Car following control method and related product

Abstract

The invention discloses a car following control method and a related product. According to the scheme, multi-source data are acquired, and the multi-source data are fused to obtain a fusion result; based on the fusion result, performing multi-modal trajectory prediction on the driving behavior of the preceding vehicle to obtain a prediction result of the driving behavior of the preceding vehicle; the multi-modal trajectory prediction comprises longitudinal motion prediction and transverse motion prediction; quantifying a car following risk by using a multi-dimensional risk index to obtain a risk level; constructing a multi-target cost function based on the fusion result, the front vehicle driving behavior prediction result and a control vector, and solving the multi-target cost function by using a preset constraint condition to obtain an initial control parameter of the vehicle; and on the basis of the risk level, initial control parameters of the vehicle are adjusted, and target control parameters of the vehicle are obtained. Compared with the problem of response lag in adaptive cruise control in the prior art, the adaptive cruise control method has obvious advantages.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, and in particular to a vehicle following control method and related products. Background Technology

[0002] With the continuous evolution of intelligent connected vehicle technology, the information interaction capabilities between vehicles have been significantly enhanced, and vehicle-to-vehicle communication (V2V) technology provides important support for achieving a higher level of cooperative driving.

[0003] Traditional adaptive cruise control (ACC) systems mainly rely on local sensors, such as millimeter-wave radar or cameras, to obtain information about the status of vehicles ahead. The control logic of ACC is mostly based on a feedback mechanism, that is, it only responds after detecting that the vehicle ahead is slowing down, which has obvious lag problems and makes it difficult to achieve active safety control.

[0004] How to control vehicles to improve the speed of risk identification in following vehicle control and ensure the safety of following vehicle is an urgent technical problem to be solved. Summary of the Invention

[0005] In view of the above problems, this application provides a vehicle following control method and related products, the purpose of which is to control vehicles, improve the risk identification speed of vehicle following control, and ensure vehicle following safety.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] The first aspect of this application provides a vehicle following control method, the method comprising:

[0008] Acquire multi-source data, fuse the multi-source data, and obtain a fusion result; the multi-source data includes local sensing data, preceding vehicle motion data, and current vehicle motion data.

[0009] Based on the fusion results, multimodal trajectory prediction is performed on the driving behavior of the vehicle in front to obtain the driving behavior prediction result of the vehicle in front; the multimodal trajectory prediction includes longitudinal motion prediction and lateral motion prediction.

[0010] The risk of following another vehicle is quantified using multi-dimensional risk indicators to obtain a risk level.

[0011] Based on the fusion result, the predicted driving behavior of the preceding vehicle, and the control vector, a multi-objective cost function is constructed. The multi-objective cost function is solved using preset constraints to obtain the initial control parameters of the vehicle.

[0012] Based on the risk level, the initial control parameters of the vehicle are adjusted to obtain the target control parameters of the vehicle.

[0013] Optionally, the step of performing multimodal trajectory prediction on the driving behavior of the vehicle in front based on the fusion result to obtain the prediction result of the driving behavior of the vehicle in front includes:

[0014] Based on the fusion results, longitudinal motion prediction is performed on the driving behavior of the vehicle in front, and the longitudinal motion prediction result of the vehicle in front is obtained.

[0015] Based on the fusion results, the lateral motion prediction of the vehicle in front is performed on the driving behavior of the vehicle in front, and the lateral motion prediction result of the vehicle in front is obtained.

[0016] The prediction results of the lateral motion and longitudinal motion of the vehicle in front are processed using a Gaussian mixture model to obtain the prediction results of the driving behavior of the vehicle in front.

[0017] Optionally, the step of predicting the longitudinal motion of the preceding vehicle based on the fusion result to obtain the longitudinal motion prediction result of the preceding vehicle includes:

[0018] Based on the actual speed of the vehicle in front at the current moment and the acceleration prediction results at a preset time step, the longitudinal motion prediction results of the vehicle in front are obtained;

[0019] The acceleration prediction result with the preset time step is obtained based on the actual acceleration of the preceding vehicle, the rate of change of the preceding vehicle's acceleration, the braking intention, and the traffic conditions ahead; the braking intention is obtained based on the preceding vehicle's acceleration and the preceding vehicle's braking state.

[0020] Optionally, based on the fusion result, the lateral motion prediction of the preceding vehicle's driving behavior is performed to obtain the preceding vehicle's lateral motion prediction result, including:

[0021] Based on multi-dimensional information, a logistic regression model is used to identify the lane-changing intention of the vehicle in front, and the lane-changing intention identification result is obtained; the multi-dimensional information includes turn signal status, lateral offset and lane line crossing time.

[0022] The driving trajectory of the vehicle in front is predicted based on the lateral motion parameters of the vehicle in front at the current moment, and the initial prediction result of the lateral motion of the vehicle in front is obtained. The lateral motion parameters of the vehicle in front include the predicted lateral position of the vehicle in front, the actual lateral position of the vehicle in front, the speed of the vehicle in front, the heading angle of the vehicle in front, and the lateral acceleration.

[0023] Based on the recognition result of the preceding vehicle's lane-changing intention, the lateral acceleration is adjusted to obtain the prediction result of the preceding vehicle's lateral movement target.

[0024] Optionally, the step of acquiring multi-source data and fusing the multi-source data to obtain a fusion result includes:

[0025] The multi-source data is processed; the processing includes time alignment, coordinate transformation, and data validity verification.

[0026] A state vector is designed based on processed multi-source data, and a nonlinear state transition function is used to predict the state to obtain the state prediction result; the state vector includes multiple vehicle-following control parameters.

[0027] The weight of each following control parameter in the state vector is quantified by information entropy to obtain the weight allocation result. The state vector and the state estimation result are then fused based on the weight allocation result to obtain the fusion result.

[0028] Optionally, the method of quantifying the risk of following another vehicle using multi-dimensional risk indicators to obtain a risk level includes:

[0029] The urgency of a collision is quantified using a time-to-collision index to obtain the collision time.

[0030] The braking safety distance index is used to quantify reaction time and braking performance, thus obtaining the braking safety distance.

[0031] Multiple risk factors are quantified using wind potential energy field indicators to obtain risk potential energy values;

[0032] The collision time, the braking safety distance, and the risk potential energy value are weighted and summed to obtain the risk value and the risk level corresponding to the risk value.

[0033] Optionally, adjusting the initial control parameters of the vehicle based on the risk level to obtain the target control parameters of the vehicle includes:

[0034] Based on the risk level, the weights corresponding to the initial control parameters of the vehicle are adjusted using a preset adaptive weight adjustment function to obtain the adjusted weight allocation result.

[0035] Based on the adjusted weight allocation results, the multi-objective cost function is solved to obtain the target control parameters of the vehicle.

[0036] A second aspect of this application provides a following control device, the device comprising:

[0037] The data acquisition and fusion module is used to acquire multi-source data, fuse the multi-source data, and obtain a fusion result; the multi-source data includes local sensing data, preceding vehicle motion data, and current vehicle motion data.

[0038] The preceding vehicle driving behavior prediction module is used to perform multimodal trajectory prediction on the preceding vehicle driving behavior based on the fusion result, and obtain the preceding vehicle driving behavior prediction result; the multimodal trajectory prediction includes longitudinal motion prediction and lateral motion prediction;

[0039] The risk level determination module is used to quantify the risk of following a vehicle using multi-dimensional risk indicators to obtain the risk level.

[0040] The initial control parameter determination module for the vehicle is used to construct a multi-objective cost function based on the fusion result, the prediction result of the driving behavior of the preceding vehicle, and the control vector, and to solve the multi-objective cost function using preset constraints to obtain the initial control parameters of the vehicle.

[0041] The vehicle target control parameter determination module is used to adjust the initial control parameters of the vehicle based on the risk level to obtain the vehicle's target control parameters.

[0042] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the following control method provided in any implementation of the first aspect.

[0043] The fourth aspect of this application provides a processor for running a computer program that, when running, executes a vehicle following control method as provided in any implementation of the first aspect.

[0044] Compared with the prior art, this application has the following beneficial effects:

[0045] The following control method provided in this application acquires multi-source data and fuses the multi-source data to obtain a fusion result. The multi-source data includes local perception data, preceding vehicle motion data, and the vehicle's own motion data. Fusing multi-source data helps improve the accuracy of state estimation and provides reliable input for subsequent prediction and control. Based on the fusion result, multimodal trajectory prediction is performed on the preceding vehicle's driving behavior to obtain a preceding vehicle driving behavior prediction result. The multimodal trajectory prediction includes longitudinal motion prediction and lateral motion prediction. By performing multimodal trajectory prediction on the preceding vehicle's driving behavior, proactive prediction of the preceding vehicle's future behavior helps to identify potential following risks in advance, significantly improving following safety. Following risks are quantified using multi-dimensional risk indicators to obtain a risk level. Based on the fusion result, the preceding vehicle driving behavior prediction result, and the control vector, a multi-objective cost function is constructed. The multi-objective cost function is solved using preset constraints to obtain the vehicle's initial control parameters. Based on the risk level, the vehicle's initial control parameters are adjusted to obtain the vehicle's target control parameters. By quantifying the risks of following another vehicle using multi-dimensional risk indicators, the risk level is obtained, and the vehicle's control parameters are adaptively adjusted according to the risk level. This transforms risk identification from a reactive response to a proactive prediction, significantly improving the sensitivity of risk perception. It also provides a time window for determining vehicle control parameters, enabling a fast and stable safety response and ensuring the safety of following another vehicle. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart of a vehicle following control method provided in an embodiment of this application;

[0048] Figure 2 This is a schematic diagram of a vehicle following control device provided in an embodiment of this application. Detailed Implementation

[0049] As described above, current traditional adaptive cruise control (ACC) systems mainly rely on local sensors, such as millimeter-wave radar or cameras, to obtain information about the status of vehicles ahead. The control logic of ACC is mostly based on a feedback mechanism, that is, it only responds after detecting that the vehicle ahead is slowing down, which has obvious lag problems and makes it difficult to achieve active safety control.

[0050] In view of the above problems, this application proposes a following control method and related products, which acquires multi-source data, fuses the multi-source data to obtain a fusion result; the multi-source data includes local perception data, preceding vehicle motion data, and the vehicle's own motion data; based on the fusion result, multimodal trajectory prediction is performed on the preceding vehicle's driving behavior to obtain a preceding vehicle driving behavior prediction result; the multimodal trajectory prediction includes longitudinal motion prediction and lateral motion prediction; following risk is quantified using multi-dimensional risk indicators to obtain a risk level; a multi-objective cost function is constructed based on the fusion result, the preceding vehicle driving behavior prediction result, and a control vector, and the multi-objective cost function is solved using preset constraints to obtain the vehicle's initial control parameters; based on the risk level, the vehicle's initial control parameters are adjusted to obtain the vehicle's target control parameters.

[0051] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0052] See Figure 1This figure is a flowchart of a vehicle following control method provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0053] S101. Acquire multi-source data, fuse the multi-source data, and obtain the fusion result.

[0054] The multi-source data includes local perception data, preceding vehicle motion data, and current vehicle motion data.

[0055] In one feasible implementation:

[0056] The multi-source data is processed; the processing includes time alignment, coordinate transformation, and data validity verification.

[0057] A state vector is designed based on processed multi-source data, and a nonlinear state transition function is used to predict the state to obtain the state prediction result; the state vector includes multiple vehicle-following control parameters.

[0058] The weight of each following control parameter in the state vector is quantified by information entropy to obtain the weight allocation result. The state vector and the state estimation result are then fused based on the weight allocation result to obtain the fusion result.

[0059] System state vector Including sensor state vectors V2V communication state vector and vehicle state vector The formula is as follows:

[0060] ;

[0061] Sensor state vector Stores local sensing data from radar and cameras, including radar ranging. Relative velocity Relative angle and camera data and The formula is as follows:

[0062] ;

[0063] V2V Communication State Vector Used to transmit the dynamic information of the vehicle ahead, including its position. ,speed acceleration Heading angle Braking state and turn signals The formula is as follows:

[0064] ;

[0065] Vehicle state vector Record the vehicle's motion parameters, including its speed. acceleration Heading angle and yaw rate The formula is as follows:

[0066] ;

[0067] Time synchronization is fundamental to multi-source data fusion. To address the discrepancy between sampling frequencies and timestamps from radar, cameras, and V2V communication, a weighted least squares estimation algorithm is used to establish a unified time reference, as shown in the following formula:

[0068] ;

[0069] in, Indicates a unified time reference point; represents the original local timestamp recorded by the i-th sensor; t represents the global synchronization time to be optimized; This represents the transmission delay experienced by the i-th sensor data from its generation to its reception; The weight of the i-th sensor is represented by ; N is the total number of sensors participating in time synchronization.

[0070] To achieve optimal time alignment, the weighting coefficients need to be dynamically adjusted based on sensor accuracy and reliability:

[0071] ;

[0072] Among them, the weighting coefficient Based on sensor accuracy and delay uncertainty Dynamic adjustments are made to ensure that high-precision sensors play a dominant role in time synchronization.

[0073] The weighted least squares estimation algorithm achieves optimal time alignment by integrating measurement error and transmission delay uncertainty, assigning greater weight to high-precision and low-latency sensors.

[0074] Coordinate transformation is achieved by using a rotation matrix to convert between global and local coordinate systems, as shown in the following formula:

[0075] ;

[0076] in, This indicates the coordinates of the preceding vehicle in the local coordinate system of this vehicle; , and These represent the angles around the Z-axis (heading angle). Y-axis (pitch angle) ) and X-axis (roll angle) The rotation matrix is ​​used to describe the vehicle's attitude; This indicates the coordinates of the vehicle in front in the Earth-Centered Earth-Fixed (ECEF) coordinate system; This indicates the coordinates of the vehicle in the ECEF coordinate system; This indicates the offset of the sensor's installation position in the local coordinate system of the vehicle.

[0077] The aforementioned transformation, through translation, rotation, and sensor offset compensation, converts the position of the preceding vehicle in the global coordinate system ECEF to the local coordinate system host of the current vehicle. This achieves spatial alignment of multi-source sensor data, such as radar, cameras, and V2V communication, ensuring high-precision spatial alignment under complex driving conditions. This is the foundation for realizing multi-source perception fusion and collaborative control.

[0078] The validity of V2V communication data is verified by constructing a state-space model based on the signal-to-noise ratio and effective range to achieve fault isolation and initial reliability assessment. The formula is as follows:

[0079] ;

[0080] in, This represents the valid representation of the i-th V2V communication data packet, where 1 indicates valid and 0 indicates invalid. If the i-th V2V communication data packet satisfies the signal-to-noise ratio... Greater than the signal-to-noise ratio threshold and distance Within the preset distance range , If the i-th V2V communication data packet is within the range of ], then the i-th V2V communication data packet is valid, providing high-quality input for subsequent fusion.

[0081] Based on multi-source data, the state vector formula is designed as follows:

[0082] ;

[0083] Where d represents the relative distance between the vehicle in front and the vehicle in front; Represents relative velocity; Indicates the acceleration of the vehicle in front; Indicates the acceleration of this vehicle; Indicates a relative angle; It represents angular velocity.

[0084] Based on the above definition of state vectors, state prediction based on kinematic models uses nonlinear state transition functions to describe the dynamic evolution of the system state:

[0085] ;

[0086] Based on the state at the current time k, predict the state at the next time k+1. Indicates a time interval; Indicates the noise term. This represents the process noise covariance matrix.

[0087] The Jacobian matrix is ​​used to locally linearize nonlinear systems, as shown in the following formula:

[0088] ;

[0089] Among them, the state transition function Taking the partial derivative with respect to the state vector X, we obtain the Jacobian matrix. Jacobian matrix It describes how changes in each state variable affect other state variables. The original state transition function is nonlinear, but it is approximated as a linear system using the Jacobian matrix, which facilitates subsequent calculation of covariance propagation.

[0090] To improve filtering accuracy, the adaptive process noise covariance mechanism dynamically adjusts according to system characteristics:

[0091] ;

[0092] in, It is the process noise covariance matrix at time k; Indicates basic process noise; It is the information covariance, which reflects the degree of difference between the predicted and observed values. yes Weighting coefficients; This is a motion detection item used to identify whether the vehicle in front has undergone drastic movements, such as changes in acceleration and angular velocity. yes The weighting coefficients.

[0093] Extended Kalman Filtering (EPF) predicts the system state through a state transition function and then corrects it using observations to achieve optimal state estimation. Compared to traditional Kalman Filtering, EPF handles nonlinear systems using the Jacobian matrix, adapting to the actual characteristics of vehicle motion. Its adaptive noise covariance mechanism dynamically adjusts process noise based on innovation and maneuver detection results, improving smoothness when the system is stationary and enhancing tracking capability during maneuvers.

[0094] After completing the state prediction, the reliability of the data is quantified using information entropy to obtain the sensor confidence level, as shown in the following formula:

[0095] ;

[0096] in, This represents the confidence level of the i-th sensor data, with a value range of [0,1]. This represents the theoretical maximum information entropy of the data from the i-th sensor. Let represent the information entropy of the data from the i-th sensor. The lower the entropy value, the higher the data reliability. The formula is as follows:

[0097] ;

[0098] in, This represents the data observed by the i-th sensor. The probability distribution; express The logarithm of is used to measure the amount of information.

[0099] Based on the information entropy quantization results, the weights are optimized according to the characteristics of different sensors. The radar weights comprehensively consider sensor confidence and ranging performance, while the V2V weights comprehensively consider communication confidence and transmission delay. The formulas are as follows:

[0100] ;

[0101] ;

[0102] in, This represents the weight of the i-th radar sensor; Indicates the confidence level of radar measurement data; This indicates the radar's ranging performance indicators, which are related to the detection range; This represents the sum of the products of the confidence scores and ranging performance metrics of all radar sensors, used for normalization. This represents the weight of the i-th V2V communication data packet; Indicates the confidence level of V2V communication data packets; Represents the reciprocal of the transmission delay of V2V communication data packets; This represents the sum of the products of the confidence level and the reciprocal of the transmission delay for all V2V communication data packets, used for normalization.

[0103] Combining the radar and V2V weights mentioned above, the final fusion result integrates the advantages of each sensor:

[0104] ;

[0105] in, The fusion result indicates that it is a weighted fusion of multi-source data, including radar and V2V data. This represents the observation state vector of the i-th sensor; express The weights; This represents the predicted system state vector at the current moment. express The weight.

[0106] The fusion weight algorithm based on information entropy uses information entropy... The uncertainty of sensor data is quantified to achieve multi-source data fusion. Confidence assessment converts entropy values ​​into a confidence index within the range of [0,1]. Dynamic weight allocation integrates sensor ranging performance (radar) and communication latency (V2V) to assign appropriate weights to different sensors. The final fusion result utilizes both measured data from each sensor and, when reliability is insufficient, relies on predicted values, ensuring the system's robustness under various operating conditions.

[0107] The V2V intent-aware predictive control architecture breaks through the limitations of traditional feedback control, enhances beyond-line-of-sight perception capabilities, effectively shortens risk identification time, improves collision avoidance success rate, and ensures driving safety.

[0108] S102. Based on the fusion result, perform multimodal trajectory prediction on the driving behavior of the vehicle in front to obtain the prediction result of the driving behavior of the vehicle in front.

[0109] The multimodal prediction includes longitudinal motion prediction and lateral motion prediction.

[0110] In one feasible implementation:

[0111] Based on the fusion results, longitudinal motion prediction is performed on the driving behavior of the vehicle in front, and the longitudinal motion prediction result of the vehicle in front is obtained.

[0112] Based on the fusion results, the lateral motion prediction of the vehicle in front is performed on the driving behavior of the vehicle in front, and the lateral motion prediction result of the vehicle in front is obtained.

[0113] The prediction results of the lateral motion and longitudinal motion of the vehicle in front are processed using a Gaussian mixture model to obtain the prediction results of the driving behavior of the vehicle in front.

[0114] In one feasible implementation, based on the fusion result, longitudinal motion prediction is performed on the driving behavior of the vehicle in front to obtain the longitudinal motion prediction result of the vehicle in front, including:

[0115] Based on the actual speed of the vehicle in front at the current moment and the acceleration prediction results at a preset time step, the longitudinal motion prediction results of the vehicle in front are obtained;

[0116] The acceleration prediction result with the preset time step is obtained based on the actual acceleration of the preceding vehicle, the rate of change of the preceding vehicle's acceleration, the braking intention, and the traffic conditions ahead; the braking intention is obtained based on the preceding vehicle's acceleration and the preceding vehicle's braking state.

[0117] Longitudinal motion prediction is based on vehicle dynamics principles, and the formula is as follows:

[0118]

[0119] in, Indicates that the car in front is Predicting speed at any given moment; This represents the actual speed of the vehicle in front at the current time t; The function representing the predicted acceleration of the vehicle in front; Indicates t to any intermediate moment; This indicates the prediction time step. Indicates that the car in front is Predicted acceleration at any given moment; This represents the actual acceleration of the vehicle in front at the current time t; This represents the rate of change of the acceleration of the vehicle in front; This represents the braking intention value, with a range of [0,1]. express The weights; This indicates the traffic condition ahead, with a value range of [0,1]. express The weight.

[0120] The braking intention function is smoothed using the Sigmoid function, as shown in the following formula:

[0121]

[0122] in, Indicates the acceleration of the vehicle in front; Indicates the acceleration threshold; This indicates the braking status of the vehicle in front; for example, if the vehicle in front has applied the brakes, then... =1, otherwise It is 0.

[0123] Longitudinal motion prediction is based on Newton's laws of motion. It predicts velocity by integrating acceleration, and then integrates velocity to obtain position. Acceleration prediction integrates current acceleration, its trend, braking intent obtained from V2V communication, and traffic conditions ahead. The braking intent function uses the sigmoid function to convert discrete braking states into continuous intent intensity, avoiding abrupt changes in prediction results and improving smoothness.

[0124] In one feasible implementation, based on the fusion result, lateral motion prediction is performed on the driving behavior of the vehicle in front to obtain the lateral motion prediction result of the vehicle in front, including:

[0125] Based on multi-dimensional information, a logistic regression model is used to identify the lane-changing intention of the vehicle in front, and the lane-changing intention identification result is obtained; the multi-dimensional information includes turn signal status, lateral offset and lane line crossing time.

[0126] The driving trajectory of the vehicle in front is predicted based on the lateral motion parameters of the vehicle in front at the current moment, and the initial prediction result of the lateral motion of the vehicle in front is obtained. The lateral motion parameters of the vehicle in front include the predicted lateral position of the vehicle in front, the actual lateral position of the vehicle in front, the speed of the vehicle in front, the heading angle of the vehicle in front, and the lateral acceleration.

[0127] Based on the recognition result of the preceding vehicle's lane-changing intention, the lateral acceleration is adjusted to obtain the prediction result of the preceding vehicle's lateral movement target.

[0128] In lateral motion prediction, a logistic regression model is used to identify lane-change intentions. The formula is as follows:

[0129] ;

[0130] in, This represents the probability that the vehicle in front intends to change lanes at the current moment. It is a bias term, representing the basic lane-changing tendency when there is no feature input; Indicates the status of the turn signal; express The weights; It is the lateral offset, which represents the lateral position deviation of the vehicle in front relative to the center line of the lane; express The weights; Indicates the time required to cross the lane lines; express The weight.

[0131] Based on the lane change intention recognition results, lateral motion prediction considers the vehicle's lateral dynamics:

[0132] ;

[0133] in, Indicates that the car in front is Predicted lateral position at any given time; This indicates the actual lateral position of the vehicle in front at the current time t; This represents the speed of the vehicle in front at the current time t; Indicates the heading angle of the vehicle in front; Indicates the prediction time step; This represents the lateral acceleration at the current time t.

[0134] lateral acceleration Determined by the principle of centripetal acceleration, the formula is as follows:

[0135] ;

[0136] in, This represents the speed of the vehicle in front at the current time t; Indicates the radius of curvature of the path of the preceding vehicle; This represents the probability that the vehicle in front intends to change lanes at the current moment.

[0137] Lateral motion prediction is based on vehicle kinematics, taking into account the influence of heading angle on lateral displacement. It adjusts lateral acceleration by combining the lane change intention recognition results to achieve accurate trajectory prediction. This can improve the recognition speed of the preceding vehicle's lane change intention and provide the control system with predictive time.

[0138] The Gaussian mixture model is used to describe the multimodal trajectory, as shown in the following formula:

[0139] ;

[0140] in, Represents the conditional probability density function; Represents the future state vector; This represents the current state vector; K is the number of Gaussian components, representing the number of driving behavior patterns; It is the weight of the k-th Gaussian component, representing the prior probability of the k-th driving behavior pattern occurring; and Let represent the mean vector and covariance matrix of the k-th Gaussian component, respectively. The calculation formulas are as follows:

[0141] ;

[0142] in, This represents the state transition matrix for the k-th driving behavior mode; This represents the bias term for the k-th driving behavior pattern; Represents the fundamental covariance matrix; Let represent the uncertainty matrix of the k-th driving behavior mode.

[0143] Multimodal trajectory prediction considers the uncertainty of driving behavior and uses a Gaussian mixture model to probabilistically describe multiple possible future trajectories. Each Gaussian component represents a typical driving behavior pattern, such as lane keeping, changing lanes left or right, etc. The mean and covariance matrices describe the expected value and uncertainty range of the trajectory, respectively. This probabilistic prediction method is more consistent with the actual driving environment than deterministic prediction, providing richer decision support information for risk assessment.

[0144] By performing multimodal trajectory prediction of the driving behavior of the vehicle ahead, and proactively predicting its future behavior, potential following risks can be identified in advance, significantly improving following safety. This perception of the vehicle's intentions makes collision avoidance strategies more proactive, significantly improving the effectiveness of active safety systems and enabling them to effectively cope with unexpected situations in complex traffic scenarios.

[0145] S103. Quantify the risk of following another vehicle using multi-dimensional risk indicators to obtain the risk level.

[0146] In one feasible implementation:

[0147] The urgency of a collision is quantified using a time-to-collision index to obtain the collision time.

[0148] The braking safety distance index is used to quantify reaction time and braking performance, thus obtaining the braking safety distance.

[0149] Multiple risk factors are quantified using wind potential energy field indicators to obtain risk potential energy values;

[0150] The collision time, the braking safety distance, and the risk potential energy value are weighted and summed to obtain the risk value and the risk level corresponding to the risk value.

[0151] Time to Collision (TTC) is used to quantify the urgency of a collision, and the formula is as follows:

[0152] ;

[0153] Where d represents the relative distance between the current vehicle and the vehicle in front; Relative speed is the difference between the speed of the vehicle in front and the speed of the vehicle in front. When... When the value is greater than 0, the vehicle is approaching the vehicle in front, posing a collision risk; when... Off-peak times refer to situations where the object is neither too close nor too far away, posing no risk of collision.

[0154] Braking Safety Distance (BSD) takes into account both reaction time and braking performance, and is calculated using the following formula:

[0155] ;

[0156] in, Represents relative velocity; Indicates the driver's reaction time; This indicates the maximum deceleration of the vehicle; Indicates the speed of the vehicle in front; This indicates the maximum deceleration of the vehicle in front.

[0157] The risk potential field integrates multiple risk factors, and the formula is as follows:

[0158] ;

[0159] in, The value represents the total risk potential energy; the larger the value, the higher the risk. d represents the relative distance between the current vehicle and the vehicle in front. Represents the normalized distance constant; Represents relative velocity; Indicates the acceleration of the vehicle in front; It is a road curvature risk item, reflecting the difficulty of handling and the impact of blind spots caused by curves; , , and These are weighting coefficients, reflecting the importance of each risk factor.

[0160] Multi-dimensional risk indicators quantify following risks from different perspectives: TTC (Time to Collision) intuitively reflects the collision time calculated based on the current relative speed; BSD (Brake Detection) is based on braking dynamics, considering driver reaction time, the vehicle's maximum braking capacity, and the braking performance of the vehicle ahead, calculating the safe braking distance required to avoid a collision; the risk potential energy field draws on physics concepts, unifying distance risk, speed risk, preceding vehicle behavior risk, and road curvature risk into a potential energy value, achieving comprehensive risk assessment. These multi-dimensional risk indicators complement each other, comprehensively covering various risk scenarios.

[0161] The risk scoring function obtains a risk value by weighted and fused multi-dimensional risk indicators, as shown in the following formula:

[0162] ;

[0163] in, This represents the total risk value; This represents the TTC component function, reflecting the collision time calculated based on the current relative velocity. for The weights; This represents the relative motion risk function. for The weights; The function representing the risk of driving intention. for The weights; Represents the environmental risk function. for The weight.

[0164] The TTC component function adopts a piecewise cosine transition form, as shown in the following formula:

[0165] ;

[0166] Here, x is an input variable representing the TTC value in seconds. The TTC value is mapped to a smooth risk function to avoid mutations.

[0167] Relative motion risk function Based on the principles of dynamics, the calculation formula is as follows:

[0168] ;

[0169] in, d represents the relative speed; d represents the relative distance between the current vehicle and the vehicle in front. This indicates the maximum deceleration of the vehicle.

[0170] Driving Intent Risk Function The calculation formula is as follows:

[0171] ;

[0172] in, It indicates the intensity of braking intent based on the acceleration and braking state of the vehicle in front; This represents the probability of changing lanes.

[0173] Braking Intent Strength The calculation formula is as follows:

[0174] ;

[0175] in, Indicates the acceleration of the vehicle in front; This indicates the threshold for recognizing braking intent; Indicates the brake pedal status; when the brake is not pressed. The value is 0 when the brake is applied. The value is 1.

[0176] Lane change probability The calculation formula is as follows:

[0177] ;

[0178] in, It is the intercept term, which represents the baseline lane-changing probability and reflects the natural lane-changing probability when there is no signal. This indicates the status of the turn signal. If the turn signal is not activated, The value is 0 if the turn signal is turned on. =1; express The state; This indicates the lateral offset, measured in meters. It represents the lateral distance between the vehicle's centerline and the lane's centerline. A positive value indicates a rightward offset, while a negative value indicates a leftward offset. express The weight.

[0179] Environmental risk function The calculation formula is as follows:

[0180] ;

[0181] in, This represents the ratio of the road surface adhesion coefficient to that of a dry road surface; This indicates rainfall intensity, with a value ranging from 0 to 100 mm / h. This represents traffic density, with a value ranging from 0 to 150 veh / km.

[0182] The comprehensive risk level assessment unifies multi-dimensional risk indicators into a single score through weighted summation, facilitating control decisions. The TTC component function is assigned a value of 1 in the danger zone when TTC < 3s and a value of 0 in the safe zone when TTC > 10s, resulting in a smooth transition. The relative motion risk function is based on the braking distance formula, reflecting the ratio of the required braking distance to the actual distance at the current relative speed. This design ensures the continuity of risk assessment and avoids abrupt changes in control caused by level jumps.

[0183] S104. Based on the fusion result, the predicted driving behavior of the preceding vehicle, and the control vector, a multi-objective cost function is constructed. The multi-objective cost function is solved using preset constraints to obtain the initial control parameters of the vehicle.

[0184] For example, the state vector of Model Predictive Control (MPC) selects a portion of the parameters from the fusion result, as shown in the following formula:

[0185] ;

[0186] Where d represents the relative distance between the vehicle in front and the vehicle in front; Represents relative velocity; Indicates the speed of this vehicle; y represents the vehicle's acceleration; y represents the lateral position offset. Indicates the heading angle; This represents the rate of change of heading angle.

[0187] The control vector is defined as:

[0188] ;

[0189] in, This indicates a longitudinal acceleration command. This indicates a lateral steering command, enabling omnidirectional vehicle control through control vectors.

[0190] Based on discrete system dynamics, the longitudinal motion prediction model is expressed as follows:

[0191] ;

[0192] in, This represents the relative distance at the k-th step. This represents the relative velocity at the k-th step; This represents the speed of the vehicle at step k; This represents the vehicle's acceleration at step k; Represents the time constant of the longitudinal dynamic system; It is the sampling time, representing the time step size for each step; This indicates the longitudinal acceleration command for the k-th step; This represents the acceleration of the vehicle in front at step k; This represents the relative distance at the (k+1)th step. This represents the relative velocity at step k+1; This represents the speed of the vehicle at step k+1. This represents the speed of the vehicle at step k+1.

[0193] The lateral motion prediction model is represented as follows:

[0194] ;

[0195] in, This indicates the lateral position offset at step k; This represents the heading angle at the k-th step; This represents the rate of change of the heading angle at the k-th step; This indicates the lateral position offset at step k+1; This represents the heading angle at step k+1; This represents the rate of change of the heading angle at the (k+1)th step; This indicates the lateral steering command at step k; Indicates the speed of this vehicle; The lateral dynamic system time constant is represented by L; the vehicle wheelbase is represented by L. This indicates the sampling time, and this indicates the time step size for each step.

[0196] The multi-objective cost function is defined as:

[0197] ;

[0198] in, It represents the total cost within the prediction time domain; N represents the length of the prediction time domain, i.e., the number of steps; This represents the stage cost function at step k, which measures the deviation between the current state and the control command. This represents the terminal cost function, which is typically a penalty term for the final state deviating from the expected value.

[0199] Among them, the stage cost The specific form is:

[0200] ;

[0201] in, This represents the expected following distance; in the weight matrix, yes The weighting measures the severity of the penalty for deviating from the expected safe following distance; yes The weighting measures the severity of the penalty for the relative speed between the vehicle and the vehicle in front; yes The weighting measures the severity of the penalty for the vehicle's acceleration. The weight of y measures the severity of the penalty for the vehicle deviating from the center of the lane. yes The weight of the heading angle measures the severity of the penalty. yes The weight of the longitudinal acceleration command measures the intensity of the penalty. yes The weight of the lateral turn instruction measures the severity of the penalty.

[0202] The state vector selects four variables: distance, relative speed, vehicle speed, and acceleration, to comprehensively describe the following state. The control vector uses jerk instead of direct acceleration, which facilitates the application of comfort constraints. The multi-objective cost function simultaneously optimizes four indices: following distance tracking, relative speed suppression, acceleration smoothing, and jerk comfort, achieving performance trade-offs through a weight matrix.

[0203] The preset constraints include safety constraints, actuator constraints, and stability constraints.

[0204] Safety constraints are expressed as follows:

[0205] ;

[0206] in, This represents the relative distance at the k-th step. Indicates the minimum safe distance threshold; Indicates the system response time; This represents the speed of the vehicle at step k; This represents the positive part of the relative velocity at step k; This indicates the maximum deceleration of the vehicle.

[0207] The actuator constraints are expressed as follows:

[0208] ;

[0209] in, This indicates the longitudinal acceleration command for the (k-1)th step; and These represent the lower and upper limits of acceleration, respectively. This indicates the lateral steering command at step k-1; and These represent the lower and upper limits of the steering angle, respectively. and These represent the lower and upper limits of the change in acceleration, respectively. and These represent the lower and upper limits of the change in steering angle, respectively.

[0210] The stability constraints are:

[0211] ;

[0212] in, This indicates the lateral position offset at step k; Indicates the maximum permissible lateral offset; This represents the heading angle at the k-th step; Indicates the maximum permissible heading angle; This represents the rate of change of the heading angle at the k-th step; This indicates the maximum permissible rate of change of heading angle.

[0213] Constraints are the core mechanism for ensuring safety in MPC. Safety constraints are based on headway and braking distance, ensuring sufficient safe distance at all times. Stability constraints limit acceleration and jerk range to ensure ride comfort. By transforming the MPC problem into a standard quadratic programming form, it can be solved using mature efficient set algorithms. Combined with warm-start techniques, the convergence speed is significantly improved.

[0214] Based on the MPC framework, the control trajectory can be optimized in advance, making the changes in inertial force experienced by passengers smoother and significantly improving the passenger riding experience. Moreover, MPC simultaneously optimizes multiple dimensions of indicators such as following distance, relative speed, and acceleration, achieving synergistic optimization of safety and experience.

[0215] S105. Based on the risk level, adjust the initial control parameters of the vehicle to obtain the target control parameters of the vehicle.

[0216] In one feasible implementation:

[0217] Based on the risk level, the weights corresponding to the initial control parameters of the vehicle are adjusted using a preset adaptive weight adjustment function to obtain the adjusted weight allocation result.

[0218] Based on the adjusted weight allocation results, the multi-objective cost function is solved to obtain the target control parameters of the vehicle.

[0219] The formula for the preset adaptive weight adjustment function is as follows:

[0220] ;

[0221] in, Indicates the current risk level; Represents the adaptive weight vector; This represents the basic weight vector under the low-risk level, corresponding to comfort and energy-saving targets; This represents the target weight vector under a high-risk level, corresponding to the security objective; Indicates the risk threshold; This represents a risk scaling factor used to control the steepness of the Sigmoid curve. The Sigmoid function maps the input to (0, 1) to achieve a smooth weight transition, avoiding ride discomfort caused by abrupt changes in the control strategy.

[0222] For example, at low-risk levels, the goals are comfort and energy efficiency. and At higher risk levels, the focus is on lane-keeping comfort, allowing for greater following distance control; at medium risk levels, the goal is to balance safety and comfort. and The weighting is moderate, balancing following distance and lateral stability; at high-risk levels, the goal is safety. and With the highest weight, The lowest weight ensures timely risk avoidance.

[0223] The following control method provided in this application acquires multi-source data and fuses the multi-source data to obtain a fusion result. The multi-source data includes local perception data, preceding vehicle motion data, and the vehicle's own motion data. Fusing multi-source data helps improve the accuracy of state estimation and provides reliable input for subsequent prediction and control. Based on the fusion result, multimodal trajectory prediction is performed on the preceding vehicle's driving behavior to obtain a preceding vehicle driving behavior prediction result. The multimodal trajectory prediction includes longitudinal motion prediction and lateral motion prediction. By performing multimodal trajectory prediction on the preceding vehicle's driving behavior, proactive prediction of the preceding vehicle's future behavior helps to identify potential following risks in advance, significantly improving following safety. Following risks are quantified using multi-dimensional risk indicators to obtain a risk level. Based on the fusion result, the preceding vehicle driving behavior prediction result, and the control vector, a multi-objective cost function is constructed. The multi-objective cost function is solved using preset constraints to obtain the vehicle's initial control parameters. Based on the risk level, the vehicle's initial control parameters are adjusted to obtain the vehicle's target control parameters. By quantifying the risks of following another vehicle using multi-dimensional risk indicators, the risk level is obtained, and the vehicle's control parameters are adaptively adjusted according to the risk level. This transforms risk identification from a reactive response to a proactive prediction, significantly improving the sensitivity of risk perception. It also provides a time window for determining vehicle control parameters, enabling a fast and stable safety response and ensuring the safety of following another vehicle.

[0224] Based on the vehicle following control method described in the preceding embodiments, this application also provides a vehicle following control device. Figure 2 This is a schematic diagram of the device. Figure 2 As shown, the following control device includes:

[0225] The data acquisition and fusion module 201 is used to acquire multi-source data, fuse the multi-source data, and obtain a fusion result; the multi-source data includes local sensing data, preceding vehicle motion data, and current vehicle motion data.

[0226] The preceding vehicle driving behavior prediction module 202 is used to perform multimodal trajectory prediction on the preceding vehicle driving behavior based on the fusion result, and obtain the preceding vehicle driving behavior prediction result; the multimodal trajectory prediction includes longitudinal motion prediction and lateral motion prediction.

[0227] The risk level determination module 203 is used to quantify the risk of following a vehicle using multi-dimensional risk indicators to obtain the risk level.

[0228] The initial control parameter determination module 204 of the vehicle is used to construct a multi-objective cost function based on the fusion result, the prediction result of the driving behavior of the preceding vehicle and the control vector, and solve the multi-objective cost function using preset constraints to obtain the initial control parameters of the vehicle.

[0229] The vehicle target control parameter determination module 205 is used to adjust the initial control parameters of the vehicle based on the risk level to obtain the vehicle's target control parameters.

[0230] Optionally, the preceding vehicle driving behavior prediction module is used for:

[0231] The longitudinal motion prediction unit is used to predict the longitudinal motion of the vehicle in front based on the fusion result, and obtain the longitudinal motion prediction result of the vehicle in front.

[0232] The lateral motion prediction unit is used to predict the lateral motion of the vehicle in front based on the fusion result, and obtain the lateral motion prediction result of the vehicle in front.

[0233] The processing unit is used to process the lateral motion prediction results and longitudinal motion prediction results of the preceding vehicle using a Gaussian mixture model to obtain the driving behavior prediction results of the preceding vehicle.

[0234] Optionally, the longitudinal motion prediction unit is used for:

[0235] Based on the actual speed of the vehicle in front at the current moment and the acceleration prediction results at a preset time step, the longitudinal motion prediction results of the vehicle in front are obtained;

[0236] The acceleration prediction result with the preset time step is obtained based on the actual acceleration of the preceding vehicle, the rate of change of the preceding vehicle's acceleration, the braking intention, and the traffic conditions ahead; the braking intention is obtained based on the preceding vehicle's acceleration and the preceding vehicle's braking state.

[0237] Optionally, the lateral motion prediction unit is used for:

[0238] Based on multi-dimensional information, a logistic regression model is used to identify the lane-changing intention of the vehicle in front, and the lane-changing intention identification result is obtained; the multi-dimensional information includes turn signal status, lateral offset and lane line crossing time.

[0239] The driving trajectory of the vehicle in front is predicted based on the lateral motion parameters of the vehicle in front at the current moment, and the initial prediction result of the lateral motion of the vehicle in front is obtained. The lateral motion parameters of the vehicle in front include the predicted lateral position of the vehicle in front, the actual lateral position of the vehicle in front, the speed of the vehicle in front, the heading angle of the vehicle in front, and the lateral acceleration.

[0240] Based on the recognition result of the preceding vehicle's lane-changing intention, the lateral acceleration is adjusted to obtain the prediction result of the preceding vehicle's lateral movement target.

[0241] Optionally, the data acquisition and fusion module is used for:

[0242] The multi-source data is processed; the processing includes time alignment, coordinate transformation, and data validity verification.

[0243] A state vector is designed based on processed multi-source data, and a nonlinear state transition function is used to predict the state to obtain the state prediction result; the state vector includes multiple vehicle-following control parameters.

[0244] The weight of each following control parameter in the state vector is quantified by information entropy to obtain the weight allocation result. The state vector and the state estimation result are then fused based on the weight allocation result to obtain the fusion result.

[0245] Optionally, the risk level determination module is used for:

[0246] The urgency of a collision is quantified using a time-to-collision index to obtain the collision time.

[0247] The braking safety distance index is used to quantify reaction time and braking performance, thus obtaining the braking safety distance.

[0248] Multiple risk factors are quantified using wind potential energy field indicators to obtain risk potential energy values;

[0249] The collision time, the braking safety distance, and the risk potential energy value are weighted and summed to obtain the risk value and the risk level corresponding to the risk value.

[0250] Optionally, the vehicle target control parameter determination module is used for:

[0251] Based on the risk level, the weights corresponding to the initial control parameters of the vehicle are adjusted using a preset adaptive weight adjustment function to obtain the adjusted weight allocation result.

[0252] Based on the adjusted weight allocation results, the multi-objective cost function is solved to obtain the target control parameters of the vehicle.

[0253] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the following control method as described in any of the method embodiments.

[0254] Furthermore, this application embodiment also provides a processor for running a computer program, which executes the following control method as described in any of the foregoing method embodiments.

[0255] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. The components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment solution according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0256] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A vehicle following control method, characterized in that, include: Acquire multi-source data, fuse the multi-source data, and obtain a fusion result; the multi-source data includes local sensing data, preceding vehicle motion data, and current vehicle motion data. Based on the fusion results, multimodal trajectory prediction is performed on the driving behavior of the vehicle in front to obtain the driving behavior prediction result of the vehicle in front; the multimodal trajectory prediction includes longitudinal motion prediction and lateral motion prediction. The risk of following another vehicle is quantified using multi-dimensional risk indicators to obtain a risk level. Based on the fusion result, the predicted driving behavior of the preceding vehicle, and the control vector, a multi-objective cost function is constructed. The multi-objective cost function is solved using preset constraints to obtain the initial control parameters of the vehicle. Based on the risk level, the initial control parameters of the vehicle are adjusted to obtain the target control parameters of the vehicle.

2. The method according to claim 1, characterized in that, Based on the fusion result, multimodal trajectory prediction of the preceding vehicle's driving behavior is performed to obtain the preceding vehicle's driving behavior prediction result, including: Based on the fusion results, longitudinal motion prediction is performed on the driving behavior of the vehicle in front, and the longitudinal motion prediction result of the vehicle in front is obtained. Based on the fusion results, the lateral motion prediction of the vehicle in front is performed on the driving behavior of the vehicle in front, and the lateral motion prediction result of the vehicle in front is obtained. The prediction results of the lateral motion and longitudinal motion of the vehicle in front are processed using a Gaussian mixture model to obtain the prediction results of the driving behavior of the vehicle in front.

3. The method according to claim 2, characterized in that, Based on the fusion result, the longitudinal motion prediction of the preceding vehicle's driving behavior is performed to obtain the preceding vehicle's longitudinal motion prediction result, including: Based on the actual speed of the vehicle in front at the current moment and the acceleration prediction results at a preset time step, the longitudinal motion prediction results of the vehicle in front are obtained; The acceleration prediction result with the preset time step is obtained based on the actual acceleration of the preceding vehicle, the rate of change of the preceding vehicle's acceleration, the braking intention, and the traffic conditions ahead; the braking intention is obtained based on the preceding vehicle's acceleration and the preceding vehicle's braking state.

4. The method according to claim 2, characterized in that, Based on the fusion result, the lateral motion prediction of the preceding vehicle's driving behavior is performed to obtain the preceding vehicle's lateral motion prediction result, including: Based on multi-dimensional information, a logistic regression model is used to identify the lane-changing intention of the vehicle in front, and the lane-changing intention identification result is obtained; the multi-dimensional information includes turn signal status, lateral offset and lane line crossing time. The driving trajectory of the vehicle in front is predicted based on the lateral motion parameters of the vehicle in front at the current moment, and the initial prediction result of the lateral motion of the vehicle in front is obtained. The lateral motion parameters of the vehicle in front include the predicted lateral position of the vehicle in front, the actual lateral position of the vehicle in front, the speed of the vehicle in front, the heading angle of the vehicle in front, and the lateral acceleration. Based on the recognition result of the preceding vehicle's lane-changing intention, the lateral acceleration is adjusted to obtain the prediction result of the preceding vehicle's lateral movement target.

5. The method according to claim 1, characterized in that, The process of acquiring multi-source data and fusing the multi-source data to obtain a fusion result includes: The multi-source data is processed; the processing includes time alignment, coordinate transformation, and data validity verification. A state vector is designed based on processed multi-source data, and a nonlinear state transition function is used to predict the state to obtain the state prediction result; the state vector includes multiple vehicle-following control parameters. The weight of each following control parameter in the state vector is quantified by information entropy to obtain the weight allocation result. The state vector and the state estimation result are then fused based on the weight allocation result to obtain the fusion result.

6. The method according to claim 1, characterized in that, The method of quantifying the risk of following another vehicle using multi-dimensional risk indicators to obtain a risk level includes: The urgency of a collision is quantified using a time-to-collision index to obtain the collision time. The braking safety distance index is used to quantify reaction time and braking performance, thus obtaining the braking safety distance. Multiple risk factors are quantified using wind potential energy field indicators to obtain risk potential energy values; The collision time, the braking safety distance, and the risk potential energy value are weighted and summed to obtain the risk value and the risk level corresponding to the risk value.

7. The method according to claim 1, characterized in that, The adjustment of the initial control parameters of the vehicle based on the risk level to obtain the target control parameters of the vehicle includes: Based on the risk level, the weights corresponding to the initial control parameters of the vehicle are adjusted using a preset adaptive weight adjustment function to obtain the adjusted weight allocation result. Based on the adjusted weight allocation results, the multi-objective cost function is solved to obtain the target control parameters of the vehicle.

8. A vehicle following control device, characterized in that, include: The data acquisition and fusion module is used to acquire multi-source data, fuse the multi-source data, and obtain a fusion result; the multi-source data includes local sensing data, preceding vehicle motion data, and current vehicle motion data. The preceding vehicle driving behavior prediction module is used to perform multimodal trajectory prediction on the preceding vehicle driving behavior based on the fusion result, and obtain the preceding vehicle driving behavior prediction result; the multimodal trajectory prediction includes longitudinal motion prediction and lateral motion prediction; The risk level determination module is used to quantify the risk of following a vehicle using multi-dimensional risk indicators to obtain the risk level. The initial control parameter determination module for the vehicle is used to construct a multi-objective cost function based on the fusion result, the prediction result of the driving behavior of the preceding vehicle, and the control vector, and to solve the multi-objective cost function using preset constraints to obtain the initial control parameters of the vehicle. The vehicle target control parameter determination module is used to adjust the initial control parameters of the vehicle based on the risk level to obtain the vehicle's target control parameters.

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

10. A processor, characterized in that, Used to run a computer program, which, when running, executes the vehicle following control method as described in any one of claims 1-7.

Images