A human-machine co-driving shared steering control method, system, medium and device

Abstract

The application discloses a kind of man-machine co-pilot shared steering control method, system, medium and equipment.The application distinguishes real control conflict from channel time delay by time delay registration;By fusing multi-dimensional consistency evaluation components into collaborative consistency index, and recursively estimating trust state from state space model with uncertainty, it overcomes the noise caused by the direct superposition of uncalibrated single trust value to control right formula, which causes control right jitter;Crucially, according to the trust state with uncertainty, the rolling optimization problem constrained by vehicle state and control barrier function is solved in the candidate scheme set, realizing the hierarchical decoupling of trust preference and safety constraint, trust preference only generates soft preference, and the control right feasible region is independently defined by both parties input their own safety cost, even if the trust decreases under high risk, the control right is always allocated to the party with lower safety cost, thereby ensuring safety priority in high-risk scenarios.

Description

Technical Field

[0001] This invention relates to the fields of human-machine shared steering control, online driver state estimation, vehicle lateral control, and on-board real-time computing technology for intelligent connected vehicles, and particularly to a human-machine co-driving shared steering control method, system, medium, and device. Background Technology

[0002] In Level 2 to Level 3 autonomous driving and certain limited-scenario human-machine co-driving vehicles, the driver and the autonomous driving system may simultaneously influence the steering actuators. The shared controller needs to coordinate the driver's intentions, the autonomous driving system's planning results, and the vehicle's state while ensuring vehicle safety and stability. Fixed-proportion control or control that only changes with road curvature is insufficient to adapt to continuous changes in driver acceptance of the system, control conflicts, and environmental risks.

[0003] To address this need, existing technical solutions mainly develop in three directions: First, control rights are allocated based on simple rules that vary only with road curvature or a fixed ratio, roughly adjusting the fusion ratio of human-machine steering input through preset fixed weights or curvature thresholds; Second, dynamic weighting schemes based on driver behavior or preference modeling are used, which estimate weights and allocate control rights by collecting the differences between driver steering input and autonomous driving system steering input (such as single error indicators like steering angle error or root mean square error) and combining them with road risk, driver ability, or degree of human-machine conflict; Third, optimization methods such as model predictive control are introduced to solve the shared steering control quantity, transforming the allocation of control rights into a rolling optimization problem constrained by vehicle state.

[0004] However, existing solutions either directly superimpose uncalibrated trust values ​​linearly into the control weight formula, causing noise and hysteresis in trust estimation to be indiscriminately translated into changes in control weight, leading to frequent control weight fluctuations and even incorrect allocation in emergency situations; or they rely solely on a single steering error index to measure human-machine consistency, failing to distinguish between diverse states such as continuous slow corrections, short-term violent confrontations, timing lags, energy differences, and different directions of risk improvement, making trust evaluation unable to truly reflect the safety situation of human-machine collaboration. As a result, when the vehicle is in a high-risk scenario (such as emergency braking, strong confrontational intent, or communication disruption), the control weight allocation decision of existing technologies may assign that share to the less safe executor, violating the fundamental principle of "safety always comes first" in human-machine co-driving design.

[0005] In summary, existing technical solutions, due to decreased trust in high-risk scenarios, lead to the incorrect allocation of control to the party with lower security, thus failing to guarantee security priority in conflict situations. Summary of the Invention

[0006] Therefore, it is necessary to provide a human-machine co-driving shared steering control method, system, medium, and device to address the aforementioned technical problems. This method can continuously and accurately assess the human's trust in autonomous driving from the differences in human-machine steering input without relying on additional physiological sensors, and accordingly generate safe and feasible shared steering control quantities when trust preferences conflict with vehicle safety.

[0007] The following technical solution is adopted in this specification: This manual provides a human-machine co-driving and shared steering control method, specifically including: The current vehicle status, driver steering input sequence, and autonomous driving system steering input sequence are obtained. Time delay registration is performed on the driver steering input sequence and the autonomous driving system steering input sequence to obtain the human-machine steering input difference sequence. The time delay registration is a sequential time delay estimation and synchronization registration. The communication or execution channel delay is eliminated by cross-correlation time shift estimation, and the difference is distinguished into real control conflict and channel delay.

[0008] Based on the human-machine steering input difference sequence, multiple consistency evaluation components are calculated; by combining the fusion weights that are adaptively updated in the driving scenario, all consistency evaluation components are fused to obtain the human-machine steering collaborative consistency index.

[0009] The human-machine steering coordination consistency index is input into a trust observation mapping with monotonic constraints, and then recursively estimated by a state-space model containing trust state variables and individual difference parameters to obtain the trust state, trust change trend and uncertainty of human-machine co-driving operation.

[0010] Based on the trust state, the trust change trend, and the uncertainty, and in conjunction with the current vehicle state and road environment, a rolling optimization problem constrained by vehicle state and control barrier function is solved within the candidate control scheme set represented by the steering input sequence of the autonomous driving system, and a shared steering control quantity is output; the candidate control scheme set is directly given by the steering input sequence of the autonomous driving system.

[0011] After the shared steering control quantity is input to the steering actuator, the vehicle status is updated, and the vehicle waits for the next round of steering control.

[0012] Furthermore, the current vehicle state includes at least one of the following: yaw rate, longitudinal acceleration, lateral acceleration, steering wheel angle, and vehicle speed, obtained through vehicle state sensors; the current vehicle state is also used in conjunction with the driving scenario to generate comprehensive driving risks.

[0013] Furthermore, the time delay estimation calculates the time shift corresponding to the maximum cross-correlation between the driver's steering angular velocity sequence and the autonomous driving system's steering angular velocity sequence within a preset time delay search range using a normalized cross-correlation function; the preset time delay search range is adaptively determined based on the vehicle's dynamic characteristics.

[0014] The synchronization registration aligns the driver's steering input sequence with the autonomous driving system's steering input sequence based on the time shift, and outputs the aligned human-machine steering input difference sequence.

[0015] Furthermore, the plurality of consistency evaluation components include at least spatial consistency components, temporal consistency components, control consistency components, risk consistency components, and trend consistency components, which are used to describe the spatial relationship, temporal relationship, control relationship, risk relationship, and trend of change between the driver's steering input sequence and the autonomous driving system steering input sequence, respectively.

[0016] Furthermore, the process of fusing all consistency evaluation components by combining the adaptively updated fusion weights of the driving scenario to obtain the human-machine steering coordination consistency index specifically includes: The objective weights calculated using the entropy weight method are weighted and synthesized with the expert prior weights to obtain the basic fusion weights. The basic fusion weights are then corrected online according to the driving scenario to obtain the corrected fusion weights. All consistency evaluation components are weighted and arithmetically fused according to the corrected fusion weights to obtain the arithmetic fusion result, and simultaneously all consistency evaluation components are weighted and geometrically fused to obtain the geometric fusion result. The human-machine steering coordination consistency index is obtained by adaptively combining the arithmetic fusion result and the geometric fusion result according to the risk gating coefficient.

[0017] The risk gating coefficient increases with the increase of comprehensive driving risk and the decrease of the minimum value among the multiple consistency evaluation components. When the risk gating coefficient increases, the proportion of the weighted geometric fusion result is increased.

[0018] Furthermore, the trust observation mapping with monotonic constraints satisfies a non-decreasing monotonic function relationship with respect to the human-machine steering coordination consistency index; the state-space model describes the trust state in the log-odds domain, including trust state variables, trust state change rate variables, and individual difference parameters; the recursive estimation is implemented using interactive multi-model Kalman filtering, which performs probability fusion among the trust rising sub-model, trust stationary sub-model, and trust falling sub-model according to mode transition probabilities, outputting the human-machine co-driving operation trust state, the trust change trend, and the uncertainty, wherein the uncertainty is the value obtained by the Jacobian transformation of the recursive estimation covariance through the corresponding observation mapping.

[0019] Furthermore, the solution of the rolling optimization problem constrained by vehicle state and control barrier function, and the output of the shared steering control quantity specifically includes: The candidate control scheme set represented by the driver's steering input sequence and the candidate control scheme set represented by the autonomous driving system's steering input sequence are respectively used as candidate control schemes for short-time domain safety cost prediction. Based on the safety cost prediction results, the safety constraint range of the autonomous driving control right coefficient is determined as the control right safe and feasible interval. The safe and feasible interval is independent of the value of the trust state and is determined only by the safety consequences input by both parties.

[0020] Within the safe and feasible control range, the desired control power, which depends on the trust state, is used as a soft constraint in the objective function, while the vehicle state constraint and the control barrier function are used as hard constraints. The rolling optimization problem is solved, and the shared steering control quantity is output. The desired control power is optimized and realized only within the safe and feasible control range. The control barrier function is used to ensure that the closed-loop system state is always within the preset safety set.

[0021] This manual provides a human-machine co-driving and shared steering control system, specifically including: The time-delay registration module is used to acquire the current vehicle state, driver steering input sequence, and autonomous driving system steering input sequence, and to perform time-delay registration on the driver steering input sequence and the autonomous driving system steering input sequence to obtain the human-machine steering input difference sequence. The time-delay registration is a time-delay estimation and synchronization registration performed sequentially on the driver steering input sequence and the autonomous driving system steering input sequence. The communication or execution channel delay is eliminated by cross-correlation time-shift estimation, and the difference is distinguished into real control conflict and channel delay.

[0022] The evaluation component calculation and fusion module is used to calculate multiple consistency evaluation components based on the human-machine steering input difference sequence; by combining the fusion weights that are adaptively updated in the driving scenario, all consistency evaluation components are fused to obtain the human-machine steering collaborative consistency index.

[0023] The trust observation mapping and recursive estimation module is used to input the human-machine steering coordination consistency index into the trust observation mapping with monotonic constraints, and then perform recursive estimation using a state-space model containing trust state variables and individual difference parameters to obtain the trust state, trust change trend and uncertainty of human-machine co-driving operation.

[0024] The shared steering control quantity calculation module is used to solve a rolling optimization problem constrained by vehicle state and control barrier function within the candidate control scheme set represented by the steering input sequence of the autonomous driving system, based on the trust state, the trust change trend, and the uncertainty, combined with the current vehicle state and road environment, and output the shared steering control quantity; the candidate control scheme set is directly given by the steering input sequence of the autonomous driving system.

[0025] The execution module is used to input the shared steering control quantity to the steering actuator, update the vehicle status, and wait for the next round of steering control.

[0026] The beneficial effects of this invention are as follows: Compared with the prior art, this invention distinguishes between real control conflicts and channel delays through time-delay registration; by integrating multi-dimensional consistency evaluation components into a collaborative consistency index and recursively estimating the trust state using a state-space model with uncertainty, it overcomes the noise-induced control jitter caused by directly superimposing a single trust value onto the control power formula without calibration; crucially, based on the trust state with uncertainty, it solves the rolling optimization problem constrained by vehicle state and control barrier function within the candidate solution set, achieving hierarchical decoupling of trust preference and safety constraints. Trust preference only generates soft preferences, and the feasible domain of control power is independently defined by the safety costs of each party. Even if trust decreases under high risk, control power is always allocated to the party with lower safety costs, thus ensuring safety priority in high-risk scenarios. Attached Figure Description

[0027] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0028] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the composition and data flow of the shared steering control system according to an embodiment of the present invention; Figure 3 This is a flowchart of the adaptive dual-time-scale window and time-delay registration process according to an embodiment of the present invention; Figure 4 This is a flowchart of the five-dimensional consistency evaluation and HMSCI fusion process according to an embodiment of the present invention; Figure 5 This is a flowchart of an online trust recursive estimation process with uncertainty, according to an embodiment of the present invention. Figure 6 This is a flowchart illustrating the control right safety feasible domain and shared steering closed-loop control in an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this invention.

[0030] This invention provides a human-machine co-driving shared steering control method, comprising five steps: time-delay registration, consistency evaluation component calculation and fusion, trust observation mapping and recursive estimation, shared steering control quantity calculation, and shared steering execution. By acquiring the steering inputs from the driver and the autonomous driving system and performing time-delay registration, differences are distinguished into real conflicts and channel delays, avoiding misjudgments of apparent differences that could mislead trust evaluation. By fusing multiple consistency components into a human-machine steering cooperative consistency index, a multi-dimensional and continuous quantification of the cooperative state is achieved, overcoming the limitation of a single error index in distinguishing diverse states. Furthermore, by inputting the consistency index into a monotonic constraint observation mapping and recursively estimating it using a state-space model,… By obtaining a continuous trust state with uncertainty, the problem of control jitter caused by the direct superposition of uncalibrated trust values ​​is overcome. By solving the rolling optimization within the set of autonomous driving candidate solutions with the trust state with uncertainty, which is constrained by vehicle state and control barrier function, the layered decoupling of trust preference and safety constraint is achieved. Trust preference only generates soft preference, and the feasible region is independently defined by the safety cost input by both parties. This ensures that in the event of a conflict, control is always allocated to the party with the lower safety cost. Thus, even if trust decreases in high-risk scenarios, the safe feasible region can still be guaranteed to be protected from being breached. This solves the problem that existing technologies cannot guarantee safety priority due to the incorrect allocation of control caused by trust decrease.

[0031] A human-machine co-driving and shared steering control method, such as Figure 1 As shown, it specifically includes: Step 1: Obtain the current vehicle status, driver steering input sequence, and autonomous driving system steering input sequence. Perform time delay registration on the driver steering input sequence and the autonomous driving system steering input sequence to obtain the human-machine steering input difference sequence. Time delay registration is performed sequentially by time delay estimation and synchronization registration. The communication or execution channel delay is eliminated by cross-correlation time shift estimation, and the difference is distinguished into real control conflict and channel delay.

[0032] Optionally, the current vehicle status includes at least one of the following: yaw rate, longitudinal acceleration, lateral acceleration, steering wheel angle, and vehicle speed, obtained through vehicle status sensors; the current vehicle status is also used in conjunction with the driving scenario to generate comprehensive driving risks.

[0033] Among them, the time delay estimation calculates the time shift corresponding to the maximum cross-correlation between the driver's steering angular velocity sequence and the autonomous driving system's steering angular velocity sequence within a preset time delay search range using a normalized cross-correlation function; the preset time delay search range is adaptively determined based on the vehicle's dynamic characteristics.

[0034] Synchronous registration aligns the driver's steering input sequence with the autonomous driving system's steering input sequence based on the time shift, and outputs the aligned human-machine steering input difference sequence.

[0035] Step 2: Calculate multiple consistency evaluation components based on the human-machine steering input difference sequence; combine all consistency evaluation components with the adaptively updated fusion weights of the driving scenario to obtain the human-machine steering collaborative consistency index.

[0036] Optionally, the multiple consistency evaluation components include at least spatial consistency components, temporal consistency components, control consistency components, risk consistency components, and trend consistency components, which are used to describe the spatial relationship, temporal relationship, control relationship, risk relationship, and changing trend between the driver's steering input sequence and the autonomous driving system steering input sequence, respectively.

[0037] Specifically, by combining adaptively updated fusion weights based on driving scenarios, all consistency evaluation components are fused to obtain the human-machine steering coordination consistency index. This includes: weighting and synthesizing objective weights calculated using the entropy weight method with expert prior weights to obtain basic fusion weights; online correction of the basic fusion weights based on driving scenarios to obtain corrected fusion weights; weighted arithmetic fusion of all consistency evaluation components based on the corrected fusion weights to obtain an arithmetic fusion result, and simultaneously weighted geometric fusion of all consistency evaluation components to obtain a geometric fusion result; and adaptive combination of the arithmetic fusion result and the geometric fusion result based on a risk gating coefficient to obtain the human-machine steering coordination consistency index.

[0038] The risk gate coefficient increases with the increase of comprehensive driving risk and the decrease of the minimum value among multiple consistency evaluation components. When the risk gate coefficient increases, the proportion of the weighted geometric fusion result is increased.

[0039] Step 3: Input the human-machine steering coordination consistency index into the trust observation mapping with monotonic constraints, and then use the state-space model containing trust state variables and individual difference parameters to recursively estimate the trust state, trust change trend and uncertainty of human-machine co-driving operation.

[0040] It should be noted that the trust observation mapping with monotonic constraints satisfies a non-decreasing monotonic function relationship with respect to the human-machine steering coordination consistency index; the state-space model describes the trust state in the log-odds domain, including trust state variables, trust state change rate variables, and individual difference parameters; the recursive estimation is implemented using interactive multi-model Kalman filtering, which performs probability fusion between the trust rising sub-model, the trust stationary sub-model, and the trust falling sub-model according to the mode transition probability, and outputs the human-machine co-driving operation trust state, trust change trend, and uncertainty. The uncertainty is the value obtained by the Jacobian transformation of the recursive estimation covariance through the corresponding observation mapping.

[0041] Step 4: Based on the trust state, trust change trend and uncertainty, combined with the current vehicle state and road environment, solve the rolling optimization problem constrained by the vehicle state and control barrier function within the candidate control scheme set represented by the steering input sequence of the autonomous driving system, and output the shared steering control quantity; the candidate control scheme set is directly given by the steering input sequence of the autonomous driving system.

[0042] Step 5: After inputting the shared steering control input to the steering actuator, update the vehicle status and wait for the next round of steering control.

[0043] Specifically, solving the rolling optimization problem constrained by vehicle state and control barrier function, and outputting the shared steering control quantity, includes: taking the candidate control scheme set represented by the driver's steering input sequence and the candidate control scheme set represented by the autonomous driving system's steering input sequence as candidate control schemes, respectively, performing short-time domain safety cost prediction, determining the safety constraint range of the autonomous driving control right coefficient based on the safety cost prediction results, and using it as the control right safe and feasible interval. The safe and feasible interval is independent of the value of the trust state and is determined only by the safety consequences input by both parties.

[0044] Within the safe and feasible control interval, the desired control right, which depends on the trust state, is used as a soft constraint in the objective function, while the vehicle state constraint and the control barrier function are used as hard constraints. The rolling optimization problem is solved, and the shared steering control quantity is output. The desired control right is only optimized within the safe and feasible control interval. The control barrier function is used to ensure that the closed-loop system state is always within the preset safe set.

[0045] The technical solutions provided by the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0046] Example The system composition and data flow of this embodiment are as follows: Figure 2 As shown.

[0047] S1: Data Acquisition and Synchronization Processing Obtain the driver steering input sequence, the autonomous driving system steering input sequence, the vehicle operating state sequence, the road environment sequence, and the driving task state, and define the raw data vector of each channel as: ; In the formula, This represents the original data vector at the k-th sampling time. / Indicates the driver's steering angle / steering torque; / This indicates the target steering angle / steering torque of the autonomous driving system; Indicates longitudinal vehicle speed; Indicates yaw rate; Indicates lateral acceleration; / Indicates lateral / heading tracking error; Indicates the curvature of the road; Indicates the collision time; This indicates the driving task status. The variables mentioned above are used to describe the vehicle status, driver input, and environmental information involved in the trust assessment process.

[0048] Based on a unified time reference, data from each channel is resampled, channel delay compensated, and quality marked. After synchronization, the data is calculated using the following formula: ; In the formula, This represents the data of the i-th channel after synchronization; Indicates the resampling interpolation kernel; This indicates the estimated delay for the i-th channel; This represents the set of samples participating in the interpolation at time k. Asynchronous data resampling and fixed delay compensation are performed on a unified time base. Total data quality components. :

[0049] ; In the formula, Represents the integrity component; Indicates the time synchronization quality component; Indicates the quality component of outliers; This represents the driving mode effectiveness component. This indicator quantifies the reliability of the current observation data and provides a basis for subsequent weight adjustments and anomaly handling.

[0050] S2: Adaptive Dual Time Scale Window and Delay Registration like Figure 3 As shown, based on the overall driving risk and the degree of vehicle dynamic changes, short-time windows and long-time windows are adaptively determined. Short-time window Used to detect sudden conflicts, long-term window Used for estimating baselines and trends:

[0051] ; ; In the formula, This indicates the overall driving risk; Indicates lateral dynamic intensity; Indicates the quantity of lane change, obstacle avoidance, or takeover events; Represents a scale factor greater than 1; This represents the Sigmoid function. In this embodiment, the short window is limited to 0.5s to 4s, and the long window is limited to 3s to 20s.

[0052] Within a short window, a normalized cross-correlation is calculated between the driver's steering angular velocity sequence and the autonomous driving system's steering angular velocity sequence within a preset time delay search range. The time offset corresponding to the maximum value of the cross-correlation function is determined as the registration time delay between the human and machine inputs. : ; In the formula, L represents the maximum time delay search range; / Represents the driver / autonomous driving steering angular velocity; corr represents the normalized cross-correlation. It is used to estimate the relative time delay between driver input and autonomous driving system input, thereby reducing misjudgment conflicts caused by communication latency, execution latency, and other factors.

[0053] The steering input sequence of the autonomous driving system is shifted and registered according to the registration delay to distinguish between communication or execution delays and conflicts with actual human-machine control. After registration, a human-machine difference vector is formed. :

[0054] ; In the formula, / / This indicates the difference between steering angle, torque, and angular velocity. This indicates the energy difference input within the window; This represents the lateral error of the candidate trajectory. A unified difference vector is used to carry out subsequent five-dimensional consistency calculations.

[0055] S3: Calculation of Five-Dimensional Consistency Components like Figure 4 As shown, the following five types of consistency evaluation components are calculated based on the registered human-machine difference vector: Spatial consistency components : ; In the formula, / Indicates non-negative weights; / Indicates the angular / lateral error scale. Evaluates the spatial similarity between the human-machine steering angle and the candidate trajectory.

[0056] Time-consistent components : ; In the formula, This represents the cross-correlation coefficient under optimal time delay; This indicates the allowable time delay scale. It also penalizes asynchronous behavior and response delays.

[0057] Control Consistency Components : ; In the formula, Indicates the direction consistency rate; Indicates the duration of the conflict; Indicates the correction frequency; / Indicates the energy input from the human / machine interface; , , , Represents non-negative weights that sum to 1. Provides a comprehensive description of direction, sustained resistance, correction frequency, and control strength.

[0058] Risk Consistency Components : ; ; In the formula, This represents the improvement in security cost resulting from the input of the i-th subject; I represents the safety cost function; J0 represents the indicator function; J0 represents the safety cost scale; it determines whether human and machine inputs simultaneously act in the direction of reducing risk and whether the magnitudes of the actions are close.

[0059] Trend Consistency Components : ; In the formula, / This represents the robust slope of human / machine input within a long window; Indicates the slope of the differential sequence; / Indicates the trend scale. Evaluates the long-term trend of human-computer input changes and whether the differences continue to widen.

[0060] S4: Adaptive Fusion of Risk Gating First, the basic fusion weights are formed offline. : Objective weights are obtained using the entropy weight method, and expert prior weights are obtained and combined using the analytic hierarchy process. ; In the formula, This represents the objective weight of the entropy weight method; Indicates expert prior weights; This represents the combination coefficients. Offline generation of basic weights takes into account both data dispersion and engineering knowledge.

[0061] Subsequently, the online phase was based on the data quality of each component. and scene vectors Perform adaptive correction: ; In the formula, Indicates online adaptive weights; This represents the scene sensitivity parameter. The contribution of each dimension is adjusted according to quality and scene, while maintaining the weights normalized.

[0062] When any data quality falls below a threshold, the fusion weight of the corresponding component is reduced. Furthermore, a risk gating coefficient is generated. :

[0063] ; In the formula, This indicates the overall driving risk; This represents one of the five consistency components. It enhances incomplete compensation fusion when risks increase or significant weaknesses emerge.

[0064] Ultimately, the Human-Machine Steering Collaborative Consistency Index (HMSCI) is obtained by combining weighted arithmetic fusion and weighted geometric fusion. The arithmetic term allows for reasonable compensation between dimensions, while the geometric term is sensitive to critical components approaching zero. The gating coefficient, when risk increases, makes the HMSCI closer to the geometric term, thus preventing critical weaknesses from being unreasonably compensated by high-value components. The core mathematical relationships of trust observation, state recursion, and model selection are as follows: Figure 5 As shown.

[0065] As shown in Table 1, under normal collaborative working conditions, the consistency of each dimension remains at a high level, with HMSCI reaching 0.934. Under short-term conflict and emergency obstacle avoidance conditions, control consistency and risk consistency are significantly reduced, with HMSCI decreasing to 0.572 and 0.390 respectively, thus enabling the differentiation of different levels of human-machine collaborative states.

[0066] Table 1. Five-dimensional consistency evaluation results S5: Online Trust State Recursion with Uncertainty After inputting HMSCI, autonomous driving task performance metrics, comprehensive risk margin, and control consistency components into a trust observation mapping with monotonic constraints, a state-space model is established by describing the trust state using the trust log-probability latent variable: ; ; ; In the formula, Represents the trust state vector; The latent variable representing the log-odds of trust; Indicates the state of the rate of change of trust; Indicates individual bias; Indicates an upward, stable, or downward pattern; Indicates the quantities that drive events, performance, and risks; This indicates the trust status of human-machine co-driving operation.

[0067] Three sub-models are set up, each corresponding to a different rate-of-change preservation coefficient and event-driven gain. Interactive multi-model (IMM) Kalman filtering is used for recursion to sequentially complete the Bayesian prediction of the mode probability and the Kalman gain correction. Then, the mode probabilities are updated based on the innovation likelihood and weighted and fused to output the human-machine co-driving operation trust state. Uncertainty of trust estimation and predicting trends : ; ; ; ; In the formula, This represents the unit vector representing the choice of the latent variable to trust. This represents a state matrix weighted by pattern probability. This indicates a predicted trend in trust changes; This indicates the sampling period. The trust recursion process for this step and the state transition relationships for the three modes (rising, stationary, and falling) are as follows: Figure 5 As shown.

[0068] As shown in Table 2, when the human-machine collaboration consistency gradually decreases from 0.91 to 0.35, the trust state decreases from 0.76 to 0.51, and the probability of the trust decline mode increases from 0.04 to 0.76. When the human-machine collaboration state recovers, the trust state and the probability of the rise mode also recover, indicating that the method can realize the dynamic update of the trust state based on the continuous human-machine interaction state.

[0069] Table 2 Online Trust Recursion Results under Typical Operating Conditions S6: Comprehensive Driving Risk Calculation and Safe Feasibility Domain Determination The overall driving risk consists of lane departure risk, collision time risk, lateral stability risk, road curvature risk, and uncertainty risk. These multiple risks are aggregated in the form of a probability union and conservatively modified. ; ; In the formula, This indicates the overall driving risk; c represents the vehicle state estimation covariance; c represents the interest state selection vector. represents the abnormal indication quantity; kP and ka both represent conservative correction coefficients.

[0070] Using both driver input and autonomous driving system input as candidate control schemes, the predicted safety costs for both schemes are calculated in the prediction time domain based on a unified vehicle prediction model. The safe and feasible range of the autonomous driving control power coefficient is determined based on the safety dominance relationship between the two channels. A positive lower boundary is formed when the autonomous driving input is significantly safer, and the upper boundary is reduced when the driver input is safer or an anomaly is detected in the autonomous driving input. This step decouples "trust preference" from "safety dominance": trust is used to generate the desired control power, and safety dominance is used to define the feasible range, thereby avoiding the incorrect allocation of control power to the less safe side in high-risk scenarios. The overall relationship between the safe and feasible range of control power and control feedback is as follows: Figure 6 As shown.

[0071] As shown in Table 3, under emergency obstacle avoidance conditions, although the trust state decreases to 0.39 and the expected control power decreases to 0.29, the safety constraint limits the control power to the range of [0.83, 1.00] because the driver's safety cost is significantly higher than the autonomous driving safety cost. The final control power is 0.83, thus ensuring that the control power allocation will not exceed the safety constraint due to changes in the trust state.

[0072] Table 3. Human-Machine Safety Costs and Feasible Range of Control Rights S7: Rolling optimization and shared steering control generation within the safe and feasible domain After determining the desired autonomous driving control, the shared steering control quantities are weighted and synthesized in the actual execution: ; ; In the formula, Indicates shared steering control quantity; Indicates the control factor for autonomous driving; This represents the safety correction amount for the control barrier function; Indicates the vehicle status.

[0073] Subsequently, within the feasible interval, a rolling optimization problem containing vehicle state constraints, control weight rate of change constraints, and discrete CBF constraints is solved. : ; In the formula, Q and R represent the trajectory error and control quantity weight matrix, respectively; , and These represent control preference, rate of change, and conflict weight, respectively. Indicates the terminal cost; This indicates the prediction time domain.

[0074] In addition, steering angle, steering angular velocity, steering torque, lateral acceleration, and actuator saturation constraints should be set. Trajectory tracking, control smoothing, trust preference, and conflict suppression should be coordinated through unified optimization. The closed-loop relationship between trust assessment and vehicle-shared steering control is as follows: Figure 6 As shown.

[0075] S8: Output, Feedback, and Online Learning The shared steering control input is output to the steering actuator, and the vehicle's operating status, human-machine steering input, and event results are fed back to the next sampling period, forming a closed loop for sustainable operation. When the data integrity rate, time synchronization error, innovation statistics, vehicle safety margin, and autonomous driving system operating status all meet the corresponding thresholds (i.e., the normalized innovation square is less than the chi-square threshold and the model does not drift significantly), the individualized non-safety parameters (preset adjustable parameters in the individual bias and monotonic observation mapping) are updated online using the recursive least squares method with a forgetting factor.

[0076] The model management layer stores the global baseline version, driver individual version, online candidate version, and rollback version. The candidate version is switched to the effective version only after being checked for scene coverage, error, uncertainty, monotonicity, and safety impact. Safety parameters such as control barrier functions, safety thresholds, rollback state machines, and actuator limits are not allowed to be rewritten online.

[0077] The anomaly detection layer monitors four levels: signal state, state estimation, model operation, and control execution. At the signal level, it detects data range and frame loss; at the estimation level, it detects exceeding limits and model drift; and at the control level, it detects optimization infeasibility and actuator saturation.

[0078] When an anomaly occurs, actions are executed in tiers according to severity: reducing the weight of the corresponding HMSCI component to maintain trust estimation, increasing observation noise, freezing online learning, tightening the rate of change of control weights, or triggering a safety rollback when optimization timeouts or infeasibility occur, the autonomous driving controller malfunctions, or communication fails. The first rollback layer solves a small quadratic programming problem with only barrier constraints near the nearest feasible control weight. The second rollback layer uses lane keeping assist, transitioning to driver control at a limited rate and issuing a warning when the autonomous driving system malfunctions and driver input is valid. The safety rollback control strategy is independent of the trust observation parameters that are allowed to be updated online and is triggered by the anomaly detection layer based on optimization feasibility, control channel health status, and vehicle safety margin.

[0079] To verify the overall control effect of the shared control method, the method of this invention was compared with the fixed proportional control method and the trust direct mapping control method. Table 4 shows that the RMS lateral error and maximum lateral error of the method of this invention were reduced to 0.137 m and 0.318 m, respectively, while the minimum safety margin was increased to 0.92 m, and no CBF safety constraint violation occurred. This indicates that, under the combined effect of trust preference, safe feasible region, and rolling optimization, the method of this invention can improve the lateral control stability of the vehicle while ensuring vehicle safety constraints and reducing frequent changes in control power.

[0080] Table 4 Vehicle control effects under different shared control strategies This manual also provides a human-machine co-driving and shared steering control system, specifically including: The time-delay registration module is used to obtain the current vehicle state, driver steering input sequence, and autonomous driving system steering input sequence. It performs time-delay registration on the driver steering input sequence and the autonomous driving system steering input sequence to obtain the human-machine steering input difference sequence. The time-delay registration is a sequential time-delay estimation and synchronization registration. It eliminates communication or execution channel delay through cross-correlation time-shift estimation and distinguishes the difference into real control conflict and channel delay.

[0081] The evaluation component calculation and fusion module is used to calculate multiple consistency evaluation components based on the human-machine steering input difference sequence; by combining the fusion weights that are adaptively updated in the driving scenario, all consistency evaluation components are fused to obtain the human-machine steering collaborative consistency index.

[0082] The Trust Observation Mapping and Recursive Estimation Module is used to input the human-machine steering coordination consistency index into the Trust Observation Mapping with monotonic constraints, and then perform recursive estimation using a state-space model containing trust state variables and individual difference parameters to obtain the trust state, trust change trend, and uncertainty of human-machine co-driving operation.

[0083] The shared steering control quantity calculation module is used to solve a rolling optimization problem constrained by vehicle state and control barrier function within the candidate control scheme set represented by the steering input sequence of the autonomous driving system, based on the trust state, trust change trend and uncertainty, combined with the current vehicle state and road environment, and output the shared steering control quantity; the candidate control scheme set is directly given by the steering input sequence of the autonomous driving system.

[0084] The execution module is used to update the vehicle status after inputting the shared steering control quantity to the steering actuator, and wait for the next round of steering control.

[0085] This specification provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0086] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0087] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A human-machine co-driving and shared steering control method, characterized in that, Includes the following steps: The current vehicle status, driver steering input sequence, and autonomous driving system steering input sequence are obtained. Time delay registration is performed on the driver steering input sequence and the autonomous driving system steering input sequence to obtain the human-machine steering input difference sequence. The time delay registration is a time delay estimation and synchronization registration performed sequentially on the driver steering input sequence and the autonomous driving system steering input sequence. Cross-correlation time shift estimation is used to eliminate communication or execution channel delays and distinguish the differences into real control conflicts and channel delays. Based on the human-machine steering input difference sequence, multiple consistency evaluation components are calculated; By combining the fusion weights that are adaptively updated in the driving scenario, all consistency evaluation components are fused to obtain the human-machine steering coordination consistency index. This index is then input into a trust observation mapping with monotonic constraints, and recursively estimated using a state-space model containing trust state variables and individual difference parameters to obtain the human-machine co-driving operation trust state, trust change trend, and uncertainty. Based on the trust state, the trust change trend, and the uncertainty, and considering the current vehicle state and road environment, a rolling optimization problem constrained by vehicle state and control barrier functions is solved within the candidate control scheme set represented by the autonomous driving system steering input sequence, outputting a shared steering control quantity. The candidate control scheme set is directly given by the autonomous driving system steering input sequence. After the shared steering control quantity is input to the steering actuator, the vehicle status is updated, and the vehicle waits for the next round of steering control.

2. The human-machine co-driving and shared steering control method as described in claim 1, characterized in that: The current vehicle status includes at least one of the following: yaw rate, longitudinal acceleration, lateral acceleration, steering wheel angle, and vehicle speed, obtained through vehicle status sensors; the current vehicle status is also used in conjunction with the driving scenario to generate comprehensive driving risks.

3. The human-machine co-driving and shared steering control method as described in claim 1, characterized in that: The time delay estimation calculates the time shift corresponding to the maximum cross-correlation between the driver's steering angular velocity sequence and the autonomous driving system's steering angular velocity sequence within a preset time delay search range using a normalized cross-correlation function; the preset time delay search range is adaptively determined based on vehicle dynamics characteristics. The synchronization registration aligns the driver's steering input sequence with the autonomous driving system's steering input sequence based on the time shift, and outputs the aligned human-machine steering input difference sequence.

4. The human-machine co-driving and shared steering control method as described in claim 1, characterized in that: The multiple consistency evaluation components include at least spatial consistency components, temporal consistency components, control consistency components, risk consistency components, and trend consistency components, which are used to describe the spatial relationship, temporal relationship, control relationship, risk relationship, and changing trend between the driver's steering input sequence and the autonomous driving system steering input sequence, respectively.

5. The human-machine co-driving and shared steering control method as described in claim 2, characterized in that, The process of fusing all consistency evaluation components by combining adaptively updated fusion weights based on driving scenarios to obtain the human-machine steering coordination consistency index specifically includes: The objective weights calculated using the entropy weight method are weighted and synthesized with the expert prior weights to obtain the basic fusion weights. The basic fusion weights are then corrected online according to the driving scenario to obtain the corrected fusion weights. All consistency evaluation components are then weighted and arithmetically fused according to the corrected fusion weights to obtain the arithmetic fusion result. Simultaneously, all consistency evaluation components are also weighted and geometrically fused to obtain the geometric fusion result. Finally, the human-machine steering coordination consistency index is obtained by adaptively combining the arithmetic fusion result and the geometric fusion result according to the risk gating coefficient. The risk gating coefficient increases with the increase of comprehensive driving risk and the decrease of the minimum value among the multiple consistency evaluation components. When the risk gating coefficient increases, the proportion of the weighted geometric fusion result is increased.

6. The human-machine co-driving and shared steering control method as described in claim 1, characterized in that: The trust observation mapping with monotonic constraints satisfies a non-decreasing monotonic function relationship with respect to the human-machine steering coordination consistency index; the state-space model describes the trust state in the log-odds domain, including trust state variables, trust state change rate variables, and individual difference parameters; the recursive estimation is implemented using interactive multi-model Kalman filtering, which performs probability fusion among the trust rising sub-model, trust stationary sub-model, and trust falling sub-model according to mode transition probabilities, outputting the human-machine co-driving operation trust state, the trust change trend, and the uncertainty, wherein the uncertainty is the value obtained by the Jacobian transformation of the recursive estimation covariance through the corresponding observation mapping.

7. The human-machine co-driving and shared steering control method as described in claim 6, characterized in that, The solution to the rolling optimization problem constrained by vehicle state and control barrier function, and the output of the shared steering control quantity, specifically include: The candidate control scheme set represented by the driver's steering input sequence and the candidate control scheme set represented by the autonomous driving system's steering input sequence are respectively used as candidate control schemes for short-time domain safety cost prediction. Based on the safety cost prediction results, the safety constraint range of the autonomous driving control right coefficient is determined as the control right safe feasible interval. The safe feasible interval is independent of the value of the trust state and is determined only by the safety consequences input by both parties. Within the safe and feasible control range, the desired control power, which depends on the trust state, is used as a soft constraint in the objective function, while the vehicle state constraint and the control barrier function are used as hard constraints. The rolling optimization problem is solved, and the shared steering control quantity is output. The desired control power is optimized and realized only within the safe and feasible control range. The control barrier function is used to ensure that the closed-loop system state is always within the preset safety set.

8. A human-machine co-driving and shared steering control system, characterized in that... include: The time-delay registration module is used to acquire the current vehicle state, driver steering input sequence, and autonomous driving system steering input sequence, and to perform time-delay registration on the driver steering input sequence and the autonomous driving system steering input sequence to obtain the human-machine steering input difference sequence; the time-delay registration is a sequential time-delay estimation and synchronization registration, and the communication or execution channel delay is eliminated by cross-correlation time-shift estimation, and the difference is distinguished into real control conflict and channel delay; The evaluation component calculation and fusion module is used to calculate multiple consistency evaluation components based on the human-machine steering input difference sequence; By combining the fusion weights that are adaptively updated in the driving scenario, all consistency evaluation components are fused to obtain the human-machine steering coordination consistency index. The trust observation mapping and recursive estimation module is used to input the human-machine steering coordination consistency index into the trust observation mapping with monotonic constraints, and then perform recursive estimation by the state space model containing trust state variables and individual difference parameters to obtain the trust state, trust change trend and uncertainty of human-machine co-driving operation. The shared steering control quantity calculation module is used to solve a rolling optimization problem constrained by vehicle state and control barrier function within the candidate control scheme set represented by the steering input sequence of the autonomous driving system, based on the trust state, the trust change trend, and the uncertainty, combined with the current vehicle state and road environment, and output the shared steering control quantity; the candidate control scheme set is directly given by the steering input sequence of the autonomous driving system. The execution module is used to input the shared steering control quantity to the steering actuator, update the vehicle status, and wait for the next round of steering control.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 7.

10. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 7.

Images