A cooperative steering control method considering human-machine trust degree

Abstract

The application discloses a cooperative steering control method considering human-machine trust degree, and aims at solving the human-machine conflict problem in the existing human-machine co-driving system. The application relates to the field of intelligent driving. Specifically, the application comprises an expected path module, a human-machine steering conflict evaluation module, a driver experience evaluation module, a weight distribution module, a non-cooperative game module and a human-machine co-driving vehicle. The expected path module generates the expected paths of the driver and the controller; the human-machine steering conflict evaluation module outputs the trust degree of the driver to the controller based on the steering action of the driver, the vehicle state and the expected paths of the driver and the controller; the driver experience evaluation module outputs the trust degree of the controller to the driver based on the vehicle state; the weight distribution module is used for outputting the driver control weight; and the non-cooperative game module combines the expected paths of the driver and the controller and the driver control weight to output the optimal steering angle, so as to control the human-machine co-driving vehicle to complete the cooperative steering action.

Description

Technical Field

[0001] This invention belongs to the field of intelligent driving, specifically a cooperative steering control method that takes into account human-machine mutual trust. Background Technology

[0002] In recent years, with the rapid development of automotive automation technology, intelligent driving systems have shown great potential in improving driving safety and reducing driver workload. However, limited by the current maturity of perception, decision-making, and execution technologies, fully autonomous driving is still difficult to achieve comprehensively. Human-machine co-driving, as a transitional form of autonomous driving, has become a current research hotspot. Although existing cooperative control technologies have achieved some success, conflicts often arise between drivers and controllers in decision-making. Game theory, due to its unique advantages in multi-agent interactive decision-making, can effectively alleviate decision-making conflicts between humans and machines. However, existing methods mostly adopt fixed or empirical weight allocation mechanisms, lacking adaptability and struggling to cope with complex and ever-changing driving scenarios. Therefore, weight allocation strategies urgently need to achieve dynamic and intelligent adjustments.

[0003] Furthermore, human-machine trust has gradually become an important basis for dynamic weight allocation. Its core ideas generally focus on how to quantify the driver's steering intention and driving ability—two key dimensions—and use these as a basis to guide the dynamic allocation of control. However, existing research relies on simplified assumptions and static rules, making it difficult to dynamically capture the driver's complex cognition and personalized style. In terms of driving ability assessment, it typically only uses indicators such as lateral tracking error and lateral acceleration for quantification, which fails to truly reflect the driver's experience level and cognitive characteristics, resulting in a one-sided and insufficiently adaptable trust assessment. To address this problem, this patent integrates an LSTM-Transformer driver angle prediction model and an inverse reinforcement learning experience quantification model to capture the spatiotemporal features of driving behavior, achieving high-precision prediction of steering intention. Simultaneously, it inverses the latent reward function from expert driving trajectories, effectively quantifying the driver's implicit experience and decision-making reasoning. Compared to existing methods, this invention significantly enhances adaptability to complex environments and individual differences, achieving more accurate and reliable dynamic weight allocation in human-machine collaboration.

[0004] Currently, various methods have been proposed for improving human-machine trust in human-machine collaborative control research, each with its own advantages but also limitations. Patent CN114537419A dynamically adjusts human-machine control weights by quantifying the driver's activity level and basic driving ability; however, the traditional driver models used in this method mostly employ static preset parameters, making it difficult to fully represent the dynamic fluctuations and individual differences in the driver's state. Furthermore, by characterizing comprehensive driving ability solely through single indicators such as lateral deviation and heading angle, it ignores the interactive influence between multiple dimensions of information, including risk perception, decision-making stability, and fatigue, resulting in insufficient model adaptability.

[0005] Patent CN120135223A identifies the driver's long-term behavioral style and comprehensive driving ability to switch human-machine control and adjust weights; however, it only calculates driving ability based on indicators such as years of driving experience and fatigue level, failing to fully utilize multi-dimensional real-time data to dynamically reflect the driver's complex and subtle state changes.

[0006] Patent CN202211540624A proposes a human-machine cooperative steering framework based on LQR and feedforward control. By establishing a second-order driver model that includes two-point preview and neuromuscular characteristics, driver parameters are identified online, and a shared behavior model is constructed. Finally, human-machine control weights are coordinated through LQR feedback, feedforward compensation, and defined stiffness parameters. Although this patent introduces two-point preview and neuromuscular characteristics, its driver model is still based on a fixed second-order linear structure, which cannot fully represent the nonlinear and highly time-varying characteristics of driving behavior. At the same time, its ability to dynamically capture individual differences and states is limited.

[0007] To address the above shortcomings, this invention proposes a collaborative steering control method that considers human-machine mutual trust. Its core lies in constructing a dynamic weight allocation strategy based on a two-way trust assessment mechanism. This not only leverages the conflict resolution capabilities of game-theoretic collaboration but also enhances the system's adaptability and user acceptance by introducing the behavioral cognitive dimension of "human-machine mutual trust." Compared to existing technologies, this method, by employing an LSTM-Transformer model and an inverse reinforcement learning model, overcomes the limitations of traditional models in representing nonlinearity and relying on a single driving ability evaluation index, achieving more accurate driver steering prediction and driving experience assessment. Furthermore, the system uses real-time human-machine mutual trust assessment as the decision-making basis, dynamically allocating driver control weights through fuzzy rules based on the trust levels between the human and machine, and optimizing the solution using non-cooperative game theory. This transforms the human-machine interaction mode from the traditional "passive balancing after conflict" to "active guidance during collaboration" based on two-way trust. This transformation not only enhances the smoothness and safety of the interaction process but also significantly improves the system's acceptability and user experience. Summary of the Invention

[0008] This invention proposes a cooperative steering control method considering human-machine mutual trust. Based on a human-machine steering conflict assessment module and a driver experience assessment module, it calculates the driver's trust in the controller and the controller's trust in the driver, respectively. Specifically, the LSTM-Transformer driver angle prediction model outputs the driver's predicted steering angle, and the MPC controller angle prediction model outputs the controller's predicted steering angle. The human-machine steering conflict assessment module quantifies the driver's trust in the controller based on the difference between the driver's predicted steering angle and the controller's predicted steering angle. The driver experience assessment module evaluates the driver's driving experience based on inverse reinforcement learning, quantifying the controller's trust in the driver. Furthermore, a weight allocation module outputs dynamic driver control weights in real time based on fuzzy rules. A non-cooperative game theory module performs optimization and outputs the optimal steering angle control to achieve human-machine cooperative steering in the co-driving vehicle. In summary, this invention integrates real-time mutual trust assessment and dynamic weight allocation within a non-cooperative game theory framework, achieving a paradigm shift from "post-conflict equilibrium" to "cooperation based on trust."

[0009] The technical solution adopted by this invention to solve the technical problem is as follows:

[0010] To address the problems mentioned above, this invention proposes a cooperative steering control method considering human-machine mutual trust. This method comprises a desired path module, a human-machine steering conflict assessment module, a driver experience assessment module, a weight allocation module, a non-cooperative game theory module, and a human-machine co-driving vehicle, forming a closed-loop control system to achieve cooperative steering control of the human-machine co-driving vehicle. Specifically, the desired path module generates the desired path between the driver and the controller; the human-machine steering conflict assessment module receives the driver's steering action, vehicle state, and the desired path between the driver and the controller, and outputs the driver's trust in the controller; the driver experience assessment module receives the vehicle state and outputs the controller's trust in the driver; the weight allocation module uses the driver's trust in the controller and the controller's trust in the driver as inputs, and outputs the driver's control weights based on fuzzy rules; the non-cooperative game theory module combines the desired path between the driver and the controller, the driver's control weights, and the vehicle state, and obtains the optimal steering angle through prediction model establishment, objective function design, and Nash equilibrium solution, thus achieving cooperative steering control of the human-machine co-driving vehicle.

[0011] The method includes the following steps:

[0012] Step 1: Design of the desired path module. Based on driving environment information, obtain the desired path of the driver and controller, including the horizontal and vertical coordinates and yaw angle of the desired path points of the driver and controller.

[0013] Step 1.1: Driver's desired path design. The driver's desired path is obtained based on the Logitech G29 steering wheel simulation bench, including the horizontal and vertical coordinates of the driver's desired path and the desired yaw angle.

[0014] Step 1.2: Controller desired path design. The controller desired path is obtained based on the CarSim simulation environment, including the horizontal and vertical coordinates of the controller desired path and the desired yaw angle.

[0015] Step 2, Design of the Human-Machine Turning Conflict Assessment Module:

[0016] Based on driver-predicted steering angle and controller predicts the turning angle The difference is used to assess the driver's trust in the controller in real time. .

[0017] Step 2.1, Design of the LSTM-Transformer driver steering angle prediction model, which includes the following sub-steps:

[0018] Step 2.1.1: Data input module, used to collect driver steering actions and vehicle status information of the human-machine co-driving vehicle, including driver steering angle. Driver's steering angle change rate and vehicle longitudinal speed Vehicle yaw rate Longitudinal acceleration Longitudinal coordinates of the driver's desired path Horizontal coordinates of the driver's desired path Yaw angle of the driver's desired path The driver's steering actions and vehicle status information will be used as feature data input to the data preprocessing module of the LSTM-Transformer driver steering angle prediction model. The expected driver steering angle will be used as label data for subsequent driver steering angle prediction. predict.

[0019] Step 2.1.2, the data preprocessing module, includes the following sub-steps:

[0020] Step 2.1.2.1: Normalize the feature data and label data using the MinMax normalization method to make the model training more stable and efficient. The MinMax normalization method preprocesses the data as shown in equation (1):

[0021] (1)

[0022] in, It is normalized data. It is the raw data of driver steering actions and vehicle status information collected by the data acquisition module. It is the maximum value of the feature data. It is the minimum value of the feature data.

[0023] Step 2.1.2.2: Use the sliding window method to select continuous... Normalized data at each time step is used as the input-output pair of the LSTM-Transformer network model, and then converted into tensor data as the input to the model.

[0024] Step 2.1.2.3: Divide the tensor data required by the LSTM-Transformer network for predicting the driver's turning angle into training set, validation set and test set.

[0025] Step 2.1.3: Design the training process of the LSTM-Transformer driver steering angle prediction model, including the following sub-steps:

[0026] Step 2.1.3.1: Define and initialize the LSTM-Transformer driver turning angle prediction model. Set the input dimension, output length, number of LSTM hidden layer units, number of LSTM network layers, number of Transformer network layers, number of Transformer attention heads, Transformer feedforward network dimension, and Transformer attention mechanism dimension. Set the loss function, optimizer, and learning rate scheduler.

[0027] Step 2.1.3.2: In each round of training, the LSTM-Transformer driver angle prediction model receives preprocessed tensor data through forward propagation, outputs the driver's predicted angle, and calculates the error between the predicted angle and the actual angle. Subsequently, the loss function is calculated based on this error, and the network weight parameters are updated through the backpropagation algorithm. After each round of training, the performance of the current model is evaluated using an independent validation set to monitor its generalization ability and training status, effectively preventing overfitting.

[0028] Step 2.1.3.3: Save the weights that minimize the verification loss of the LSTM-Transformer driver angle prediction model as the current optimal LSTM-Transformer driver angle prediction model.

[0029] Step 2.1.3.4: After the LSTM-Transformer driver angle prediction model is trained, load the trained network model, use the test set to predict the driver's turning angle, and calculate the evaluation index.

[0030] Step 2.1.4: Design a real-time prediction module and load the optimal LSTM-Transformer driver angle prediction model obtained from the network training module. During vehicle operation, this module receives real-time information from the data acquisition module regarding the driver's steering actions, vehicle status, and the driver's desired path with the controller. Tensor data is generated through the data preprocessing module. This tensor data is then input into the loaded LSTM-Transformer driver angle prediction model. The LSTM layer extracts the temporal dynamic features of the input sequence, capturing the driver's short-term habits and state evolution. Further, the Transformer encoder layer performs self-attention weighting on the temporal features, fusing global contextual information to focus on key decision moments. Finally, the predicted driver angle is output through regression via a fully connected layer. .

[0031] Step 2.2, Design of the MPC controller corner prediction model, which includes the following sub-steps:

[0032] Step 2.2.1: Establish the prediction model. The vehicle dynamics model is used as the prediction model, as shown in equation (2):

[0033] (2)

[0034] in, , Including vehicle lateral speed Vehicle yaw rate lateral coordinates of the vehicle Vehicle yaw angle ; For the controller to turn, For road curvature, , and The state coefficient matrix, This is the output coefficient matrix.

[0035] Step 2.2.2, Objective function design: Design the following multi-objective cost function to optimize path tracking accuracy and control quantity, as shown in equation (3):

[0036] (3)

[0037] in, Let Q be the desired path for the controller, and let H be the weight matrices.

[0038] Step 2.2.3: Optimization and solution. Solve the multi-objective optimization problem shown in formula (3) in each control cycle to obtain the controller's predicted rotation angle. .

[0039] Step 2.3, Driver's trust in the controller The design is as shown in equation (4):

[0040] (4)

[0041] in, To adjust the human-machine cornering conflict sensitivity factor, Human-machine corner conflict factor This represents the acceptable level of conflict.

[0042] Step 2.3.1, the human-machine angle conflict factor is designed based on the absolute value of the difference between the driver's angle and the controller's angle, as shown in equation (5):

[0043] (5)

[0044] in, and For the driver and controller respectively, in the prediction time domain, the first... Predict the turning angle step by step. For the current moment, Predicting human-machine corner angles in the time domain.

[0045] Step 3: Driver Experience Assessment Module Design:

[0046] Based on an inverse reinforcement learning-based driver experience evaluation model, it receives vehicle status and outputs driver experience reward values. Furthermore, based on driver experience reward values Assess the controller's reliability with the driver .

[0047] Step 3.1, Design of the Inverse Reinforcement Learning Driver Experience Evaluation Model, which includes the following sub-steps:

[0048] Step 3.1.1, State Input and Feature Construction:

[0049] The inverse reinforcement learning driver experience evaluation model receives vehicle state data as input, specifically including lateral tracking error. longitudinal speed of vehicles Collision time between the vehicle and the vehicle in front Front wheel steering angle of the vehicle Furthermore, a feature vector representing driver experience is constructed, specifically including safety, tracking accuracy, longitudinal ride comfort, and lateral stability.

[0050] Step 3.1.2: Design of driver experience reward function, based on safety, tracking accuracy, longitudinal ride comfort, and lateral stability; where safety considers the collision time between the vehicle and the vehicle in front, tracking accuracy considers the lateral tracking error, longitudinal ride comfort considers the longitudinal speed change rate, and lateral stability considers the front wheel steering angle change rate, as shown in equations (6) to (11):

[0051] (6)

[0052] (7)

[0053] (8)

[0054] (9)

[0055] (10)

[0056] (11)

[0057] in, For security reward function, For tracking accuracy reward function, The reward function is for vertical smoothness. For horizontal stability reward function, and The first The collision time between the vehicle and the vehicle in front in the frame and the safe collision time threshold. For the first Lateral tracking error of the frame, For the first The rate of change of longitudinal vehicle speed in the frame, For the first The rate of change of the vehicle's front wheel steering angle in the frame. For the current frame number, The weights of each reward function, Let be the standard deviation of each reward function, where , , The total number of sampled frames is used to assess driver experience; the reward functions for safety, tracking accuracy, longitudinal ride comfort, and lateral stability are all standardized and weighted. It is learned from expert data through inverse reinforcement learning.

[0058] Step 3.1.3, the training process of the inverse reinforcement learning driver experience evaluation model, includes the following sub-steps:

[0059] Step 3.1.3.1: Construction of expert dataset. Based on a large amount of publicly available natural driving data from experienced drivers in various scenarios, an expert demonstration trajectory set is formed.

[0060] Step 3.1.3.2: Design the learning objective function and optimize iteratively to ensure that the learned reward function can explain expert behavior.

[0061] Step 3.1.3.3: Training process of the inverse reinforcement learning driver experience evaluation model. The maximum entropy inverse reinforcement learning framework is adopted. Through iterative optimization algorithm, the weights and parameters of the reward function are adjusted to maximize the expected cumulative reward of the expert policy under the function. Thus, the optimal reward function model that can explain the expert driving preferences and decision-making logic is deduced in reverse.

[0062] Step 3.1.3.4: Output the optimal driver experience reward function. After training, the model obtains a set of optimal reward function weights. This leads to the optimal driver experience reward function.

[0063] Step 3.1.3.5: Calculate the real-time reward value. Input the historical and current vehicle state data into the trained driver experience reward function to calculate the driver's experience reward value for the current time period. .

[0064] Step 3.2, Controller's trust in the driver The design is as shown in equation (12):

[0065] (12)

[0066] in, To adjust the sensitivity factor of driver experience reward value, This is a driver experience reward value. The median reward value for the dataset of experienced drivers.

[0067] Step 4, Weight Allocation Module Design:

[0068] This module assigns weights based on fuzzy rules, taking into account the driver's trust in the controller. And the controller's trust in the driver As input, the driver control weights are dynamically output.

[0069] The specific steps for designing fuzzy rules are as follows:

[0070] Step 4.1: Confirm input variable: Driver's trust level with the controller And the controller's trust in the driver .

[0071] Step 4.2, Blur processing: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] and Fuzzification, mapping to the corresponding fuzzy set .

[0072] Step 4.3: Fuzzy rule reasoning. Reasoning is performed based on preset fuzzy rules. There are nine fuzzy rules in total, including:

[0073] Article 1, if for low and If the value is low, the driver's control weight is low;

[0074] Article 2, if for low and If the driver's control weight is set to medium, then the driver's control weight is also medium.

[0075] Article 3, if for low and If the value is high, the driver's control weight is high;

[0076] Article 4, if For the middle and If the value is low, the driver's control weight is low;

[0077] Article 5, if For the middle and If the driver's control weight is set to medium, then the driver's control weight is also medium.

[0078] Article 6, if For the middle and If the value is high, the driver's control weight is high;

[0079] Article 7, if For high and If the value is low, the driver's control weight is low;

[0080] Article 8, if For high and If the driver's control weight is set to medium, then the driver's control weight is also medium.

[0081] Article 9, if For high and If the value is high, then the driver's control weight is high.

[0082] Step 4.4: Defuzzification. The centroid method is used to defuzzify the fuzzy set, converting it into precise driver control weight values. As shown in equation (13):

[0083] (13)

[0084] in, This represents the total number of fuzzy rules triggered under the input variables. For the triggered first The membership degree of a fuzzy rule. , The center value of the corresponding weight interval. .

[0085] Step 4.5: Output variable - Driver control weight Driver path tracking weights as a non-cooperative game module As shown in equation (14):

[0086] (14)

[0087] Step 5: Design of the non-cooperative game module, which includes the following sub-steps:

[0088] Step 5.1: Prediction model establishment. Construct a discretized driver-controller interaction model, as shown in equation (15):

[0089] (15)

[0090] in, Turn for the driver; For controller rotation angle; state variables , ,That , , and These are the vehicle's lateral velocity, vehicle's yaw rate, vehicle's lateral coordinate, and vehicle's yaw angle, respectively. , , , It is a coefficient matrix.

[0091] Step 5.2, Objective function design: In order to reduce human-machine decision-making conflicts and improve driving safety in the human-machine co-driving mode, while ensuring vehicle stability, the objective function is designed as shown in equation (16).

[0092] (16)

[0093] In the formula,

[0094]

[0095] in, and These are the objective functions for the driver and the controller, respectively. To control the output, and These are the desired paths for the driver and the controller, respectively. and These are the control inputs for the driver and the controller, respectively. and These are the single-step path information for the driver and the controller, respectively. and These are the single-step path tracking weights for the driver and the controller, respectively. and These are the single-step control input weights for the driver and the controller, respectively. , This represents the prediction time domain of the human-computer interaction model.

[0096] Step 5.3, Solving for Nash Equilibrium:

[0097] Step 5.3.1: First, define the error variables for the driver and the controller, as shown in equation (17):

[0098] (17)

[0099] in, and These are the error variables for the driver and the controller, respectively. The state coefficient matrix, and These are the control input coefficient matrices for the driver and the controller, respectively.

[0100] Step 5.3.2: Based on the least squares method and the convex iteration method, the optimal control input sequence of the driver and the controller can be obtained; further, the first value of each sequence is selected as the optimal solution, as shown in equation (18):

[0101] (18)

[0102] in, and These represent the optimal solutions for the driver and the controller, respectively. and These represent the driver's optimal control input sequence and the controller's optimal control input sequence, respectively.

[0103] Step 5.4 Finally, the steering of the human-machine co-driving vehicle is as shown in equation (19):

[0104] (19)

[0105] in, The final steering is performed by the human-machine co-driving vehicle.

[0106] The beneficial effects of this invention are as follows: First, this invention constructs a closed-loop, two-way trust mechanism, significantly improving the accuracy of evaluating driver intent and driving experience. Through the human-machine steering conflict assessment module, the spatiotemporal features of driving steering behavior are captured using the LSTM-Transformer driver angle prediction model and compared with the MPC controller's angle prediction results, thereby accurately quantifying the driver's trust in the controller. Simultaneously, a driver experience assessment module is introduced, which uses inverse reinforcement learning to invert the latent reward function from expert driving trajectories. Driver experience is objectively evaluated through safety, tracking accuracy, longitudinal stability, and lateral stability indicators, further assessing the controller's trust in the driver. This overcomes the limitations of traditional methods that rely solely on single indicators such as lateral tracking accuracy and lateral acceleration, and are difficult to characterize nonlinear features. Second, this method uses fuzzy rules for dynamic weight allocation, taking the trust levels of both parties as input. Fuzzy rules are used to output the driver's control weights in real time, and Nash equilibrium is solved using non-cooperative game theory, effectively mitigating human-machine conflict. Finally, this invention achieves a fundamental improvement in the human-machine interaction paradigm. The system can adjust control weights in advance based on mutual trust states, enhancing the system's adaptability in complex environments and the driver's acceptance of the intelligent system. Attached Figure Description

[0107] Figure 1 This is a schematic diagram of a cooperative steering control system that considers human-machine trust in this invention.

[0108] Figure 2 This is a flowchart of the LSTM-Transformer driver angle prediction model of the present invention.

[0109] Figure 3 This is a flowchart of the MPC controller corner prediction model of the present invention.

[0110] Figure 4 This is a flowchart illustrating the design of the inverse reinforcement learning driver experience evaluation model of the present invention. Detailed Implementation

[0111] The present invention will now be described in detail with reference to the accompanying drawings.

[0112] This invention proposes a cooperative steering control method considering human-machine mutual trust. This method comprises a closed-loop control system consisting of six modules: a desired path module, a human-machine steering conflict assessment module, a driver experience assessment module, a weight allocation module, a non-cooperative game theory module, and a human-machine co-driving vehicle module. This system enables cooperative steering control of the human-machine co-driving vehicle. Specifically, the desired path module generates the desired paths for the driver and controller; the human-machine steering conflict assessment module receives the driver's steering action, vehicle state, and the desired paths for both the driver and controller. It predicts the driver's steering angle using an LSTM-Transformer model and simultaneously predicts the controller's steering angle using an MPC model. By comparing the differences between the two predictions, the driver's steering angle is quantified. The driver's trust in the controller is assessed through a driver experience evaluation module that receives the vehicle state, extracts safety, tracking accuracy, longitudinal ride comfort, and lateral stability features from the vehicle state using an inverse reinforcement learning model, calculates the driver's experience reward value, and evaluates the controller's trust in the driver. A weight allocation module, based on fuzzy rules, takes the driver's trust in the controller and the controller's trust in the driver as inputs and outputs the driver's control weights. Finally, a non-cooperative game theory module combines the driver's and controller's desired paths, the driver's control weights, and the vehicle state, and obtains the optimal steering angle through three steps: prediction model establishment, objective function design, and Nash equilibrium solution, thus achieving cooperative steering control of the human-machine co-driving vehicle.

[0113] Step 1, Design of the Expected Path Module:

[0114] The desired path module is used to generate the driver's desired path and the controller's desired path.

[0115] Step 1.1, Driver's desired route design:

[0116] Driver's Expected Path In the joint simulation platform of CarSim and Simulink, input is obtained through the Logitech G29 steering wheel module based on the driving environment; among them, and These are the longitudinal and lateral coordinates of the driver's desired path, respectively. This represents the yaw angle of the driver's desired path.

[0117] Step 1.2, Controller Desired Path Design:

[0118] Controller Desired Path Acquired based on the CarSim simulation environment; among which, and These are the vertical and horizontal coordinates of the controller's desired path, respectively. This represents the yaw angle of the controller's desired path.

[0119] Step 2: Design of the human-machine steering conflict assessment module, including the design of the LSTM-Transformer driver angle prediction model and the design of the MPC controller angle prediction model.

[0120] Step 2.1, Design of LSTM-Transformer Driver Steering Angle Prediction Model:

[0121] Step 2.1.1: The data input module is used to collect driver steering actions and vehicle status information under normal operating conditions of the human-machine co-driving vehicle, including driver steering angle. Driver's steering angle change rate and vehicle longitudinal speed Vehicle yaw rate Longitudinal acceleration Longitudinal coordinates of the driver's desired path Horizontal coordinates of the driver's desired path Yaw angle of the driver's desired path The above information will be used as feature data input to the data preprocessing module for the LSTM-Transformer driver steering angle prediction model, to predict the desired driver steering angle. This data is then used as a label for subsequent driver-predicted steering angles. predict.

[0122] Step 2.1.2, the data preprocessing module, includes the following sub-steps:

[0123] Step 2.1.2.1: Normalize the feature data and label data using the MinMax normalization method to make the model training more stable and efficient. The MinMax normalization method preprocesses the data as shown in the invention formula (1).

[0124] Step 2.1.2.2: Use the sliding window method to select continuous... Normalized data at each time step is used as the input-output pair of the LSTM-Transformer network model, and then converted into tensor data as the input of the model.

[0125] Step 2.1.2.3: Divide the tensor data required by the LSTM-Transformer network for predicting the driver's turning angle into a training set of 80%, a validation set of 10%, and a test set of 10%.

[0126] Step 2.1.3: Design the training process of the LSTM-Transformer driver steering angle prediction model, which includes the following sub-steps:

[0127] Step 2.1.3.1: Define and initialize the LSTM-Transformer network model:

[0128] Its input is three-dimensional tensor data with dimensions of 1. ,in, Indicates batch size, Representing the feature dimension (driver's turning angle) Driver's steering angle change rate longitudinal speed of vehicles Vehicle yaw rate Longitudinal acceleration Longitudinal coordinates of the driver's desired path Horizontal coordinates of the driver's desired path Yaw angle of the driver's desired path ), This indicates the length of the input sequence (i.e., the model predicts the driver's turn angle for the next moment based on data from 20 consecutive historical moments, corresponding to a historical window of 0.4 seconds).

[0129] The network structure design is as follows:

[0130] LSTM layer: Contains a stacked LSTM structure with two layers, designed to capture the long-term and short-term dependencies of time series, with 128 hidden units in each layer.

[0131] Transformer encoder layer: Connected after the LSTM layer, used to extract global attention features; the number of attention heads is set to 4, the dimension of the feedforward network is 256, and the number of network layers is 1.

[0132] Fully connected layer: Output dimension is 1, directly outputting the predicted driver steering angle. .

[0133] The training hyperparameters are set as follows:

[0134] The loss function uses mean squared error to measure the deviation between the driver's predicted steering angle and the driver's actual steering angle.

[0135] The optimizer used is AdamW, with an initial learning rate of 0.001 and a weight decay coefficient of 0.0001.

[0136] The learning rate is scheduled using a cosine annealing strategy, with a minimum learning rate of 0.000001.

[0137] Step 2.1.3.2: Perform the model training process.

[0138] The total training period was set to 500 rounds. In each round, the normalized tensor data was input into the hybrid model, which output the driver's predicted turning angle through forward propagation and calculated its gradient with the label data. Subsequently, the network weight parameters were updated through the backpropagation algorithm.

[0139] Step 2.1.3.3: If the validation loss in the current round is minimized, save the weights of the current LSTM-Transformer driver corner prediction model and use it as the current optimal LSTM-Transformer driver corner prediction model.

[0140] Step 2.1.3.4: After the LSTM-Transformer driver angle prediction model is trained, load the trained network model, use the test set to predict the driver's turning angle, and calculate the evaluation index.

[0141] Step 2.1.4: Design a real-time prediction module and load the optimal LSTM-Transformer driver angle prediction model obtained from the network training module. During vehicle operation, this module receives real-time information from the data acquisition module regarding the driver's steering actions, vehicle status, and the driver's desired path with the controller. Tensor data is generated through a data preprocessing procedure consistent with offline training. This tensor data is then input into the loaded LSTM-Transformer driver angle prediction model. First, the LSTM layer extracts the temporal dynamic features of the input sequence to capture the driver's short-term habits and state evolution. Further, the Transformer encoder layer performs self-attention weighting on the temporal features, fusing global contextual information to focus on key decision moments. Finally, the predicted driver angle is output through a fully connected layer regression. .

[0142] Step 2.2, Design of the MPC controller corner prediction model, which includes the following sub-steps:

[0143] Step 2.2.1: Establish the prediction model. The vehicle dynamics model is used as the prediction model, as shown in the invention formula (2).

[0144] The vehicle dynamics parameters and controller parameters are set as follows:

[0145] Vehicle dynamics model prediction time domain This corresponds to a vehicle status prediction of 0.4 seconds.

[0146] Control Time Domain This means optimizing only the control quantities for the next five steps.

[0147] Step 2.2.2, Objective function design: Design the following multi-objective cost function to optimize path tracking accuracy and control quantity, as shown in the invention formula (3).

[0148] Step 2.2.3: Optimization Solution. Solve the above optimization problem in each control cycle to obtain the controller's predicted rotation angle. .

[0149] Step 2.3, Driver's trust in the controller The design is as shown in the invention description (4).

[0150] Step 2.3.1, the human-machine angle conflict factor is designed based on the absolute value of the difference between the driver's angle and the controller's angle, as shown in the invention formula (5).

[0151] Step 3: Driver Experience Assessment Module Design:

[0152] Based on an inverse reinforcement learning-based driver experience evaluation model, it receives vehicle status and outputs driver experience reward values. Furthermore, based on driver experience reward values Assess the controller's reliability with the driver .

[0153] Step 3.1, Design of the Inverse Reinforcement Learning Driver Experience Evaluation Model, which includes the following sub-steps:

[0154] Step 3.1.1, Feature Extraction and State Input:

[0155] The inverse reinforcement learning driver experience evaluation model receives vehicle state data as input, including lateral tracking error. longitudinal speed of vehicles Collision time between the vehicle and the vehicle in front Front wheel steering angle of the vehicle Furthermore, a feature vector representing driver experience is constructed, with a feature dimension of 4, including safety, tracking accuracy, longitudinal ride comfort, and lateral stability.

[0156] Step 3.1.2: Design of driver experience reward function, based on safety, tracking accuracy, longitudinal ride comfort and lateral stability; where safety considers the collision time between the vehicle and the vehicle in front, tracking accuracy considers the lateral tracking error, longitudinal ride comfort considers the longitudinal speed change rate, and lateral stability considers the front wheel steering angle change rate, as shown in the invention content equations (6) to (11).

[0157] Step 3.1.3, the training process of the inverse reinforcement learning driver experience evaluation model, includes the following sub-steps:

[0158] Step 3.1.3.1: Construction of expert dataset. Based on the publicly available NGSIM dataset, the dataset is classified by K-means clustering into three categories: novice drivers, intermediate-experience drivers, and highly experienced drivers. The highly experienced driver dataset is selected to form the expert demonstration trajectory set.

[0159] Step 3.1.3.2: Design the learning objective function. Through iterative optimization, the learned reward function should be able to explain the expert's behavior. That is, under this reward function, the expected cumulative reward of the expert policy should be significantly higher than that of other random policies.

[0160] Step 3.1.3.3: Training process of the inverse reinforcement learning driver experience evaluation model, setting the number of iterations to 200; adopting the maximum entropy inverse reinforcement learning framework, through iterative optimization algorithm, adjusting the weight parameters of the reward function to maximize the expected cumulative reward of the expert policy under the function, thereby inversely deriving the optimal reward function model that can explain the expert driving preferences and decision-making logic.

[0161] Step 3.1.3.4: Output the optimal driver experience reward function. After training, the model obtains a set of optimal reward function weights. This leads to the optimal driver experience reward function.

[0162] Step 3.1.3.5: Calculate the real-time reward value. Input the historical and current vehicle state data into the trained reward function to calculate the driver experience reward value for the current time period. .

[0163] Step 3.2, Controller's trust in the driver The design is as shown in the invention description (12).

[0164] Step 4, Weight Allocation Module Design:

[0165] This module assigns weights based on fuzzy rules, taking into account the driver's trust in the controller. And the controller's trust in the driver As input, the driver control weights are dynamically output.

[0166] The specific steps for designing fuzzy rules are as follows:

[0167] Step 4.1: Confirm input variable: Driver's trust level with the controller And the controller's trust in the driver The domain of discourse is .

[0168] Step 4.2, Blur processing: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] and Fuzzification, mapping to the corresponding fuzzy set .

[0169] The membership function is a Gaussian membership function, where the low, medium, and high values ​​of the input variables correspond to equations (20), (21), and (22), respectively:

[0170] (20)

[0171] (twenty one)

[0172] (twenty two)

[0173] in For input variables, As the standard deviation of the Gaussian function.

[0174] Step 4.3: Fuzzy rule reasoning. Reasoning is performed according to preset fuzzy rules. There are a total of nine fuzzy rules, as shown in the first to ninth rules in step 4.3 of the invention content.

[0175] Step 4.4: Defuzzification. The centroid method is used to defuzzify the fuzzy set, converting it into precise driver control weight values. As shown in the invention content formula (13).

[0176] Step 4.5: Output variable: Driver control weight Driver path tracking weights as a non-cooperative game module As shown in the invention description (14).

[0177] Step 5: Design of the non-cooperative game module, which includes the following sub-steps:

[0178] Step 5.1: Vehicle model establishment. The vehicle kinematic model is described using the vehicle coordinate system. The vehicle dynamics model only considers the lateral and yaw motions of the vehicle, while ignoring the effects of aerodynamics and suspension, thus obtaining a two-degree-of-freedom model. The vehicle kinematics and dynamics model is constructed as shown in equation (23):

[0179] (twenty three)

[0180] in, For vehicle quality; and These are the vehicle's longitudinal speed and lateral speed, respectively. The vehicle's yaw rate; and These are the lateral forces of the front and rear wheels, respectively. The steering angle of the vehicle's front wheels; and These are the distances from the vehicle's center of gravity to the front and rear axles, respectively. Let yaw moment of inertia be the moment of inertia about the vertical axis of the car's center of mass. This refers to the vehicle's yaw angle; and Here are the horizontal and vertical coordinates of the vehicle.

[0181] Step 5.2: Assuming that the tire force does not enter the nonlinear region during vehicle movement, that is, the magnitude of the tire force and the sideslip angle have a linear relationship, equation (23) is linearized as shown in equation (24):

[0182] (twenty four)

[0183] in, , These are the front and rear tire stiffness, respectively. This is the steering gear ratio.

[0184] Step 5.3: Transform equation (24) into state-space form, as shown in equation (25):

[0185] (25)

[0186] In the formula,

[0187]

[0188] in, The state matrix; , and These are the state coefficient matrix, the control input coefficient matrix, and the control output coefficient matrix, respectively.

[0189] Step 5.4: Prediction model establishment. Construct a discrete driver-controller interaction model, as shown in the invention content formula (15).

[0190] Step 5.5, Objective function design: In order to reduce human-machine decision-making conflicts and improve driving safety in the human-machine co-driving mode, and at the same time ensure vehicle stability, the objective function is designed as shown in the invention content formula (16).

[0191] Step 5.6, Solving for Nash Equilibrium:

[0192] Step 5.6.1 First, define the error variables for the driver and the controller, as shown in the invention formula (17).

[0193] Step 5.6.2: Substitute formula (17) into formula (16) to obtain the objective functions for the driver and controller, as shown in formula (26):

[0194] (26)

[0195] in, and These are the control input coefficient matrices for the driver and the controller, respectively.

[0196] Step 5.6.3: For non-cooperative game optimization control problems, the solution can be obtained by solving for the Nash equilibrium; the Nash equilibrium condition is shown in equation (27):

[0197] (27)

[0198] in, and These represent the entire control set for the driver and the controller, respectively. This is called a set of Nash equilibrium solutions.

[0199] Step 5.6.4: Based on the least squares method and the convex iteration method, the optimal control input sequence of the driver and the controller can be obtained; further, the first value of each sequence is selected as the optimal solution, as shown in the invention formula (18).

[0200] Step 6: Finally, the human-machine co-driving vehicle performs the final steering maneuver. As shown in the invention description (19).

[0201] In summary, this invention proposes a cooperative steering control method considering human-machine mutual trust. By constructing a closed-loop two-way trust perception mechanism and a trust-based dynamic weight decision framework, it achieves a fundamental shift in human-machine co-driving systems from passive conflict coordination to proactive trust collaboration. On one hand, the system utilizes an LSTM-Transformer model to accurately predict the driver's steering intentions and assesses the driver's trust in the controller in real time by comparing the predicted steering angle with that predicted by the MPC controller. On the other hand, it employs an inverse reinforcement learning model to invert the reward function from multi-dimensional driving behavior, quantifying the controller's trust in the driver's experience. Based on this, it fuses the trust levels of both parties through fuzzy rules, dynamically allocates driver control weights in real time, and introduces non-cooperative game theory to solve for Nash equilibrium, generating the optimal steering angle that integrates the intentions of both parties. This controls the human-machine co-driving vehicle to complete the cooperative steering maneuver, providing an innovative and feasible solution for the practical application of human-machine cooperative driving technology.

Claims

1. A cooperative steering control method considering human-machine mutual trust, characterized in that: This method constructs a closed-loop control system consisting of a desired path module, a human-machine steering conflict assessment module, a driver experience assessment module, a weight allocation module, a non-cooperative game theory module, and a human-machine co-driving vehicle to achieve cooperative steering control of the human-machine co-driving vehicle. Specifically, the desired path module generates the desired path between the driver and the controller; the human-machine steering conflict assessment module receives the driver's steering action, vehicle state, and the desired path between the driver and the controller, and outputs the driver's trust level with the controller; the driver experience assessment module receives the vehicle state and outputs the controller's trust level with the driver; the weight allocation module combines the driver's trust level with the controller's trust level with the controller's trust level, and outputs the driver's control weight based on fuzzy rules; the non-cooperative game theory module, based on the desired path between the driver and the controller, the driver's control weight, and the vehicle state, obtains the optimal steering angle through prediction model establishment, objective function design, and Nash equilibrium solution, thus achieving cooperative steering control of the human-machine co-driving vehicle. The LSTM-Transformer driver angle prediction model in the human-machine steering conflict assessment module receives the driver's steering actions. Vehicle status Driver's desired route Output driver's predicted steering angle The human-machine steering conflict assessment module uses an MPC controller-based angle prediction model to receive vehicle status data. Controller expected path Output controller predicts rotation angle Ultimately, the turning angle is determined based on the driver's prediction. and controller predicts the turning angle Determine the driver's level of trust in the controller. ;in, For the driver to turn the corner, For the rate of change of the driver's steering angle, and These are the vehicle's longitudinal speed and lateral speed, respectively. and These are the lateral tracking error and the yaw angle error, respectively. Let yaw rate be the vehicle's angular velocity. For road curvature, For the longitudinal acceleration of the vehicle, and These are the longitudinal and lateral coordinates of the driver's desired path, respectively. The yaw angle is the driver's desired path. and These are the vertical and horizontal coordinates of the controller's desired path, respectively. The yaw angle of the controller's desired path. The current moment; The driver's trust in the controller The design is shown below: ， in, To adjust the human-machine cornering conflict sensitivity factor, Human-machine corner conflict factor The acceptable level of conflict is defined as follows; the human-machine angle conflict factor is designed based on the absolute value of the difference between the driver's angle and the controller's angle, and the specific formula is as follows: ， in, and For the driver and controller respectively, in the prediction time domain, the first... Predict the turning angle step by step. For the current moment, Predicting human-machine corner rotation in the time domain; The driver experience assessment module includes a driver experience assessment model based on inverse reinforcement learning, which receives vehicle status data. Output driver experience reward value Ultimately, the reward value is based on the driver's experience. Assess the controller's reliability with the driver ;in, For lateral tracking error, For the longitudinal speed of the vehicle, The time of collision between the vehicle and the vehicle in front. The steering angle of the vehicle's front wheels; The controller's level of trust in the driver The design is shown below: ， in, To adjust the sensitivity factor of driver experience reward value, This is a driver experience reward value. The median reward value for the dataset of experts with extensive driving experience; The driver experience reward function is designed based on safety, tracking accuracy, longitudinal ride comfort, and lateral stability. Safety considers the collision time between the vehicle and the vehicle in front; tracking accuracy considers the lateral tracking error; longitudinal ride comfort considers the longitudinal speed change rate; and lateral stability considers the front wheel steering angle change rate. The specific formulas are as follows: ， ， ， ， ， in, For security reward function, For tracking accuracy reward function, The reward function is for vertical smoothness. For horizontal stability reward function, and The first The collision time between the vehicle and the vehicle in front in the frame and the safe collision time threshold. For the first Lateral tracking error of the frame, For the first The rate of change of longitudinal vehicle speed in the frame, For the first The rate of change of the vehicle's front wheel steering angle in the frame. For the current frame number, The weights of each reward function, Let be the standard deviation of each reward function, where , , Total number of sampling frames for driver experience assessment; The weight allocation module assigns weights based on fuzzy rules, taking into account the driver's trust in the controller. And the controller's trust in the driver To dynamically allocate control weights and output driver control weights Driver path tracking weights as a non-cooperative game module ; Finally, the non-cooperative game theory module combines the driver's and controller's desired paths, the driver's control weights, and the vehicle's state. Through prediction model establishment, objective function design, and Nash equilibrium solution, the optimal turning angle is obtained, thereby achieving cooperative steering control of the human-machine co-driving vehicle.

Images