A method for human-machine co-driving takeover decision-making, stability control, and optimization.

Abstract

This invention belongs to the field of vehicle safety control and human-machine interaction technology, specifically involving a human-machine co-driving takeover decision-making, stability control, and optimization method, including: A. Constructing a driving safety field model to reflect the impact of dynamic environment and road elements on the vehicle and provide real-time basis for takeover decisions; B. Constructing a driver's human factor capability field to assess the driver's takeover capability through multimodal data collection and real-time analysis; C. Dynamic coupling of the safety field and human factor capability field in takeover decision-making; D. Human-machine takeover mode discrimination and control allocation method; E. Takeover scheme screening and optimization, further screening and adjustment through the comprehensive cost function of the optimization algorithm for execution in actual operation. The advantages of this invention are: by real-time assessment of environmental risks, driver capability, and takeover timing, it intelligently controls the handover between the autonomous driving system and the driver, ensuring a safe, timely, and efficient transfer of driving power.

Description

Technical Field

[0001] This invention belongs to the field of vehicle safety control and human-machine interaction technology, specifically relating to a human-machine co-driving takeover decision-making, stability control and optimization method and system that dynamically couples the safety field and the human factor capability field. Background Technology

[0002] Under Level 3 human-machine co-driving conditions, autonomous driving systems can already undertake stable longitudinal / lateral control on structured roads. However, in sudden or complex situations (such as sudden stopping of the target ahead, lane closure obstruction, low-adhesion curves, etc.), the driver still needs to take over within a limited time window. As shown in "Research on Takeover Integrating Driver's Takeover Capability and Driving Risk", existing takeover triggering strategies are mostly based on a single threshold (such as fixed TTC, fixed warning time, or risk classification threshold) and make binary decisions within the rule base. This type of method does not adequately characterize the correlation between factors such as "risk evolution over time", "fluctuations in the driver's immediate ability", and "the reshaping of behavior by HMI prompts", resulting in inconsistent effectiveness of the same threshold for different drivers and under different operating conditions. Taking the study of integrating driving risk and driver ability as an example, the literature points out that in the deterministic region of high and low risk, rules or Bayesian networks can achieve high accuracy. However, once entering the "medium risk-uncertain region", the discrete conditional probability table is difficult to express the nonlinear coupling and trend information of ability and risk, resulting in boundary jitter and insufficient generalization. This causes the decision to repeatedly jump between "takeover / maintain" under the same working condition, making it difficult to stably support the determination of takeover timing.

[0003] The previously reported study, "Characteristics of Driver Takeover Behavior and Road Adaptability in Human-Machine Co-driving Mode," indicates that driver takeover behavior is significantly influenced by human-dependent variables such as attention allocation, situational awareness (SA), reaction time, and adaptation to the road environment. Related research, using eye-tracking attention models and SA-time functions to characterize the behavior before takeover, shows that when a driver quickly switches back to their primary task from a non-driving task, there is a significant time delay between attention shift and SA recovery, which varies with road complexity, curve radius, and gaze shift probability. If takeover is triggered before SA recovery, although the request is perceived promptly, the quality of the takeover is prone to fluctuation. These results suggest that takeover triggering and window allocation should not rely solely on single-point thresholds but should be matched with the individual's current SA and reaction chain parameters.

[0004] Regarding HMI takeover prompt channels and prompt levels, the previously reported research, "Research on Multi-Sensory Interaction Design for Human-Machine Co-driving for Takeover," proposed a multi-sensory (visual / auditory / tactile) strategy to improve takeover accessibility and perceptibility. However, experiments show that excessively strong vibration and acoustic prompts can easily trigger a startle effect, while conversely, they are easily ignored in high-load multimedia environments. Pushing prompts with uniform intensity or fixed levels is difficult to adapt to different drivers' current alertness levels and cognitive loads, resulting in "prompt arrival-action mismatch." Although existing designs propose design principles such as "tiered takeover process" and "timely and effective," they lack a joint closed loop with risk evolution and capability status. The prompt level and timing are still mainly based on preset rules, making it difficult to achieve stable migration for heterogeneous groups.

[0005] Existing Chinese patent CN202010690235.3 discloses a multi-factor fusion method for allocating driving rights in human-machine co-driving. This method collects driver cognitive load, muscle driving ability recovery, and environmental information to calculate the "driving rights to be allocated," and then uses a weighted fusion to obtain the final allocation result. It emphasizes considering multiple factors in parallel to improve takeover safety and smoothness. While this approach is more comprehensive than a single threshold, its core remains a static weighting of multiple indicators: first, it lacks continuous field modeling and trend quantification of driving risks in the spatiotemporal dimension, making it difficult to distinguish risk trajectories with "same value but different potential"; second, the inclusion of driver ability is mainly reflected in point-in-time estimation, and has not yet formed a linkage with the SA recovery-reaction chain time-series parameters; and third, there is no adaptive coupling strategy based on risk-ability state between HMI prompts and control allocation, and the selection of prompt levels and intensity remains too rule-based.

[0006] In summary, existing takeover strategies based primarily on fixed TTC / time thresholds or static weighting are insufficient to fully reflect the coupling mechanism of "risk situation - human capability - HMI guidance": Under medium-risk and boundary conditions, rule / probability models fail to adequately express risk trends and individual capability differences, resulting in inconsistencies ("fast reaction but insufficient perception" corresponds to scenarios where SA has not yet recovered); under uniform thresholds or uniform prompting strategies, individual capabilities may not reach the threshold when the prompt arrives ("prompt has arrived but capability has not reached," see the discussion on decision failure and individual differences in medium-risk areas); even if time seems ample, takeover quality remains unstable if the prompt intensity, channel, and individual state are mismatched ("ample time but unstable takeover quality," see empirical evidence of multi-sensory prompts being startled or ignored).

[0007] Therefore, it is urgent to dynamically couple the spatiotemporal continuous representation of driving risks with the temporal recovery process of driver capabilities, and form a closed loop with the multimodal HMI hierarchical strategy to achieve adaptive optimization of takeover window, prompting strategy and responsibility allocation. Summary of the Invention

[0008] In view of the above-mentioned technical problems and defects, the purpose of this invention is to provide a human-machine co-driving takeover decision-making, stability control and optimization method, which intelligently controls the handover between the autonomous driving system and the driver by real-time assessment of environmental risks, driver ability and takeover timing, so as to ensure safe, timely and efficient completion of the transfer of driving rights, thereby overcoming the shortcomings of the above-mentioned prior art.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] A method for human-machine co-driving takeover decision-making, stability control, and optimization includes the following steps:

[0011] A. Construct a driving safety field model to reflect the impact of dynamic environment and road elements on the vehicle and provide real-time basis for takeover decisions;

[0012] B. Construct a driver human factor capability field, and evaluate the driver's takeover capability through multimodal data collection and real-time analysis. This requires a comprehensive evaluation of the driver's physiological signals, behavioral characteristics, and psychological state information, and derive the driver's takeover capability index through a weighted linear combination model.

[0013] C. The dynamic coupling of the safety field and human capability field in takeover decision-making, including: takeover urgency Calculation and derivation, feasibility index Calculations and safety margins Calculation;

[0014] D. Methods for determining human-machine takeover modes and allocating control rights, including: takeover mode selection, allocation of human-machine control rights, and HMI prompt design;

[0015] E. Takeover scheme screening and optimization: By comprehensively evaluating the takeover mode selection, takeover urgency, takeover feasibility, driver status, and environmental risk factors, the optimal takeover scheme is screened out. The scheme is then further screened and adjusted using the comprehensive cost function of the optimization algorithm for implementation in actual operation.

[0016] As a preferred embodiment of the present invention, step A further includes the following step:

[0017] A1. Acquire the vehicle's status data relative to the environment, including the vehicle's relative position, speed, and acceleration information with respect to surrounding vehicles;

[0018] A2. Obtain static information about the road, including environmental visibility, lane width, road adhesion coefficient, and speed limit information;

[0019] A3. Based on the dynamic information of the self-vehicle and the target vehicle obtained by the vehicle perception layer, a spatiotemporal safety field model is constructed. The spatiotemporal safety field model reflects the risks faced by the self-vehicle based on the relative position, speed, acceleration and target behavior of the target vehicle.

[0020] A4. Improve the safety field model by considering the influence of the static road environment. The formula for calculating the static safety field is:

[0021] ;

[0022] In the formula: This represents the static road field strength. This is the relative distance between the center of the lane and the vehicle. The standard deviation of lane width; Indicator for restricted areas; This is a visibility correction factor, reflecting the impact of visibility on the safety field; For visibility measurement; The weight for centering out of lane; Hard constraint weights;

[0023] A5. Taking into account both the dynamic safety field and the static road safety field, the spatiotemporal comprehensive safety field of the vehicle's location is calculated. The formula for calculating the comprehensive safety field, based on the union of probabilities, is as follows:

[0024] ;

[0025] In the formula: For overall field strength; and The weighting coefficients determine the contribution of risk factors to the overall security field; It is a safety field related to road speed limits, measuring the relationship between the current vehicle speed and the speed limit; Weights for bounded constraints;

[0026] Through a comprehensive safety field, the vehicle can fully assess the safety of its surroundings and provide real-time data for takeover decisions.

[0027] As a preferred embodiment of the present invention, step A3 further includes the following step:

[0028] A3.1. Derive the influence of relative position and relative speed. The safety field assesses risk based on the relative position and speed between the vehicle and the target vehicle. The longitudinal distance between the vehicle and the target vehicle... and lateral distance To improve security, the longitudinal distance is represented using a Gaussian kernel function. and lateral distance The impact on safety;

[0029] A3.2. Derivation of Collision Time The collision time between the target vehicle and the vehicle itself is a key factor in assessing collision risk. The shorter the collision time, the closer the target vehicle and the vehicle itself are, the greater the probability of a collision and the greater the risk faced by the vehicle itself.

[0030] A3.3. Derivation of the impact of speed difference: Relative speed directly affects the degree of threat posed by the target vehicle to the vehicle itself. If the approach is at high speed, the risk of collision increases.

[0031] A3.4. Introduce risk factors. Used to quantify the degree of danger of target behavior; if the target exhibits dangerous behavior, then the risk factor... This amplifies the perceived safety threat posed by the target to the vehicle, increasing the risk factor. This is derived by dynamically evaluating the behavior of the target.

[0032] As a preferred embodiment of the present invention, step B further includes the following step:

[0033] B1. Multimodal data acquisition,

[0034] The system collects multimodal data and operational behavior data of the driver in real time through onboard sensors and physiological monitoring equipment. The multimodal data includes: heart rate variability, skin conductance response, eye tracking and fixation stability; the operational behavior data includes: the force and frequency of the accelerator and brake pedals, and the steering wheel rotation angle. Based on the multimodal data and operational behavior data, the system determines whether the driver is in a normal driving state and whether there are any sudden reactions or abnormal operations.

[0035] B2. Calculation of the driver takeover capability index

[0036] First, by integrating multimodal data and operational behavior data, the takeover capability index is used as a comprehensive measure to reflect whether the driver can successfully take over vehicle control at a specific moment. A weighted linear combination model is used, and the results are normalized by the Sigmoid function so that the takeover capability index TCI value is between [0,1].

[0037] The weighted linear combination model weights physiological signals, behavioral and psychological characteristics to calculate a takeover capability index that reflects the driver's ability to take over. , The closer the value is to 1, the stronger the driver's ability to take over; the closer it is to 0, the weaker the driver's ability to take over.

[0038] B3. Driver reaction time prediction model

[0039] The reaction time is modeled using a Gaussian distribution, where the mean and standard deviation of the reaction time need to be dynamically adjusted according to the driver's current state. The prediction of the reaction time is based on the driver's physiological state, emotional fluctuations, and current driving environment factors.

[0040] By collecting multimodal data from drivers and combining it with Gaussian distribution modeling, driver reaction time can be predicted.

[0041] As a preferred embodiment of the present invention, step C further includes the following step:

[0042] C1. Urgency of Takeover The calculation and derivation,

[0043] Takeover assessment and decision-making are based on the spatiotemporal safety field and the human factor capability field. By convolutional accumulation of hazard intensity and time, and then through accessibility analysis of capability and time window, the urgency of takeover is determined. Feasibility index of takeover With safety margin Based on this, mode selection and thresholding decisions are made. The urgency of takeover is used to measure the cumulative degree of hazard intensity over time within a preset short-term prediction window. The hazard intensity originates from the comprehensive safety field or the risk density equivalent to the comprehensive safety field. ,Will Interpreted as the intensity of danger per unit time, over a length of... Within the short-term horizon, a time kernel is introduced. From union probability to time-weighted integral;

[0044] The current urgency of takeover is calculated by weighting and averaging environmental risks over a preset period of time.

[0045] C2. Feasibility Index The calculation,

[0046] Feasibility Index This measure assesses the probabilistic accessibility of a driver's ability to successfully take over in real time within the safe horizon. The core variables are the capability field, reaction time, and the safe horizon. The feasibility of takeover is assessed by evaluating the driver's ability to take over and their reaction time, taking into account the driver's physiological state, reaction time, and the current takeover window. The feasibility index is calculated accordingly. The higher the value, the more successfully the driver can take over.

[0047] Let the key sub-events for a successful takeover be: awareness / understanding; reaction / action; and environmental accommodation. The probability of success can be expressed as:

[0048] ;

[0049] In the formula: Probability of success; To perceive / understand the probability of success; This represents the probability of a successful reaction. To determine the probability of successful environmental containment;

[0050] Log-odds linearization, followed by Sigmoid mapping, yields:

[0051] ;

[0052] In the formula: Feasibility index; The driver takeover capability index; This refers to the driver's reaction time. This is the safe time window for takeover, and also the current assessment moment; This is to reduce the driver's workload. As a measure of the driver's trust in the system; , , , These are the coupling weights, i.e., the weight coefficients of the behavioral features; To prevent constants with a denominator of zero;

[0053] C3. Safety margin The calculation,

[0054] Safety margin Used to measure the safety of a vehicle's current operation, taking into account both longitudinal braking and lateral stability, the safety margin is divided into longitudinal margin and lateral margin. The longitudinal and lateral margins together determine whether the vehicle has enough safety space to perform a takeover operation under the current driving conditions without a collision or loss of control.

[0055] As a preferred embodiment of the present invention, step C3 further includes the following step:

[0056] C3.1. Calculating the longitudinal margin involves the longitudinal clearance between the vehicle and the vehicle in front, as well as the vehicle's braking capacity. When taking over, the longitudinal margin is estimated using a stopping distance model, assuming the vehicle can quickly decelerate and stop at the current speed.

[0057] The parking distance model consists of two parts: the inertial coasting distance during the reaction time and the braking distance;

[0058] Longitudinal margin It is the gap between your vehicle and the vehicle in front. Subtract the parking distance, then compare it with the reference distance. Compare;

[0059] when When the value is negative, it indicates that the distance between the vehicle in front and the vehicle itself is insufficient, making it impossible to stop safely under the current conditions. In this case, it is necessary to apply the brakes or take over immediately.

[0060] C3.2. Calculate lateral margin. Lateral margin measures the stability of a vehicle when turning or changing lanes. The lateral acceleration of a vehicle is related to the curvature of the road, vehicle speed, and coefficient of friction. The maximum lateral acceleration that a vehicle can withstand is determined by the coefficient of friction of the road and the vehicle's braking capacity. Calculate the maximum value of lateral acceleration. ;

[0061] The actual lateral acceleration of a vehicle when changing lanes or turning. Calculate the lateral margin based on the current driving conditions;

[0062] C3.3. Calculate the overall safety margin. It is determined by the longitudinal margin and lateral margin The calculated minimum value is used to achieve security in both the longitudinal and lateral directions;

[0063] The final safety margin is the minimum of the longitudinal and lateral margins. When the safety margin is negative, minimum risk control or takeover will be triggered.

[0064] As a preferred embodiment of the present invention, step D further includes the following step:

[0065] D1. Takeover Mode Selection

[0066] Feature quantity , , , These represent the urgency of takeover, the feasibility index of takeover, the safety margin, and the safety horizon, respectively, with three candidate models. , For emergency takeover, In order to coordinate the takeover, To take over gradually;

[0067] D2. Allocation of human-machine control.

[0068] The allocation of control must meet the following conditions ∈[0,1], which represents the proportion of driver control. That is, the proportion of system control allocated; with the urgency of takeover As the feasibility index increases, control tends to shift towards the system; Increased control tends to shift towards the driver;

[0069] Human-machine control ratio As a function of the urgency and feasibility of takeover, it is represented by a linear combination of logarithmic odds:

[0070] ;

[0071] In the formula: The logarithmic probability represents the relative advantage of the driver in control. , To control the impact of the takeover feasibility index and urgency on the allocation of control; The minimum threshold for the feasibility index of takeover; The minimum threshold for the urgency of takeover; For bias terms; Feasibility index;

[0072] Use the Sigmoid function to convert log odds into a control weight ratio:

[0073] ;

[0074] In the formula: It is the Sigmoid activation function. Between 0 and 1; The logarithmic probability represents the relative advantage of the driver in control. It is the hyperbolic tangent function;

[0075] D3.HMI design prompts,

[0076] Based on the values ​​of two indicators, takeover urgency and takeover feasibility, the system dynamically determines the level of prompts received by the driver. The takeover prompt levels are designed into four levels: L1: When the takeover urgency is below the set value, the system only prompts the driver to remain alert; L2: When the takeover urgency is above the set value, the system prompts the driver to prepare for takeover; L3: When the takeover urgency is above the set value, the system prompts the driver to prepare for takeover; When the warning level is reached, the system issues a warning to remind the driver that takeover is about to occur; L4: When the urgency of takeover is the highest or the safety margin is lower than the set value, the system forces takeover and issues an emergency warning;

[0077] The intensity and frequency of takeover notifications are dynamically adjusted based on the urgency of the takeover.

[0078] As a preferred embodiment of the present invention, step E further includes the following step:

[0079] E1. Preliminary Screening: After the takeover mode is selected and control allocation is determined, the system will conduct a preliminary screening of takeover plans based on the urgency, feasibility index, safety margin, and environmental factors. By evaluating multiple takeover plans, the system will select the plan that meets the takeover conditions and safety requirements. The core objective of the initial screening is to ensure that the takeover plan meets the following conditions:

[0080] Comprehensive assessment of the spatiotemporal safety field: Determining that the relative risks between the vehicle and the environment are within an acceptable range within the predetermined time window;

[0081] Real-time assessment of driver takeover capability: The driver's takeover capability is sufficient to complete the takeover task;

[0082] Handover quality assurance: Safety margins are used to ensure that there is sufficient safe operating space in both the longitudinal and lateral directions during the handover process;

[0083] The initial screening criteria determine the basic safety and rationality of the takeover task. Takeover plans that do not meet the initial screening criteria will be eliminated, resulting in a set of candidate plans for the initial screening.

[0084] E2. Scheme Optimization and Selection: After initial screening, the system optimizes multiple takeover schemes based on a comprehensive cost function. This comprehensive cost function considers factors such as takeover safety, driver comfort, and operational smoothness. The cost of each takeover scheme is calculated using a weighted summation method, and the optimal scheme is selected. The optimization objective balances multiple dimensions of the takeover as follows:

[0085] Safety: Ensure that no collisions or accidents occur during the takeover process, and maintain the longitudinal and lateral stability of the vehicle;

[0086] Driver comfort: Reduce driver workload and prevent discomfort or tension caused by taking over too early or too late;

[0087] Smoothness of operation: Ensures a smooth transition of vehicle control and prevents overreaction or instability caused by uneven control switching;

[0088] The comprehensive cost function for optimizing the takeover plan is as follows:

[0089] ;

[0090] In the formula: For the takeover plan The overall cost; , , , These are weighting coefficients, representing the weights for controlling safety, comfort, speed stability, and HMI prompts, respectively. A unified timestamp for the perception and control stack; The cost of HMI alerts is measured by weighing the burden of alerts against the costs of false positives and false negatives. To assess the overall risk field strength; The integral of the rate of change of acceleration; Integral for the vehicle's speed; For the desired speed;

[0091] By calculating the comprehensive cost function, the system selects the optimal takeover scheme and passes it to the execution layer for control.

[0092] Advantages and benefits of the present invention:

[0093] (1) This invention proposes a human-machine co-driving takeover strategy based on a safety field. This strategy combines the spatiotemporal safety field of the vehicle's surrounding environment with the human factor capability field. By assessing environmental risks, driver capabilities, and takeover timing in real time, it intelligently controls the handover between the autonomous driving system and the driver, ensuring a safe, timely, and efficient transfer of driving control.

[0094] (2) The present invention adopts a dynamic takeover timing and mode determination method based on the coupling of "safety field × capability field", which combines the spatiotemporal continuous risk field and the driver's capability recovery status to replace the traditional takeover determination mechanism based on fixed threshold.

[0095] (3) This invention adopts a takeover decision mechanism that combines capability recovery progress with takeover window constraints. In the coupled framework, a feasibility index of "safety horizon-capability ratio" is constructed. The takeover window is dynamically constrained in combination with the driver's capability recovery progress to avoid misjudgment of "seemingly in time but not stable control". At the same time, the longitudinal braking achievable limit and the lateral stability limit are measured by the safety margin to ensure the accuracy of the takeover decision.

[0096] (4) This invention adopts a multimodal HMI adaptive grading and continuous control allocation mechanism. It proposes a multimodal HMI adaptive grading and continuous control allocation mechanism, which uses the three-dimensional index surface of "urgency-feasibility-safety margin" to drive the linkage upgrade / downgrade of prompt level and channel combination (visual / auditory / tactile). It also uses continuous human-machine weight mapping to replace discrete switching, so that prompts and responsibility allocation are coordinated in real time according to the scene and individual status, ensuring the effectiveness of prompts and the quality of takeover.

[0097] (5) In addition to the framework of “static weighting”, this invention adopts the joint measurement of spatiotemporal continuous security field and capability time series model, and forms a closed loop through adaptive HMI hierarchical + continuous control rights allocation, which solves the coupling and adaptive optimization problem between “takeover window-prompt strategy-rights and responsibilities allocation”. Attached Figure Description

[0098] Other objects and results of the invention will become more apparent and readily understood with reference to the following description taken in conjunction with the accompanying drawings. In the drawings:

[0099] Figure 1 The following is a logic block diagram provided for this embodiment. Detailed Implementation

[0100] This embodiment provides a method for human-machine co-driving takeover decision-making, stability control, and optimization, specifically including the following steps:

[0101] A. Construct a driving safety field model. In order to accurately assess the safety risks of the vehicle and the surrounding traffic, a safety field model is constructed to reflect the impact of dynamic environment and road elements on the vehicle and to provide real-time basis for takeover decisions.

[0102] A1. Acquire the vehicle's status data in relation to the environment, including information such as the relative position, speed, and acceleration of the vehicle and surrounding vehicles;

[0103] A2. Obtain static road information, including environmental visibility, lane width, road adhesion coefficient, and speed limit information. The collected dataset is represented as follows:

[0104] ;

[0105] In the formula: The relative position of the target vehicle; , , These are the vehicle's speed, acceleration, and heading angle, respectively. The curvature of the road; Speed ​​limits for roads;

[0106] A3. Based on the dynamic information of the self-vehicle and the target vehicle obtained by the vehicle perception layer, a spatiotemporal safety field model is constructed. The spatiotemporal safety field model reflects the risks faced by the self-vehicle based on the relative position, speed, acceleration and target behavior (such as emergency braking, acceleration, lane changing, etc.) of the target vehicle.

[0107] A3.1. Deriving the influence of relative position and relative speed, the core of the safety field is to assess the risk based on the relative position and speed between the vehicle and the target vehicle, and the longitudinal distance between the vehicle and the target vehicle. and lateral distance To improve security, the longitudinal distance is represented using a Gaussian kernel function. and lateral distance Safety impact:

[0108] ;

[0109] In the formula: To unify the timestamps for the perception and control stacks, a monotonic clock is used. To avoid jumps and callbacks; The goal Dynamic safety field strength; and These are weighting coefficients that control the degree of influence of longitudinal and lateral distances on the safety field; and Each is related to the target Longitudinal and lateral relative displacement; and The variance of the distance determines the range of the distance's influence on the safety field;

[0110] A3.2. Derivation of Collision Time The collision time between the target vehicle and the vehicle itself is a key factor in assessing collision risk. The shorter the collision time, the closer the target vehicle and the vehicle are, the greater the probability of a collision, and the increased risk faced by the vehicle. Specifically, as follows:

[0111] ;

[0112] In the formula: It is the difference in longitudinal distance between the target and the vehicle; and These are the speeds of the vehicle and the target, respectively. It is a constant used to prevent the denominator from being zero;

[0113] A3.3. Derivation of the impact of speed difference: Relative speed directly affects the threat level posed by the target vehicle to the vehicle itself, especially when approaching at high speeds, the risk of collision increases dramatically. Therefore, the speed difference term is introduced as follows:

[0114] ;

[0115] In the formula: The weighting coefficient for the effect of speed difference; For speed difference;

[0116] A3.4. Introduce risk factors. Used to quantify the degree of danger of target behavior; if the target exhibits dangerous behavior (such as sudden braking, sudden lane change, etc.), then the risk factor... This amplifies the perceived safety threat posed by the target to the vehicle, increasing the risk factor. This is derived by dynamically evaluating the target's behavior.

[0117] Taking all the above factors into account, the formula for calculating the safe field is as follows:

[0118] ;

[0119] In the formula: and The relative longitudinal and lateral distances between the vehicle and the target vehicle; , , The weighting coefficients of the risk factors represent the contributions of different factors to the safety field; The time to collision required for a collision with the target vehicle. To prevent constants with a denominator of zero; As a risk amplification factor, it reflects the degree of risk of the target behavior; As a risk factor; and The variance of the distance determines the range of the distance's influence on the safety field;

[0120] A4. Further improve the safety field model by considering the influence of the static road environment, such as lane width and speed limit. The calculation formula for the static safety field is as follows:

[0121] ;

[0122] In the formula: This represents the static road field strength. This is the relative distance between the center of the lane and the vehicle. The standard deviation of lane width; This is an indicator for restricted areas (1 for restricted areas, 0 otherwise); This is a visibility correction factor, reflecting the impact of visibility on the safety field; For visibility measurement; The weight for centering out of lane; Hard constraint weights;

[0123] A5. Taking into account both the dynamic safety field and the static road safety field, the spatiotemporal comprehensive safety field of the vehicle's location is calculated. The formula for calculating the comprehensive safety field, based on the union of probabilities, is as follows:

[0124] ;

[0125] In the formula: For overall field strength; and The weighting coefficients determine the contribution of risk factors to the overall security field; It is a safety field related to road speed limits, measuring the relationship between the current vehicle speed and the speed limit; Weights for bounded constraints;

[0126] Through a comprehensive safety field, the vehicle can fully assess the safety of its surroundings and provide real-time data for takeover decisions.

[0127] B. Construct a driver's human factor capability field. Through multimodal data collection and real-time analysis, assess the driver's takeover capability. A driver's takeover capability is not only related to their physiological state and cognitive abilities, but also closely related to factors such as reaction time, attention allocation, and psychological state in specific driving tasks. To determine in real-time whether a driver can successfully take over vehicle control, it is necessary to comprehensively evaluate information such as the driver's physiological signals, behavioral characteristics, and psychological state, and derive the Driver Takeover Capability Index (TCI) through a weighted linear combination model.

[0128] B1. Multimodal data acquisition

[0129] To comprehensively assess the driver's takeover ability, this step collects multimodal data from the driver in real time using onboard sensors and physiological monitoring equipment. The multimodal data includes: heart rate variability (HRV), electrical skin response (EDA), and eye tracking and fixation stability (ATT); the operational behavior data includes: the force and frequency of the accelerator and brake pedals, and the steering wheel rotation angle; by using multimodal data and operational behavior data, it is determined whether the driver is in a normal driving state and whether there are any sudden reactions or abnormal operations.

[0130] B2. Calculation of Driver Takeover Capability Index (TCI)

[0131] To transform multimodal data into a quantifiable takeover capability index, the takeover capability index (TCI) is first integrated with operational behavior data. As a comprehensive measure, the TCI reflects a driver's ability to successfully take over vehicle control at a specific moment. A weighted linear combination model is used, and the results are normalized using the Sigmoid function to ensure the TCI value is between [0,1]. The weighted linear combination model is then used to further refine the TCI index. The calculation formula is as follows:

[0132] ;

[0133] In the formula: To unify the timestamps between the perception and control stack; , , , , For at any time The acquired physiological signals and behavioral characteristics; , , , , These are the weighting coefficients for each behavioral characteristic, reflecting the degree of influence of different characteristics on the takeover capability index; For bias terms; This is the Sigmoid activation function, used to map the output to the range [0,1].

[0134] The weighted linear combination model weights multiple physiological signals, behavioral and psychological characteristics to calculate a takeover capability index that reflects the driver's ability to take over. , The closer the value is to 1, the stronger the driver's ability to take over; the closer it is to 0, the weaker the driver's ability to take over.

[0135] B3. Driver reaction time prediction model

[0136] Reaction time is modeled using a Gaussian distribution, where the mean and standard deviation of the reaction time need to be dynamically adjusted according to the driver's current state. Reaction time is a key indicator for driver takeover control. The prediction of reaction time is based on factors such as the driver's physiological state, emotional fluctuations, and the current driving environment (e.g., traffic density, road conditions). The formula for predicting reaction time is as follows:

[0137] ;

[0138] In the formula: This refers to the driver's reaction time. The average reaction time of a driver is usually related to the driver's physiological signals and the difficulty of the current driving task; The standard deviation is the value related to the driver's current state, reflecting the uncertainty of the driver's reaction time;

[0139] A driver's reaction time is closely related to their physiological and behavioral characteristics; for example, driver fatigue and anxiety levels typically affect their reaction speed. Therefore, by collecting multimodal data from drivers and combining it with Gaussian distribution modeling, we can more accurately predict drivers' reaction times.

[0140] C. The dynamic coupling of the safety field and human capability field in takeover decision-making, including: takeover urgency Calculation and derivation, feasibility index Calculations and safety margins Calculation;

[0141] C1. Urgency of Takeover The calculation and derivation,

[0142] Takeover assessment and decision-making are based on the spatiotemporal safety field and the human factor capability field. By convolutional accumulation of hazard intensity and time, and then through accessibility analysis of capability and time window, the urgency of takeover is determined. Feasibility index of takeover With safety margin Three types of indicators are used to select patterns and make thresholding decisions. The urgency of takeover is used to measure the cumulative degree of danger intensity over a pre-set short-term prediction window in the future. The danger intensity is derived from the comprehensive safety field given by A5 or its equivalent risk density. ,Will Interpreted as the intensity of danger per unit time, over a length of... Within the short-term horizon, to highlight that near-terminal dangers are more urgent than far-terminal dangers, a time core is introduced. From union probability to time-weighted integral, we define the urgency of takeover:

[0143]

[0144] In the formula: This indicates the level of risk between the vehicle and its surrounding environment; It is a time decay factor, representing the decrease in the urgency of takeover over time. This is the starting point of the current evaluation window; It is the length of the time window; To unify timestamps between the perception and control stack,

[0145] The current urgency of takeover is calculated by weighting and averaging environmental risks over a preset time period (e.g., 1-3 seconds); a time decay factor is used to ensure that risks farther away from the current moment have a smaller impact.

[0146] The current urgency of takeover is calculated by weighting and averaging environmental risks over a preset time period (e.g., 1-3 seconds); a time decay factor is used to ensure that risks farther away from the current moment have a smaller impact.

[0147] C2. Feasibility Index The calculation,

[0148] Feasibility Index This measure assesses the probabilistic accessibility of a driver's ability to successfully take over in real time within the safe horizon. The core variables are the capability field, reaction time, and the safe horizon. The feasibility of takeover is assessed by evaluating the driver's ability to take over and their reaction time, taking into account the driver's physiological state, reaction time, and the current takeover window. The feasibility index is calculated accordingly. The higher the value, the more successfully the driver can take over.

[0149] Let the key sub-events for a successful takeover be: awareness / understanding (determined by attention, trust, and workload); reaction / operation (determined by reaction time and action quality); and environmental accommodation (determined by...). (The given time and margin of safety are determined by these factors, which are approximately independent. The probability of success can be expressed as:)

[0150] ;

[0151] In the formula: Probability of success; To perceive / understand the probability of success; This represents the probability of a successful reaction. To determine the probability of successful environmental containment;

[0152] Log-odds linearization, followed by Sigmoid mapping, yields:

[0153] ;

[0154] In the formula: Feasibility index; The driver takeover capability index (derived from physiological data); For driver reaction time, for the current assessment moment, keep the alignment within the same frame; This is the safe time window for takeover, and also the current assessment time; This is to reduce the driver's workload. As a measure of the driver's trust in the system; , , , These are the coupling weights, i.e., the weight coefficients of the behavioral features; To prevent constants with a denominator of zero, the system is specifically a driver-to-vehicle autonomous driving and takeover coordination system (including modules such as perception fusion, environmental modeling and safety field, decision planning and longitudinal and lateral control, takeover management and HMI prompts).

[0155] This represents the driver's current ability to take over; the higher the value, the more suitable the driver's physiological and cognitive state is for taking over. This refers to the time it takes for a driver to react from perception to decision; the shorter the reaction time, the easier it is to successfully take over. The safety horizon indicates the timeframe for takeover. The smaller the size, the shorter the time required for takeover; This is the Sigmoid activation function, used to map the output to the range [0,1].

[0156] C3. Safety margin The calculation,

[0157] safety margin Used to measure the safety of a vehicle's current operation, taking into account both longitudinal braking and lateral stability, it is one of the key indicators determining whether a takeover operation can proceed. Specifically, the safety margin is divided into longitudinal margin and lateral margin, and the minimum value of both determines the final safety margin. The longitudinal and lateral margins together determine whether the vehicle has enough safety space to perform a takeover operation without a collision or loss of control under the current driving conditions.

[0158] C3.1. Calculating the longitudinal margin involves the longitudinal clearance between the vehicle and the vehicle in front, as well as the vehicle's braking capacity. When taking over, the longitudinal margin is estimated using a stopping distance model, assuming the vehicle can quickly decelerate and stop at the current speed.

[0159] The parking distance model consists of two parts: the inertial coasting distance during the reaction time, and the braking distance; the parking distance formula is as follows:

[0160] ;

[0161] In the formula: Braking distance; The speed of the vehicle; This refers to the driver's reaction time; The coefficient of adhesion is the friction force between the wheel and the road surface. It is the acceleration due to gravity;

[0162] Longitudinal margin It is the gap between your vehicle and the vehicle in front. Subtract the parking distance, then compare it with the reference distance. Comparison; the calculation formula is:

[0163] ;

[0164] In the formula: For longitudinal margin; The longitudinal clearance between the vehicle in front and the vehicle behind; A standard used to measure safe distances; Braking distance; The speed of the vehicle; This refers to the driver's reaction time; The coefficient of adhesion is the friction force between the wheel and the road surface. It is the acceleration due to gravity;

[0165] when When the value is negative, it indicates that the distance between the vehicle in front and the vehicle itself is insufficient, making it impossible to stop safely under the current conditions. In this case, it is necessary to apply the brakes or take over immediately.

[0166] C3.2. Calculate lateral margin. Lateral margin measures the stability of a vehicle when turning or changing lanes. The lateral acceleration of a vehicle is related to the curvature of the road, vehicle speed, and coefficient of friction. The maximum lateral acceleration that a vehicle can withstand is determined by the coefficient of friction of the road and the vehicle's braking capacity. Calculate the maximum value of lateral acceleration. The maximum value of lateral acceleration Given by the following formula:

[0167] ;

[0168] In the formula: The road adhesion coefficient; It is the acceleration due to gravity;

[0169] The actual lateral acceleration of a vehicle when changing lanes or turning. The lateral margin is calculated based on the current driving conditions, using the following formula:

[0170] ;

[0171] In the formula: For longitudinal margin; This refers to the vehicle's actual lateral acceleration; The maximum permissible lateral acceleration;

[0172] lateral margin It indicates whether the vehicle can safely perform lane change or turn operations. If the lateral margin is negative, it means that the vehicle cannot safely complete the lateral operation in the current state.

[0173] C3.3. Calculate the overall safety margin. It is determined by the longitudinal margin and lateral margin The calculated minimum value is used to achieve security in both the longitudinal and lateral directions. The calculation formula is as follows:

[0174] ;

[0175] In the formula: To provide a comprehensive safety margin, the value can be negative; a positive value indicates that there is sufficient feasible space. This refers to the vehicle's actual lateral acceleration; The maximum permissible lateral acceleration; This refers to the longitudinal clearance between the vehicle in front and the vehicle behind it. A standard used to measure safe distances; The speed of the vehicle; The coefficient of adhesion is the friction force between the wheel and the road surface. It is the acceleration due to gravity;

[0176] Safety margin measures the safety of takeover in both longitudinal and lateral dimensions, ensuring that the vehicle has sufficient safety space to execute the takeover operation in its current state. Longitudinal margin calculation considers the distance between the vehicle and the vehicle in front, as well as the vehicle's braking capability; lateral margin measures the vehicle's stability during cornering or lane changes. The final safety margin is the minimum of the longitudinal and lateral margins. When the safety margin is negative, minimum risk control (MRC) or takeover operation will be triggered to ensure the safety of the vehicle.

[0177] D. Methods for determining human-machine takeover modes and allocating control rights, including: takeover mode selection, allocation of human-machine control rights, and HMI prompt design;

[0178] D1. Takeover Mode Selection

[0179] Feature quantity , , , These represent the urgency of takeover, the feasibility index of takeover, the safety margin, and the safety horizon (operable time window), respectively. The objective of mode selection is to ensure feasibility. >0) and time accessibility ( relatively Under the premise of (sufficiency), to minimize risk and cost and reduce the burden on both humans and machines, three candidate modes are proposed. , For emergency takeover, In order to coordinate the takeover, For gradual takeover, the cost function for each mode :

[0180] ;

[0181] In the formula: , , , For pattern-related weights; The shortest available reaction time is the fastest possible reaction time for the driver from a valid HMI prompt to generating a valid takeover input, plus the inherent system and HMI latency. Option:

[0182] ;

[0183] when and When taking large weights, the equivalent thresholding rule is derived:

[0184] ;

[0185] D2. Allocation of human-machine control.

[0186] The allocation of control must meet the following conditions ∈[0,1], which represents the proportion of driver control. That is, the proportion of system control allocated; with the urgency of takeover As the feasibility index increases, control should shift more towards the system; With increased power, control should be tilted more towards the driver;

[0187] Human-machine control ratio As a function of the urgency and feasibility of takeover, it is represented by a linear combination of logarithmic odds:

[0188] ;

[0189] In the formula: The logarithmic probability represents the relative advantage of the driver in control. , To control the impact of the takeover feasibility index and urgency on the allocation of control; The minimum threshold for the feasibility index of takeover; The minimum threshold for the urgency of takeover; For bias terms; Feasibility index;

[0190] Use the Sigmoid function to convert log odds into a control weight ratio:

[0191] ;

[0192] In the formula: It is the Sigmoid activation function. Between 0 and 1; The logarithmic probability represents the relative advantage of the driver in control. It is the hyperbolic tangent function;

[0193] D3.HMI design prompts,

[0194] Based on the values ​​of two indicators, takeover urgency and takeover feasibility, the system dynamically determines the level of prompts received by the driver. The takeover prompt level design is divided into four levels: L1 (Information Prompt): When the takeover urgency is below the set value, the system only prompts the driver to remain alert; L2 (Attention Prompt): When the takeover urgency is above the set value, the system prompts the driver to prepare for takeover; L3 (Warning Prompt): When the takeover urgency is above the set value, the system prompts the driver to prepare for takeover; L4 (Attention Prompt): When the takeover urgency is above the set value, the system prompts the driver to prepare for takeover. When the warning level is reached, the system issues a warning to remind the driver that takeover is about to occur; L4 (Critical Warning): When the urgency of takeover is the highest or the safety margin is lower than the set value, the system forces takeover and issues an emergency warning;

[0195] The relationship between the level of takeover notification and the urgency and feasibility index of takeover is as follows:

[0196]

[0197] In the formula: , The threshold for urgency determines the transition of the alert level; This represents the minimum value of the takeover feasibility index. Emergency takeover;

[0198] To ensure a smooth transition and avoid excessive or insufficient prompting, the intensity and frequency of takeover prompts are dynamically adjusted based on the urgency of the takeover; specifically:

[0199]

[0200]

[0201] In the formula: The intensity of the warning increases with the urgency of the situation. The frequency of notifications increases with the urgency of takeover. , Thresholds for urgency levels; , These are the lower and upper bounds of the frequency. This is a numerical clipping function (returns 0 for values ​​less than 0 and 1 for values ​​greater than 1).

[0202] E. Takeover Scheme Screening and Optimization: Takeover scheme screening is a crucial step in ensuring the effectiveness of takeover decisions. Through a comprehensive evaluation of takeover mode selection, urgency, feasibility, driver condition, and environmental risk factors, this step ensures that the takeover scheme meets safety standards and effectively adapts to the driver's capabilities and risk profile. This step will select the optimal takeover scheme and further refine and adjust it using a comprehensive cost function derived from an optimization algorithm before implementation in actual operation.

[0203] E1. Preliminary Screening: After the takeover mode is selected and control allocation is determined, the system will conduct a preliminary screening of takeover plans based on the urgency, feasibility index, safety margin, and environmental factors. By evaluating multiple takeover plans, the system will select the plan that meets the takeover conditions and safety requirements. The core objective of the initial screening is to ensure that the takeover plan meets the following conditions:

[0204] Comprehensive assessment of the spatiotemporal safety field: Determining that the relative risks between the vehicle and the environment are within an acceptable range within the predetermined time window;

[0205] Real-time assessment of driver takeover capability: The driver's takeover capability (assessed by the Driver Takeover Capability Index) is sufficient to complete the takeover task;

[0206] Handover quality assurance: Safety margins are used to ensure that there is sufficient safe operating space in both the longitudinal and lateral directions during the handover process;

[0207] The specific filtering criteria are as follows:

[0208] ;

[0209] In the formula: It is in time The safety field value at any given time is used to assess the risks between the vehicle and the environment; To ensure the maximum permissible safe field strength and prevent excessive control; To ensure a safe horizon, ensure that the time window for takeover is sufficient; For the predicted driver reaction time; For safety margin; For safety margin; To determine the length of the scroll evaluation window;

[0210] The initial screening criteria determine the basic safety and rationality of the takeover mission, preventing takeover operations from being carried out in dangerous environments. Takeover plans that do not meet the initial screening criteria will be eliminated, resulting in a set of candidate plans for the initial screening.

[0211] E2. Scheme Optimization and Selection: After initial screening, the system optimizes multiple takeover schemes based on a comprehensive cost function. This comprehensive cost function considers factors such as takeover safety, driver comfort, and operational smoothness. The cost of each takeover scheme is calculated using a weighted summation method, and the optimal scheme is selected. The optimization objective balances multiple dimensions of the takeover as follows:

[0212] Safety: Ensure that no collisions or accidents occur during the takeover process, and maintain the longitudinal and lateral stability of the vehicle;

[0213] Driver comfort: Reduce driver workload and prevent discomfort or tension caused by taking over too early or too late;

[0214] Smoothness of operation: Ensures a smooth transition of vehicle control and prevents overreaction or instability caused by uneven control switching;

[0215] The comprehensive cost function for optimizing the takeover plan is as follows:

[0216] ;

[0217] In the formula: For the takeover plan The overall cost; , , , These are weighting coefficients, representing the weights for controlling safety, comfort, speed stability, and HMI prompts, respectively. To use a unified timestamp for the perception and control stack, avoiding jumps and rollbacks; The cost of HMI alerts is measured by weighing the burden of alerts against the costs of false positives and false negatives. To assess the overall risk field strength; The integral of the rate of change of acceleration; Integral for the vehicle's speed; For the desired speed;

[0218] By calculating the comprehensive cost function, the system selects the optimal takeover scheme and passes it to the execution layer for control.

[0219] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for human-machine co-driving takeover decision-making, stability control, and optimization, characterized in that, Includes the following steps: A. Construct a driving safety field model to reflect the impact of dynamic environment and road elements on the vehicle and provide real-time basis for takeover decisions; B. Construct a driver human factor capability field, and evaluate the driver's takeover capability through multimodal data collection and real-time analysis. This requires a comprehensive evaluation of the driver's physiological signals, behavioral characteristics, and psychological state information, and derive the driver's takeover capability index through a weighted linear combination model. C. The dynamic coupling of the safety field and human capability field in takeover decision-making, including: takeover urgency Calculation and derivation, feasibility index Calculations and safety margins Calculation of feasibility index This measure assesses the probabilistic accessibility of a driver successfully taking over within a safe takeover window, using immediate capabilities. The core variables are the human capability field, reaction time, and the safe takeover window. The feasibility of takeover is assessed by evaluating the driver's ability to take over and their reaction time, taking into account the driver's physiological state, reaction time, and the current takeover window. The feasibility index is calculated accordingly. The higher the value, the more successfully the driver can take over. D. Methods for determining human-machine takeover modes and allocating control rights, including: takeover mode selection, allocation of human-machine control rights, and HMI prompt design; E. Screening and optimization of takeover plans E1. Preliminary Screening: After the takeover mode is selected and control allocation is determined, the system will conduct a preliminary screening of takeover plans based on the urgency, feasibility index, safety margin, and environmental factors. By evaluating multiple takeover plans, the system will select the plan that meets the takeover conditions and safety requirements. The core objective of the initial screening is to ensure that the takeover plan meets the following conditions: Comprehensive assessment of the spatiotemporal safety field: Determining that the relative risks between the vehicle and the environment are within an acceptable range within the predetermined time window; Real-time assessment of driver takeover capability: The driver's takeover capability is sufficient to complete the takeover task; Handover quality assurance: Ensure sufficient longitudinal and lateral safe operating space during handover by using safety margins; The initial screening criteria determine the basic safety and rationality of the takeover task. Takeover plans that do not meet the initial screening criteria will be eliminated, resulting in a set of candidate plans for the initial screening. E2. Scheme Optimization and Selection: After initial screening, the system optimizes multiple takeover schemes based on a comprehensive cost function. This comprehensive cost function considers factors such as takeover safety, driver comfort, and operational smoothness. The cost of each takeover scheme is calculated using a weighted summation method, and the optimal scheme is selected. The optimization objective balances multiple dimensions of the takeover as follows: Safety: Ensure that no collisions or accidents occur during the takeover process, and maintain the longitudinal and lateral stability of the vehicle; Driver comfort: Reduce driver workload and prevent discomfort or tension caused by taking over too early or too late; Smoothness of operation: Ensures a smooth transition of vehicle control and prevents overreaction or instability caused by uneven control switching; The comprehensive cost function for optimizing the takeover plan is as follows: ； In the formula: For the takeover plan The overall cost; , , , These are weighting coefficients, representing the weights for controlling safety, comfort, speed stability, and HMI prompts, respectively. A unified timestamp for the perception and control stack; The cost of HMI alerts is measured by weighing the burden of alerts against the costs of false positives and false negatives. To assess the overall risk field strength; The rate of change of acceleration; For the vehicle's speed; For the desired speed; By calculating the comprehensive cost function, the system selects the optimal takeover scheme and passes it to the execution layer for control.

2. The human-machine co-driving takeover decision-making, stability control, and optimization method according to claim 1, characterized in that, Step A also includes the following steps: A1. Acquire the vehicle's status data relative to the environment, including the vehicle's relative position, speed, and acceleration information with respect to surrounding vehicles; A2. Obtain static information about the road, including environmental visibility, lane width, road adhesion coefficient, and speed limit information; A3. Based on the dynamic information of the self-vehicle and the target vehicle obtained by the vehicle perception layer, a spatiotemporal safety field model is constructed. The spatiotemporal safety field model reflects the risks faced by the self-vehicle based on the relative position, speed, acceleration and target behavior of the target vehicle. A4. Improve the safety field model by considering the influence of the static road environment. The formula for calculating the static safety field is: ； In the formula: This represents the static road field strength. This is the relative distance between the center of the lane and the vehicle. The standard deviation of lane width; Indicator for restricted areas; This is a visibility correction factor, reflecting the impact of visibility on the safety field; For visibility measurement; The weight for centering out of lane; Hard constraint weights; A5. By comprehensively considering the dynamic safety field and the static road safety field, the spatiotemporal comprehensive safety field of the area where the vehicle is located is calculated. Through the comprehensive safety field, the vehicle can fully assess the safety of the environment and provide real-time basis for takeover decisions.

3. The human-machine co-driving takeover decision-making, stability control, and optimization method according to claim 2, characterized in that, Step A3 also includes the following steps: A3.

1. Derive the influence of relative position and relative speed. The safety field assesses risk based on the relative position and speed between the vehicle and the target vehicle. The longitudinal distance between the vehicle and the target vehicle... and lateral distance To improve security, the longitudinal distance is represented using a Gaussian kernel function. and lateral distance The impact on safety; A3.

2. Derivation of Collision Time The collision time between the target vehicle and the vehicle itself is a key factor in assessing collision risk. The shorter the collision time, the closer the target vehicle and the vehicle itself are, the greater the probability of a collision and the greater the risk faced by the vehicle itself. A3.

3. Derivation of the impact of speed difference: Relative speed directly affects the degree of threat posed by the target vehicle to the vehicle itself. If the approach is at high speed, the risk of collision increases. A3.

4. Introduce risk factors. Used to quantify the degree of danger of target behavior; if the target exhibits dangerous behavior, then the risk factor... This amplifies the perceived safety threat posed by the target to the vehicle, increasing the risk factor. This is derived by dynamically evaluating the behavior of the target.

4. The human-machine co-driving takeover decision-making, stability control, and optimization method according to claim 1, characterized in that, Step B also includes the following steps: B1. Multimodal data acquisition The system collects multimodal data and operational behavior data of the driver in real time through onboard sensors and physiological monitoring equipment. The multimodal data includes: heart rate variability, skin conductance response, eye tracking and fixation stability; the operational behavior data includes: the force and frequency of the accelerator and brake pedals, and the steering wheel rotation angle. Based on the multimodal data and operational behavior data, the system determines whether the driver is in a normal driving state and whether there are any sudden reactions or abnormal operations. B2. Calculation of the driver takeover capability index First, by integrating multimodal data and operational behavior data, the takeover capability index is used as a comprehensive measure to reflect whether the driver can successfully take over vehicle control at a specific moment. A weighted linear combination model is used, and the results are normalized by the Sigmoid function so that the takeover capability index TCI value is between [0,1]. The weighted linear combination model weights physiological signals, behavioral and psychological characteristics to calculate a takeover capability index that reflects the driver's ability to take over. , The closer the value is to 1, the stronger the driver's ability to take over; the closer it is to 0, the weaker the driver's ability to take over. B3. Driver reaction time prediction model The reaction time is modeled using a Gaussian distribution, where the mean and standard deviation of the reaction time need to be dynamically adjusted according to the driver's current state. The prediction of the reaction time is based on the driver's physiological state, emotional fluctuations, and current driving environment factors. By collecting multimodal data from drivers and combining it with Gaussian distribution modeling, driver reaction time can be predicted.

5. The human-machine co-driving takeover decision-making, stability control, and optimization method according to claim 1, characterized in that, Step C also includes the following steps: C1. Urgency of Takeover The calculation and derivation, Takeover assessment and decision-making are based on the spatiotemporal safety field and the human factor capability field. By convolutional accumulation of hazard intensity and time, and then through accessibility analysis of capability and time window, the urgency of takeover is determined. Feasibility index of takeover With safety margin Based on this, mode selection and thresholding decisions are made. The urgency of takeover is used to measure the cumulative degree of hazard intensity over time within a preset short-term prediction window. The hazard intensity originates from the comprehensive safety field or the risk density equivalent to the comprehensive safety field. ,Will Interpreted as the intensity of danger per unit time, over a length of... Within the short-time horizon, a time kernel is introduced. From union probability to time-weighted integral; The current urgency of takeover is calculated by weighting and averaging environmental risks over a preset period of time. C2. Feasibility Index The calculation, Let the key sub-events for a successful takeover be: awareness / understanding; reaction / action; and environmental accommodation. The probability of success can be expressed as: ； In the formula: Probability of success; To perceive / understand the probability of success; This represents the probability of a successful reaction. To determine the probability of successful environmental containment; Log-odds linearization, followed by Sigmoid mapping, yields: ； In the formula: Feasibility index; The driver takeover capability index; This refers to the driver's reaction time. This is the safe time window for takeover, and also the current assessment moment; This is to reduce the driver's workload. As a measure of the driver's trust in the system; , , , These are the coupling weights, i.e., the weight coefficients of the behavioral features; To prevent constants with a denominator of zero; C3. Safety margin The calculation, Safety margin Used to measure the safety of a vehicle's current operation, taking into account both longitudinal braking and lateral stability, the safety margin is divided into longitudinal margin and lateral margin. The longitudinal and lateral margins together determine whether the vehicle has enough safety space to perform a takeover operation under the current driving conditions without a collision or loss of control.

6. The human-machine co-driving takeover decision-making, stability control, and optimization method according to claim 5, characterized in that, Step C3 also includes the following steps: C3.

1. Calculating the longitudinal margin involves the longitudinal clearance between the vehicle and the vehicle in front, as well as the vehicle's braking capacity. When taking over, the longitudinal margin is estimated using a stopping distance model, assuming the vehicle can quickly decelerate and stop at the current speed. The parking distance model consists of two parts: the inertial coasting distance during the reaction time and the braking distance; Longitudinal margin It is the gap between your vehicle and the vehicle in front. Subtract the parking distance, then compare it with the reference distance. Compare; when When the value is negative, it indicates that the distance between the vehicle in front and the vehicle itself is insufficient, making it impossible to stop safely under the current conditions. In this case, it is necessary to apply the brakes or take over immediately. C3.

2. Calculate lateral margin. Lateral margin measures the stability of a vehicle when turning or changing lanes. The lateral acceleration of a vehicle is related to the curvature of the road, vehicle speed, and coefficient of friction. The maximum lateral acceleration that a vehicle can withstand is determined by the coefficient of friction of the road and the vehicle's braking capacity. Calculate the maximum value of lateral acceleration. ; The actual lateral acceleration of a vehicle when changing lanes or turning. Calculate the lateral margin based on the current driving conditions; C3.

3. Calculate the overall safety margin. It is determined by the longitudinal margin and lateral margin The calculated minimum value is used to achieve security in both the longitudinal and lateral directions; The final safety margin is the minimum of the longitudinal and lateral margins. When the safety margin is negative, minimum risk control or takeover will be triggered.

7. The human-machine co-driving takeover decision-making, stability control, and optimization method according to claim 1, characterized in that, Step D also includes the following steps: D1. Takeover Mode Selection Feature quantity , , , These represent the urgency of takeover, the feasibility index of takeover, the safety margin, and the safe time window for takeover, respectively, with three candidate models. , For emergency takeover, In order to coordinate the takeover, To take over gradually; D2. Allocation of human-machine control. The allocation of control must meet the following conditions ∈[0,1], which represents the proportion of driver control. That is, the proportion of system control allocated; with the urgency of takeover As the feasibility index increases, control tends to shift towards the system; Increased control tends to shift towards the driver; Human-machine control ratio As a function of the urgency and feasibility of takeover, it is represented by a linear combination of logarithmic odds: ； In the formula: The logarithmic probability represents the relative advantage of the driver in control. , To control the impact of the takeover feasibility index and urgency on the allocation of control; The minimum threshold for the feasibility index of takeover; The minimum threshold for the urgency of takeover; For bias terms; Feasibility index; Use the Sigmoid function to convert log odds into a control weight ratio: ； In the formula: It is the Sigmoid activation function. Between 0 and 1; The logarithmic probability represents the relative advantage of the driver in control. It is the hyperbolic tangent function; D3.HMI design prompts, Based on the values ​​of two indicators, takeover urgency and takeover feasibility, the system dynamically determines the level of prompts received by the driver. The takeover prompt levels are designed into four levels: L1: When the takeover urgency is below the set value, the system only prompts the driver to remain alert; L2: When the takeover urgency is above the set value, the system prompts the driver to prepare for takeover; L3: When the takeover urgency is above the set value, the system prompts the driver to prepare for takeover; When the warning level is reached, the system issues a warning to remind the driver that takeover is about to occur; L4: When the urgency of takeover is the highest or the safety margin is lower than the set value, the system forces takeover and issues an emergency warning; The intensity and frequency of takeover notifications are dynamically adjusted based on the urgency of the takeover.

Images