the interaction layer projects a flashing red border and the text "Emergency Braking" on the windshield, and the intelligent voice assistant simultaneously issues a clear prompt that "the system has taken over braking," allowing the student to be aware of the current status in a timely manner.

[0093] In an emergency steering and avoidance scenario, taking the example of a vehicle about to collide with the corner of a parking space due to the student turning the steering wheel too late while reversing into a parking space, the system uses lidar and ultrasonic radar to monitor in real time that the distance between the left rear side of the vehicle and the corner of the parking space is less than the safety threshold, and the collision is predicted to be unavoidable based on the current trajectory. After the tiered arbitration logic module triggers strong takeover, the steer-by-wire unit immediately sends precise steering angle commands to the electric power steering system according to the preset avoidance path, automatically correcting the steering wheel to the right by a certain angle to move the vehicle away from the corner of the parking space; at the same time, the brake-by-wire unit applies appropriate braking force to reduce the vehicle speed, ensuring a smooth and controllable avoidance process. Once the distance between the vehicle and the corner of the parking space returns to a safe range, the system automatically releases steering control and, through a voice prompt "Danger has been eliminated, please continue practicing," returns the vehicle to the student for operation.

[0094] In complex scenarios involving multi-sensor fusion, such as when a vehicle suddenly cuts in at high speed from a blind spot during training, the system detects the risk of a side collision through the fusion of millimeter-wave radar and surround-view cameras, triggering a strong takeover command. At this point, the system may simultaneously execute a combined intervention of braking deceleration and steering avoidance: the brake-by-wire unit applies partial braking force to reduce vehicle speed, buying more time for avoidance; the steering-by-wire unit calculates the optimal avoidance path based on the obstacle's trajectory and the vehicle's state, performs a small steering maneuver, avoids the hazard, and then automatically returns to the correct direction. The entire process is completed within hundreds of milliseconds, far faster than the reaction time of a human driver.

[0095] It is worth emphasizing that the system consistently adheres to the principle of "smooth intervention and safe exit" throughout the entire intervention process. Braking force is dynamically adjusted according to the level of danger, avoiding abrupt "nodding" braking that could cause discomfort to the learner or loss of vehicle control. Steering intervention is applied with progressive torque, allowing the learner to perceive the system's corrective intent rather than a sudden takeover of steering. Once the hazard is successfully avoided and the vehicle is back to a safe state, the system does not immediately relinquish control completely. Instead, it continuously confirms the safety of the surrounding environment through the perception module. After confirming there are no further risks, the tiered arbitration logic module gradually reduces the intervention level from Level 3 back to Level 0 or Level 1, providing voice prompts such as "Danger has been eliminated, please continue practicing" or "Control has been returned," along with the gradual disappearance of the prompt icon on the AR-HUD, ensuring the learner clearly perceives the smooth transfer of control. After each intervention, the system records complete intervention process data in the cloud, including the triggering reason, intervention actions, vehicle response, and subsequent student actions. This data is used to generate a safety event analysis in the learning report. After training, a voice assistant briefly explains the intervention reasons to the student, helping them understand the dangerous scenario and avoid similar mistakes from recurring. Through this complete "monitoring-judgment-intervention-recovery-explanation" closed loop, the system maximizes student learning opportunities while ensuring absolute safety, achieving an organic unity of proactive safety protection and educational value.

[0096] S5. After the training is completed, upload the data to the cloud to generate a personalized analysis report and update the trainee's competency profile.

[0097] Specifically, once a complete training course concludes and the student parks and turns off the vehicle, the system automatically triggers the post-training processing flow. Its core task is to upload the massive amounts of raw data collected during the training to a cloud-based monitoring service platform. After in-depth analysis and data mining, a personalized learning report is generated for the student, and the student's competency profile in the cloud database is updated simultaneously. This step is not only a summary of a single training session but also a crucial step in enabling the system to achieve continuous evolution and a closed loop of personalized teaching, maximizing the value of each driving practice session.

[0098] At the data upload level, the vehicle terminal uses a 5G / V2X communication module to package and encrypt the diverse data generated during this training session before uploading it to the cloud server. The uploaded data packets are massive in size and rich in structure, containing three core dimensions: First, video stream data, including road condition videos captured by the forward-facing camera, panoramic images synthesized from the surround-view cameras, and in-vehicle facial videos of the trainees recorded by the driver monitoring module. This video data undergoes keyframe extraction and compression processing, preserving a complete record of the training process while controlling upload bandwidth usage; second, vehicle dynamic data stream, including CAN bus data recorded with millisecond-level timestamps, such as vehicle speed, engine speed, throttle opening, braking status, steering wheel angle, clutch position, gear information, etc., completely reproducing the details of each trainee's operation; and finally, high-precision trajectory data, namely the centimeter-level latitude and longitude coordinate sequence output by the perception layer positioning module, which corresponds one-to-one with the spatial position in the digital twin model, enabling the cloud to accurately reconstruct the complete driving path of the vehicle within the training ground. This data can be uploaded within seconds of training completion thanks to the high bandwidth and low latency of 5G networks, ensuring that the cloud can process and provide timely analysis results.

[0099] In the process of generating personalized analysis reports in the cloud, after receiving the raw data, the server first calls the big data analysis engine for multi-dimensional in-depth analysis. The trajectory replay module overlays the trainee's real-time driving trajectory with the ideal trajectory in the digital twin model, generating a visual dynamic replay animation. Different colors are used to distinguish perfect trajectories (green), slight deviations (yellow), and serious deviations (red), allowing trainees to intuitively see their performance at each curve and each parking space. The deduction point statistics module, based on the loaded examination rules engine, counts and classifies all violations triggered during training, such as "reversing into a parking space - 3 times", "rolling back after starting on a slope - 2 times", and "changing lanes without using turn signals - 1 time". The deduction distribution for each item is presented in the form of bar charts or pie charts to help trainees quickly identify their weaknesses. The smoothness analysis module quantifies the trainee's fine motor skills. For example, it calculates the smoothness score of clutch engagement by analyzing the correlation between the clutch release speed curve and engine speed fluctuations; it assesses the stability of steering control by analyzing the correction frequency and magnitude of steering wheel angle; and it evaluates the linearity of acceleration by analyzing the rate of change of throttle opening. These analysis results are ultimately integrated into a graphical learning report, pushed to the trainee's mobile phone via a mobile app or WeChat mini-program. The report not only includes data charts but also a summary comment from the voice assistant, such as, "You have made significant progress in reversing into a parking space this week, but there is still a risk of rolling back on the hill start. It is recommended that you focus on the timing of clutch and throttle coordination in the next training session," providing trainees with clear, specific, and actionable improvement suggestions.

[0100] In updating the trainee's competency profile, the cloud-based monitoring service platform integrates the quantitative indicators of the current training session with the trainee's historical data, dynamically updating the trainee's multi-dimensional competency profile in the cloud database. The competency profile is not a static set of labels, but a continuously evolving dynamic data model containing multiple quantitative indicators: the skill dimension covers completion quality scores for each exam item, operational smoothness index, and distribution of common error types; the psychological dimension covers tension index, attention span, and fatigue tendency analyzed through the driver monitoring module; and the learning habit dimension covers training frequency, single training duration, and error correction speed. After each training session, newly generated data is incorporated into the profile using a weighted average, ensuring the profile accurately reflects the trainee's current level and progress trend. For example, if a trainee initially frequently made the mistake of crossing the right line in the reverse parking exercise, after multiple training sessions, the frequency of crossing the right line gradually decreases, while occasional left-side parking errors occur. The error type distribution in the competency profile will dynamically adjust accordingly, allowing the system to more specifically focus on the risk of left-side parking errors in the next training session and trigger teaching prompts in advance at appropriate locations. More importantly, the updated competency profile will serve as input parameters for the expert system's adaptive teaching module, directly influencing the detail of teaching instructions, the timing of intervention, and the intervention threshold in the next training session, thus achieving true personalized instruction. For example, if the system detects that a trainee's anxiety level remains consistently high in their mental profile, it will automatically switch to a gentler novice mode in the next training session, reducing the speaking speed, increasing encouraging prompts, and relaxing the tolerance for some non-critical deviations to help the trainee gradually build driving confidence.

[0101] Through this complete "upload-analysis-report-update" closed loop, the cloud-based monitoring service platform transforms each training session into a data node in the trainee's growth trajectory, enabling trainees to clearly see their progress and allowing the system to accurately understand the trainees' needs. Ultimately, this achieves a leapfrog transformation from experience-driven traditional teaching to data-driven intelligent teaching.

[0102] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. An intelligent driving training assistance and safety control system, characterized in that, include: The perception layer is used to collect data on the vehicle's surrounding environment, the vehicle's own status, and the driver's behavior. The decision planning layer, connected to the perception layer, is used to process the environmental data and its own state data, construct a dynamic scene model, and generate teaching instructions or safety control instructions based on the expert system. The control execution layer, connected to the decision planning layer, is used to receive the safety control commands and execute the vehicle's braking, steering, or power control operations. The interaction layer, connected to the decision planning layer, is used to output the teaching instructions and receive voice or operation feedback from the driver.

2. The intelligent driving training assistance and safety control system according to claim 1, characterized in that, The sensing layer includes: LiDAR is used to acquire three-dimensional point cloud data and distance information of surrounding obstacles; The visual sensor array, including surround-view and forward-view cameras, is used to identify lane lines, traffic signs, and traffic lights. Millimeter-wave radar, positioned at the four corners of the vehicle, is used for blind spot monitoring and lane change assistance; Ultrasonic radar, installed on the front and rear bumpers of the vehicle, is used for near-field collision protection; The positioning module, including an RTK-GNSS receiver, a dual-antenna direction-finding unit, and an inertial measurement unit, is used to acquire the vehicle's centimeter-level absolute position, heading angle, and attitude information in a stationary state, and automatically switch to SLAM positioning mode based on lidar point cloud matching or visual odometry in areas where GNSS signals are blocked. The driver monitoring module is used to capture the driver's facial expressions, gaze direction, and fatigue level.

3. The intelligent driving training assistance and safety control system according to claim 1, characterized in that, The control execution layer includes a drive-by-wire chassis modification mechanism: The brake-by-wire unit, using a co-pilot auxiliary brake motor or an electro-hydraulic braking system, is used to receive emergency braking commands and perform active braking. The steer-by-wire unit obtains the electric power steering protocol by parsing the electric power steering protocol or establishing a direct communication connection, and directly sends steering angle or torque commands to achieve automatic steering or force feedback teaching. The drive-by-wire power unit includes an electronic throttle signal interceptor and a clutch travel sensor, used to limit maximum throttle output, block accidental throttle input signals, or automatically replenish throttle when engine stall is anticipated.

4. The intelligent driving training assistance and safety control system according to claim 1, characterized in that, The decision-making and planning layer has a built-in expert system engine, which includes: The ideal trajectory generation module stores demonstration trajectory data from experienced coaches; The deviation analysis module uses dynamic time warping or Frescher distance algorithm to calculate the trajectory deviation between the student's actual trajectory and the ideal trajectory; The tiered arbitration logic module is used to trigger different levels of intervention strategies based on the level of deviation value, and is defined as follows: Level 0: When the trajectory deviation value is less than the monitoring threshold, only monitoring is performed, without intervention; Level 1: When the trajectory deviation value is greater than or equal to the monitoring threshold and less than the first preset threshold, and there is no risk of collision, a visual or auditory prompt is issued through the interaction layer. Level 2: When the trajectory deviation value is greater than or equal to the first preset threshold and less than the second preset threshold, or when overspeeding or crossing the line is detected, the throttle output is limited or a reverse torque is applied to the steering wheel. Level 3: When the trajectory deviation value is greater than or equal to the second preset threshold, or the collision time is less than the safety threshold, or the student is detected to have accidentally stepped on the accelerator, the vehicle control is forcibly taken over and emergency braking or steering is performed to avoid the collision. Among them, the monitoring threshold is less than the first preset threshold and less than the second preset threshold.

5. The intelligent driving training assistance and safety control system according to claim 1, characterized in that, The expert system engine also includes an adaptive teaching module: This module builds a competency profile based on the trainee's historical training data and dynamically adjusts the level of detail in teaching instructions and the timing of intervention. In beginner mode, frequent and detailed step-by-step instructions and early intervention protection are provided; In skilled mode, streamlined error correction instructions are provided and control transfer conditions are relaxed.

6. The intelligent driving training assistance and safety control system according to claim 1, characterized in that, The interaction layer includes an augmented reality head-up display and an intelligent voice assistant: The augmented reality head-up display is used to project virtual guide lines, hazard markings, and visual animations of wheel corners onto the windshield, achieving a fusion display of information and the real road surface; The intelligent voice assistant supports multi-turn dialogue and can explain the reasons for deductions or answer students' questions about the timing of operations based on current environmental data.

7. The intelligent driving training assistance and safety control system according to claim 1, characterized in that, It also includes a cloud-based monitoring service platform: The vehicle video stream, dashboard data and location information are uploaded to the cloud in real time via the 5G / V2X communication module. It supports remote safety operators to take over control when the in-vehicle system is unable to handle complex scenarios. The cloud-based monitoring service platform is used to generate learning reports that include trajectory playback, deduction point statistics, and operation smoothness analysis, and uses the collected training data to iteratively optimize the expert system model.

8. The intelligent driving training assistance and safety control system according to claim 1, characterized in that, It also includes a high-precision map module and an electronic fence module: The high-precision map module pre-stores a semantically annotated digital map of the training ground, including lane lines, stop lines, project areas, and virtual walls, which is used to compare with real-time perception data to determine the current teaching stage of the vehicle. The electronic fence module uses RTK positioning to determine whether a vehicle has driven out of the preset training area boundary. If it does, it automatically cuts off the throttle and triggers a slow braking stop. It also sets dynamic speed thresholds for different training projects to intervene in speeding.

9. The intelligent driving training assistance and safety control system according to claim 1, characterized in that, It also includes a failover protection mechanism: It features hardware redundancy design, and key sensor signals are input via dual channels; A heartbeat monitoring mechanism is provided. When the communication between the central computing unit and the underlying controller is interrupted, the underlying controller automatically executes the safety parking logic of cutting off power, applying maximum braking and activating hazard lights. Physical emergency stop buttons are provided inside the vehicle, outside the vehicle, and remotely. Pressing the button will directly cut off the power to the actuator and trigger the mechanical brake.

10. A driving training assistance method, based on the intelligent driving training assistance and safety control system according to any one of claims 1-9, characterized in that, Includes the following steps: Load the pre-built digital twin model of the training site based on multi-source sensor fusion, and load the rule engine for the corresponding examination item; Real-time collection of driving operation data and environmental data from trainees, and calculation of the deviation between the vehicle's current position and the ideal trajectory; Based on the level of deviation and potential risks, provide human-like teaching guidance through AR-HUD and voice assistant; When the system detects that a student's operation is about to cause a safety accident or violate the examination rules, it will automatically intervene to control the vehicle according to the hierarchical arbitration logic until the danger is eliminated. After the training is completed, the data is uploaded to the cloud to generate a personalized analysis report and update the trainees' competency profiles.

Images