CN112896188B - Automatic driving decision control system considering front vehicle encounter - Google Patents

Abstract

The invention discloses an automatic driving decision control system considering a front vehicle encounter, which comprises: establishing a 'man-vehicle-road' dynamic model of the front vehicle according to the type of the front vehicle, and calculating the response of the front vehicle under the non-emergency road condition; comparing the response of the front vehicle under the non-emergency road condition and the actual road condition, and judging whether the driving behavior of the front vehicle is abnormal or not; the method comprises the steps of estimating the operation input and the interference input of a front vehicle according to the actual response of the front vehicle relative to the current road surface based on a front vehicle human-vehicle-road dynamic model, predicting the emergency road condition encountered by the front vehicle, and correcting the decision, the trajectory planning and the control of automatic driving according to the prediction result. The system overcomes the defect that the shielding of the front vehicle limits the effective sensing range of the sensor, can acquire road condition information which is shielded by the front vehicle and cannot be acquired by predicting and estimating the specific details encountered by the front vehicle, expands the sensing capability range of the automatic driving vehicle, corrects the automatic driving decision and control strategy, and improves the driving safety and comfort.

Description

Automatic driving decision control system considering front vehicle encounter

Technical Field

The invention belongs to the technical field of intelligent networked vehicles and automatic driving, and particularly relates to an automatic driving decision control system considering a front vehicle encounter.

Background

The system can sense the front road condition on a driving path in time during driving, and is a basic guarantee for driving safety. However, since the vehicles suddenly brake on the front emergency road conditions, such as pits in urban roads and expressways, the vehicles behind the vehicles are not able to find, estimate and predict the road conditions in time, and therefore, the vehicles behind the vehicles are likely to turn over or rear-end accidents. Under the condition of urban congestion, due to the fact that the following distance is short, the vehicle is like a zebra crossing is shielded by a bus in front of an adjacent lane, and due to the fact that pedestrians on the zebra crossing cannot be observed, serious accidents of people and vehicles are likely to happen. Therefore, the driving behavior and the motion state of the front vehicle are directly reflected by the front vehicle driver facing the front road condition, and the front vehicle encounters the situation that the front vehicle is the 'reference of the front vehicle' of the vehicle, so that the reference significance is provided for the decision control of the vehicle.

The improvement of safety and comfort is a typical target of Advanced Driving Assistance Systems (ADAS) such as Adaptive Cruise Control (ACC), Forward Collision Warning System (FCW), automatic Emergency Collision Avoidance (AEB), and the like, and is also a core target in Automatic Driving Systems (ADS). The design scheme of the disclosed ADAS and ADS system comprises single-vehicle intelligence and networking intelligence, wherein the single-vehicle intelligence scheme senses surrounding road conditions by a vehicle sensor, and realizes safe driving by high-speed calculation and active control of steering, braking, driving and the like. The ADAS and ADS system of the intelligent network connection scheme realizes the information distribution and sharing of roads and vehicles through the vehicle network connection technology, finds obstacles in road conditions based on the information distribution and sharing, and plans and implements the movement of the vehicles. Taking an adaptive cruise ACC system as an example, the system senses a vehicle ahead through sensors such as radar or a camera. When a vehicle is in front, according to the cruising speed and the head time distance set by the driver, the brake, the accelerator and the front vehicle are controlled to stably run at the cruising speed as close as possible to the set cruising speed on the premise of keeping the safe vehicle distance so as to reduce the driving burden of the driver and reduce the driving error.

However, a front vehicle, particularly a large vehicle, blocks the view of a rear vehicle and greatly limits the effective sensing range of a rear vehicle sensor. When the front vehicle encounters an emergency, such as a road pit and an emergency rear-end collision, the rear vehicle often cannot make timely and accurate judgment due to insufficient information acquired through the sight line and the vehicle-mounted sensor. The perception system of the disclosed ADAS and ADS only focuses on the direct perception of the road conditions around the vehicle, especially senses and predicts the obstacles nearest to the vehicle, and obtains the prediction of the current motion state and the future motion state, but there is no report of predicting the road conditions encountered by the front vehicle based on a model. As an example, in the ACC system, the vehicle only focuses on the motion state of the preceding vehicle, and makes an acceleration decision control command such as braking, acceleration, and uniform speed, but a technology of "predicting and estimating specific details of the road condition encountered by the preceding vehicle according to the speed of the preceding vehicle, and correcting the acceleration decision control command according to the estimation result" has not been reported in public.

In summary, there is no report related to predicting the preceding vehicle encounter according to the driving behavior and motion state of the preceding vehicle, and performing decision control on the vehicle based on the prediction.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provide a system for automatic driving decision control considering the encounter of a front vehicle. The invention considers the driving behavior and the motion state of the front vehicle in an emergency and refers to the decision control of the vehicle.

The purpose of the invention is realized by the following technical scheme: a system for automatic driving decision control considering a preceding vehicle encounter comprises a sensing module, a preceding vehicle road dynamics modeling module, a preceding vehicle response calculation module under actual road conditions, a non-emergency road condition preceding vehicle response calculation module, a preceding vehicle abnormal driving behavior judgment module, a preceding vehicle operation input and interference input estimation module, a preceding vehicle encounter prediction module and an automatic driving decision and control module.

The sensing module acquires traffic road condition information, vehicle state data and front vehicle state data through a sensor. The traffic road condition information comprises obstacles, road surface adhesion coefficients, lane widths, lane slopes, lane curvatures, road markings, traffic signs, signal lamp states, traffic flow information and weather conditions; the vehicle state data comprises a vehicle body six-degree-of-freedom dynamic state Y _ ego of the vehicle relative to the road surface and vehicle manipulation input; the front vehicle state data comprises a six-degree-of-freedom dynamic state Y _ relative of the front vehicle relative to the vehicle body, vehicle appearance parameters, tail lamp states, steering indicator lamp states, warning horn sounds, tire noises, power transmission system noises and impact noises. The vehicle body six-degree-of-freedom dynamic state comprises linear velocity, angular velocity, linear acceleration, angular acceleration, position and attitude angle of six motion degrees of freedom of a vehicle body rigid body; the operation input comprises the operations of steering, braking, driving, indicating lamps and warning horns.

The leading vehicle and road dynamics modeling module comprises a vehicle type identification submodule and a human and vehicle road modeling submodule. The vehicle type recognition sub-module determines the type of the front vehicle by combining the vehicle type database based on the vehicle appearance parameters in the state data of the front vehicle output by the sensing module. The man-vehicle-road modeling submodule establishes a front vehicle 'man-vehicle-road' dynamic model:

X(k+1)＝F(X(k),Uk,Wk)

Y(k)＝H(X(k),Uk,Wk)

wherein X (k) is the state data of the front vehicle at the time k; uk is the front vehicle operation input within the time of 0-k; wk is the interference input within the time range of 0-k; y (k) is the output response at time k; f is the state transfer function and H is the output response function.

And the actual road condition front vehicle response calculation module superposes Y _ ego and Y _ relative based on the vehicle state data and the front vehicle state data output by the sensing module to obtain a vehicle body six-degree-of-freedom dynamic state Yreal of the front vehicle relative to the road surface under the actual road condition.

The non-emergency road condition front vehicle response calculation module calculates a six-degree-of-freedom dynamic state Ynormal of a front vehicle relative to a vehicle body of a road surface when the front vehicle is in a non-emergency road condition on the basis of traffic road condition information acquired by the sensing module and a front vehicle human-vehicle-road dynamic model in a front vehicle human-vehicle-road dynamic modeling module. The non-emergency road condition refers to a condition that any obstacle outside the perception capability range of the vehicle is not encountered during the driving of the vehicle in front on a dry and flat road surface.

The front vehicle abnormal driving behavior judging module performs difference on dimensions of YReal and Ynormal to obtain multi-dimensional motion difference, and judges whether the front vehicle driving behavior is abnormal or not according to the motion difference.

And the front vehicle operation input and interference input estimation module estimates the front vehicle operation input Uk and the front vehicle interference input Wk according to YReal based on a front vehicle human-vehicle-road dynamic model in the front vehicle road dynamic modeling module and the traffic road condition information obtained by the sensing module after the front vehicle abnormal driving behavior judgment module judges that the front vehicle driving behavior is abnormal.

When the preceding vehicle encounter prediction module judges that the driving behavior of the preceding vehicle is abnormal, the obstacle attribute theta of the preceding vehicle encounter is predicted based on the traffic road condition information and the preceding vehicle state data output by the sensing module, the preceding vehicle operation input and the preceding vehicle interference input estimated by the interference input estimation module, wherein the obstacle attribute theta comprises the type, the position and the collision force with the preceding vehicle of an obstacle, the depth or the height, the length and the width of a static obstacle, the road surface unevenness and the speed and the track of a dynamic obstacle.

The automatic driving decision and control module is specifically that firstly, according to the traffic road condition information Info _ traffic acquired by the sensing module, an automatic driving algorithm gives an initial vehicle decision control strategy Policy _ init ═ Vx not considering the encounter of a front vehicle; vy; yawrate; FuturePath; SusMode ] including a longitudinal speed Vx, a lateral speed Vy, a yaw rate Yawrate, a future travel trajectory FuturePath, and a suspension control mode SusMode of the host vehicle for a future period of time; the suspension control modes comprise a comfort mode, a conventional mode and a controllability mode, and the corresponding assignments are sequentially-1, 0 and 1. Then, according to the attribute theta of the obstacle encountered by the front vehicle predicted by the front vehicle encounter prediction module and the traffic road condition information output by the sensing module, calculating the corrected delta Policy of the decision control strategy of the vehicle by using a correction strategy, wherein the corrected delta Policy is [ delta Vx; Δ Vy; Δ Yawrate; Δ FuturePath; Δ SusMode ]. Finally, Policy _ init is corrected based on Δ Policy to implement the desired motion of the vehicle.

Further, the judging method in the module for judging the abnormal driving behavior of the preceding vehicle is a method based on a set threshold, a classification method based on machine learning or a regression method based on machine learning.

The method based on the set threshold specifically comprises the steps that when the motion difference of a certain dimensionality is larger than the corresponding set threshold, the dynamic state of the corresponding dimensionality of the front vehicle is abnormal, and as long as the dynamic state of the front vehicle of one dimensionality is abnormal, the driving behavior of the front vehicle is abnormal; otherwise, the dynamic state of the front vehicle is not abnormal, and the decision control of the vehicle is not required to be corrected.

The classification method based on machine learning specifically comprises the steps of inputting motion differences into a machine learning classification model, and classifying the driving behaviors of the front vehicle, wherein the classification result comprises abnormity and normality.

The regression method based on machine learning specifically comprises the steps of taking motion difference as input of a machine learning regression model, outputting abnormal degree vectors lambda, wherein lambda corresponds to each dimension of a six-degree-of-freedom dynamic state of a vehicle body one by one, and elements of the vectors lambda are all in [0,1 ]; when the degree of abnormality of a certain dimension is greater than a correspondingly set threshold value, the dynamic state of the corresponding dimension is abnormal, and if the dynamic state of a front vehicle of one dimension is abnormal, the driving behavior of the front vehicle is abnormal; otherwise, the dynamic state of the front vehicle is not abnormal, and the decision control of the vehicle is not required to be corrected;

the training of the machine learning regression model is specifically to simulate the motion difference between a front vehicle and a non-emergency road condition when the front vehicle encounters different emergency road conditions through a vehicle dynamics model, and train the machine learning regression model based on abnormal degree vectors corresponding to different motion differences; defining each dimensionality of an abnormality degree vector lambda when an accident occurs as 1, and defining each dimensionality of the abnormality degree lambda when the automobile is normally driven as 0; counting the six-degree-of-freedom dynamic state change of a vehicle body caused by corresponding deceleration and steering operations when a front vehicle encounters an emergency road condition and is successfully avoided, obtaining the maximum variable quantity of each dimension of the dynamic state, and forming a vector delta ysimmax; defining each dimension lambda of the abnormal degree vector of the dynamic state_iComprises the following steps:

the system comprises a simulation system, a road surface simulation system and a road surface simulation system, wherein Ysimreal is a six-degree-of-freedom dynamic state of a vehicle body of a front vehicle relative to the road surface in the simulation system under actual road conditions; ysimnormal is a six-degree-of-freedom dynamic state of a vehicle body of a front vehicle relative to a road surface under a non-emergency road condition in simulation;

is a constant greater than zero to ensure λ_iBetween 0 and 1; subscript i denotes the ith dimension in the vector; the emergency road condition refers to the situation that a preceding vehicle encounters an obstacle outside the perception capability range of the vehicle during running.

Further, the estimation strategy in the front vehicle manipulation input and interference input estimation module is longitudinal-transverse-vertical direction decoupling estimation or longitudinal-transverse-vertical direction joint estimation.

The longitudinal-transverse-vertical joint estimation adopts a six-freedom-degree vehicle dynamic model to estimate the control input of braking, driving and steering of a front vehicle and the interference input.

The longitudinal-transverse-vertical decoupling estimation specifically comprises the following steps: according to longitudinal dynamics states in YReal and longitudinal lane gradients in traffic road condition information, a vehicle longitudinal dynamics model is adopted to estimate a brake control input Com _ brk, a drive control input Com _ drv and a longitudinal component Fxe of interference input of a front vehicle; according to the transverse dynamic state in YReal and the transverse lane gradient in the traffic road condition information, a vehicle steering control dynamic model is adopted to estimate a front vehicle steering control input Com _ str, a lateral component force Fye of an interference input and a yaw component moment Mze of the interference input; removing six-degree-of-freedom motion response Yhorizontal of a vehicle body caused by Com _ brk, Com _ drv, Com _ str, Fxe, Fye and Mze from Yreal to obtain a difference value Yreal-Yhorizontal, and estimating a vertical component force and a component force of interference input, including a vertical component force Fze, a roll component force Mxe and a pitch component force Mye of the interference input, as input of a vehicle vertical vibration model; the final estimated preceding vehicle maneuver inputs include a brake maneuver input Com _ brk, a drive maneuver input Com _ drv, and a steering maneuver input Com _ str; the estimated front vehicle disturbance inputs include the longitudinal component Fxe, the lateral component Fye, the yaw component moment Mze, the vertical component Fze, the roll component moment Mxe, and the pitch component moment Mye of the disturbance input.

Further, the prediction strategy in the preceding vehicle encounter prediction module is a rule-based prediction strategy or a deep neural network learning-based prediction strategy.

The rule-based prediction strategy comprises rule base query and fuzzy rule reasoning, and the obstacle attribute theta encountered by the preceding vehicle in the past time period is comprehensively judged according to the estimated values of the operation input and the interference input of the preceding vehicle in the past time period, tire noise and impact noise.

The prediction strategy based on the deep neural network learning specifically comprises the steps that traffic road condition information in the past time period, an estimated value of the previous vehicle operation input and an estimated value of the previous vehicle interference input are used as the input of the deep neural network, and the prediction result of the deep neural network is the obstacle attribute theta encountered by the previous vehicle in the past time period.

Further, the correction strategy in the automated driving decision and control module is a rule-based correction strategy or a machine learning-based correction strategy.

The rule-based correction strategy specifically comprises: the type and position of the obstacle in the obstacle attribute theta encountered by the front vehicle and the depth, height, length, width and road surface unevenness of the static obstacle are input into a machine learning model ML _ Comfort for judging the influence of Comfort, and whether the influence on the future travelling Comfort of the vehicle is output. All dimensions in the obstacle attribute theta encountered by the front vehicle are input into the machine learning model ML _ Safe for judging the influence of safety, and whether the influence of safety on future driving of the vehicle is output. As long as one of the Comfort judgment model ML _ Comfort and the safety judgment model ML _ Safe has an influence, the decision control strategy of the vehicle needs to be corrected, otherwise, the decision control strategy of the vehicle does not need to be corrected; according to the attribute of the obstacle encountered by the front vehicle and the traffic road condition information, solving the correction quantity of the vehicle decision control strategy:

ΔPolicy＝f_correct(θ,Info_traffic)

wherein f _ correct is a correction calculation function.

The correction strategy based on machine learning specifically comprises the steps of inputting the attribute theta of the obstacle encountered by the front vehicle in the past time period and the traffic road condition information output by the sensing module into a machine learning model, and outputting the information as the correction quantity delta Policy of the decision and control of the vehicle.

Further, in the automatic driving decision and control module, the construction rule of f _ correct in the machine learning-based correction strategy is as follows: under the given traffic road condition information Info _ traffic, when the output of the Comfort judgment model ML _ Comfort is significant, the larger the road surface unevenness, the closer the position of the obstacle, and the larger the size of the static obstacle in the obstacle attribute θ, the larger and negative the values of the vehicle longitudinal vehicle speed correction amount Δ Vx and the suspension mode correction amount Δ susMode are. When the output of the safety calculation model ML _ Safe has an influence, a future travel trajectory correction amount Δ FuturePath is determined to avoid an imminent collision based on the position of the obstacle in the obstacle attribute θ, the depth or height of the static obstacle, the length or width of the static obstacle, or the speed or trajectory of the dynamic obstacle, and the correction amounts Δ Vx, Δ Vy, Δ Yawrate, and Δ SusMode are determined by an automatic driving control algorithm based on the corrected future travel trajectory FuturePath + Δ FuturePath.

Further, the types of the obstacles include traffic control signals, static obstacles and dynamic obstacles.

Further, the disturbance inputs include road surface excitation inputs, aerodynamic forces, and obstacle impact force excitations.

Further, the forms of the functions F and H comprise a logic rule model, a transfer function model, a linear state space model, a nonlinear state space model and a neural network model; the linear state space model comprises a vehicle six-degree-of-freedom dynamic model, a vehicle longitudinal dynamic model, a vehicle steering control dynamic model, a vehicle single-degree-of-freedom vertical vibration model, a vehicle two-degree-of-freedom vertical vibration model and a vehicle vertical-pitching-rolling vibration model.

Further, the travel track includes the change of the positions of each point of the vehicle body with time and the change of the posture of the vehicle body with time.

Compared with the prior art, the invention has the beneficial effects that:

1. the influence of driving behaviors and motion states of the front vehicle on the decision control of the vehicle is considered when the front vehicle encounters an emergency road condition, and the decision, trajectory planning and control of automatic driving are corrected by predicting the attribute of the obstacle encountered by the front vehicle and the emergency road condition of a front road section, so that the riding comfort of passengers of the vehicle is improved;

2. on the basis of a 'human-vehicle-road' dynamic model of the front vehicle, the control input and the interference input of the front vehicle in case of an emergency road condition are estimated and used for predicting the attribute of the barrier outside the sensing capability range of the front vehicle and the emergency road condition, so that the information acquisition and predictability of automatic driving decision control are increased, the effectiveness of the decision control is improved, and the driving safety can be improved; particularly, under the condition of night or haze weather, the sensing range of the sensor is limited, and the capability boundary of automatic driving sensing and prediction can be remarkably expanded;

3. the invention can be applied to manually driven vehicles, semi-automatically driven vehicles and fully-automatically driven vehicles, and improves the safety, the comfort, the energy conservation and the user experience of the vehicles. The invention considers the output response of the front vehicle under the non-emergency road condition, compares the output response with the response of the front vehicle under the actual road condition, judges whether the driving behavior of the front vehicle is abnormal or not, and can remind passengers of the possible emergency road condition in front in advance.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention;

FIG. 1 is a schematic diagram of the system components and process of the present invention;

FIG. 2 is a schematic diagram of a vehicle encountering a speed bump before going on an uphill turn;

FIG. 3 is a schematic side view of a vehicle in relation to a preceding vehicle when the preceding vehicle encounters a deceleration strip on a horizontal road segment;

FIG. 4 is a schematic diagram of a vehicle body side-tipping when a front vehicle encounters a single-side depression of a road (from the view of a front camera of the vehicle);

fig. 5 is a schematic diagram of the side inclination of the vehicle body when the front vehicle encounters a single-side bump on the road (from the view angle of the front camera of the vehicle).

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the following embodiments and accompanying drawings. The exemplary embodiments and descriptions of the present invention are provided to explain the present invention, but not to limit the present invention.

As shown in fig. 1, the invention provides a system for automatic driving decision control considering a preceding vehicle encounter, which comprises a sensing module, a preceding vehicle road dynamics modeling module, a preceding vehicle response calculation module under actual road conditions, a preceding vehicle response calculation module under non-emergency road conditions, a preceding vehicle abnormal driving behavior judgment module, a preceding vehicle operation input and interference input estimation module, a preceding vehicle encounter prediction module and an automatic driving decision and control module.

(1) The sensing module acquires traffic road condition information, vehicle state data and front vehicle state data through a sensor; the data information sources comprise vehicle-mounted sensors, high-precision maps, road traffic facilities and vehicle networking information. The traffic road condition information comprises obstacles, road surface adhesion coefficients, lane widths, lane slopes, lane curvatures, road markings, traffic signs, signal lamp states, traffic flow information and weather conditions; the vehicle state data comprises a vehicle body six-degree-of-freedom dynamic state and vehicle operation input; the front vehicle state data comprises a six-degree-of-freedom dynamic state of a front vehicle body, vehicle appearance parameters, a tail lamp state, a steering indicator lamp state, warning horn sound, tire noise, power transmission system noise and impact noise; the vehicle body six-degree-of-freedom dynamic state comprises linear velocity, angular velocity, linear acceleration, angular acceleration, position and attitude angle of six motion degrees of freedom of a vehicle body rigid body; the vehicle body six-degree-of-freedom dynamic state Y _ ego is generated by the traction motion of the vehicle relative to the current road surface, and the front vehicle body six-degree-of-freedom dynamic state Y _ relative is the six-degree-of-freedom dynamic state data of the front vehicle relative to the vehicle and is generated by the relative motion of the front vehicle and the vehicle; the operation input comprises the operations of steering, braking, driving, indicating lamps and warning horns.

The obstacles comprise driving instructions for strict violation of traffic regulations and the fact that the vehicle cannot drive an approaching object under safety constraints; specific characteristics of the obstacles include traffic control signals, static obstacles and dynamic obstacles. The static barriers comprise road surface depressions, road surface bulges, deceleration strips, tracks, inspection well covers, road surface foreign bodies and road boundaries; dynamic obstacles include pedestrians, non-motorized vehicles, motor vehicles, animals.

(2) The front vehicle and vehicle road dynamics modeling module comprises a vehicle type recognition sub-module and a human and vehicle road modeling sub-module, wherein the vehicle type recognition sub-module determines the type of the front vehicle based on vehicle appearance parameters in the front vehicle state data output by the sensing module and the vehicle type database; the man-vehicle-road modeling sub-module establishes a front vehicle 'man-vehicle-road' dynamic model; the vehicle appearance parameters comprise appearance size and appearance pictures; the vehicle types include saloon cars, SUVs, pickup trucks, medium size trucks, heavy duty trucks.

The front vehicle human-vehicle-road dynamic model is as follows:

X(k+1)＝F(X(k),Uk,Wk)

Y(k)＝H(X(k),Uk,Wk)

wherein, X (k) is the previous vehicle state data at the time k, and X (k +1) is the previous vehicle state data at the time k + 1; uk is the front vehicle operation input within the time of 0-k; wk is the interference input within 0-k time, including road surface excitation input, aerodynamic acting force and obstacle collision force excitation; y (k) is the output response of the human-vehicle-road system at the moment k; f is a state transfer function, H is an output response function, and the forms of the F and H model functions comprise a logic rule model, a transfer function model, a linear state space model, a nonlinear state space model and a neural network model. The linear state space model comprises a vehicle six-degree-of-freedom dynamic model, a vehicle longitudinal dynamic model, a vehicle steering control dynamic model, a vehicle single-degree-of-freedom vertical vibration model, a vehicle two-degree-of-freedom vertical vibration model and a vehicle vertical-pitching-rolling vibration model.

(3) The actual road condition front vehicle response calculation module regards the vehicle and the front vehicle as rigid bodies based on the vehicle state data and the front vehicle state data output by the sensing module, the vehicle body six-degree-of-freedom dynamic state Y _ ego in the vehicle state data and the front vehicle body six-degree-of-freedom dynamic state Y _ relative in the front vehicle state data are superposed with the traction motion state Y _ ego and the relative motion state Y _ relative according to the motion synthesis principle of the rigid bodies, and the actual road condition response Yreal of the front vehicle generated by the absolute motion of the front vehicle relative to the current road surface under the actual road condition is obtained and comprises the front vehicle body six-degree-of-freedom dynamic state; where Yreal is a multi-dimensional vector that varies over time.

(4) The non-emergency road condition front vehicle response calculation module is responsible for calculating the output response of the front vehicle under the non-emergency road condition, and the response Ynormal of the front vehicle under the non-emergency road condition is calculated by using a front vehicle human-vehicle-road dynamic model in the front vehicle road dynamic modeling module by considering the traffic road condition information acquired by the sensing module, wherein the response Ynormal comprises the six-degree-of-freedom dynamic state of the front vehicle body; where Ynormal is a multi-dimensional vector that varies over time.

The non-emergency road condition refers to that the front vehicle does not encounter any obstacle outside the perception capability range of the vehicle during running on a dry and flat road surface.

(5) And the front vehicle abnormal driving behavior judging module is used for comparing the motion difference of the front vehicle response Yreal on the actual road condition with the front vehicle response Ynormal on the non-emergency road condition and judging whether the front vehicle driving behavior is abnormal or not.

The motion difference of the six-degree-of-freedom dynamic state of the vehicle body under the actual road condition and the non-emergency road condition within a period of time is expressed in a multi-dimensional form and comprises vehicle running track difference, linear velocity difference, angular velocity difference, linear acceleration difference and angular acceleration difference. The driving track comprises the change of the positions of all points of the vehicle body along with time and the change of the posture of the vehicle body along with time.

The judgment method is based on a threshold setting method, a machine learning based classification method or a machine learning based regression method.

The method based on the set threshold specifically includes that motion difference is compared with the set threshold, and when the motion difference of a certain dimension is larger than the difference threshold correspondingly set for the dimension, the dynamic state of the corresponding dimension of the front vehicle is indicated to be abnormal. If the dynamic state of the front vehicle in one dimension is abnormal, the driving behavior of the front vehicle is abnormal; if all the dimensions of the dynamic state of the vehicle in the prior art are normal, the driving behavior is not abnormal, and the decision control of the vehicle is not required to be corrected.

The classification method based on the machine learning specifically comprises the step of classifying the driving behaviors of the front vehicle by inputting the motion difference into a machine learning model, wherein the classification result comprises abnormity and normality.

The regression method based on machine learning specifically comprises the steps of taking the motion difference as the input of a machine learning model, outputting an abnormal degree vector lambda in the range of 0 to 1, wherein the dimension of the abnormal degree vector lambda is the same as the dimension of the six-degree-of-freedom dynamic state of the vehicle body. λ corresponding to the motion difference of a certain dimension_iIs less thanWhen the dynamic state of the dimension is equal to a set threshold value, the dynamic state of the dimension is normal, and when the dynamic state of the dimension is larger than the set threshold value, the dynamic state of the dimension is abnormal; if the dynamic state of the front vehicle in one dimension is abnormal, the driving behavior of the front vehicle is abnormal; if all the dimensions of the dynamic state of the vehicle in the prior art are normal, the driving behavior is not abnormal, and the decision control of the vehicle is not required to be corrected.

Acquiring data of a front vehicle under different emergency road conditions through vehicle dynamics model simulation, and acquiring abnormal degree training machine learning regression models corresponding to different motion differences; the emergency road condition refers to the situation that a preceding vehicle encounters an obstacle outside the perception capability range of the vehicle during running.

Defining each dimension of the abnormality degree lambda as a maximum value 1 when an accident occurs, and defining each dimension of the abnormality degree lambda as a minimum value 0 when normal driving occurs; and counting the six-degree-of-freedom dynamic state change of the vehicle body caused by corresponding deceleration and steering operations when the front vehicle encounters an emergency road condition and the avoidance is successful, and obtaining the maximum variable quantity of each dimension of the dynamic state to form a vector delta Ysimmax.

Defining each dimension lambda of the abnormal degree of the dynamic state_iComprises the following steps:

wherein Ysimreal is the response of the front vehicle of the actual road condition in the simulation; ysimnormal is the response of the front vehicle of the non-emergency road condition in the simulation;

is a constant greater than zero to ensure λ_iBetween 0 and 1; the index i denotes the ith dimension in the vector.

(6) The front vehicle operation input and interference input estimation module estimates a front vehicle operation input Uk and a front vehicle interference input Wk according to the actual response Yreal of a front vehicle relative to the current road surface based on a front vehicle human-vehicle-road dynamic model established in a front vehicle road dynamic modeling module and traffic road condition information obtained by a sensing module after the front vehicle abnormal driving behavior judgment module judges that the front vehicle driving behavior is abnormal; the estimation strategy comprises decoupling estimation of the vertical-horizontal-vertical direction and joint estimation of the vertical-horizontal-vertical direction.

The joint estimation of longitudinal-transverse-vertical directions adopts a six-degree-of-freedom vehicle dynamic model to estimate the control input of braking, driving and steering of a front vehicle and the interference input.

The longitudinal-transverse-vertical decoupling estimation method specifically comprises the following steps:

according to the longitudinal dynamics state in the actual road condition response YReal of the front vehicle and the longitudinal lane gradient in the traffic road condition information, a vehicle longitudinal dynamics model is adopted to estimate the brake control input Com _ brk, the drive control input Com _ drv and the longitudinal component Fxe of the interference input of the front vehicle.

According to the transverse dynamic state in the actual road condition response YReal of the front vehicle and the transverse lane gradient in the traffic road condition information, a vehicle steering control dynamic model is adopted to estimate a steering control input Com _ str of the front vehicle, a lateral component force Fye of an interference input and a yaw component moment Mze of the interference input.

The six-degree-of-freedom motion response Yhorizontal of the vehicle body caused by Com _ brk, Com _ drv, Com _ str, Fxe, Fye and Mze is removed from the actual road condition front vehicle response Yreal, a difference value Yreal-Yhorizontal is obtained, and the vertical component force and the component moment of the disturbance input, including the vertical component force Fze, the roll component moment Mxe and the pitch component moment Mye of the disturbance input, are estimated and obtained as the input of the vehicle vertical vibration model.

The final estimated previous vehicle control input comprises a braking control input Com _ brk, a driving control input Com _ drv and a steering control input Com _ str; the estimated front vehicle disturbance inputs include the longitudinal component Fxe, the lateral component Fye, the yaw component Mze, the vertical component Fze, the roll component Mxe, and the pitch component Mye of the disturbance input.

(7) And the front vehicle encounter prediction module predicts the emergency road condition encountered by the front vehicle based on the traffic road condition information and the state data of the front vehicle output by the sensing module and the front vehicle operation input and the front vehicle interference input estimated by the interference input estimation module. The prediction strategy comprises a rule-based prediction strategy and a deep neural network learning-based prediction strategy.

The rule-based prediction strategy comprises rule base query and fuzzy rule reasoning, and the obstacle attribute theta encountered by the front vehicle in the past time period tau is comprehensively judged according to the estimated values of the control input and the interference input of the front vehicle in the past time period tau, tire noise and impact noise. The barrier attribute theta is multidimensional information and comprises the type and position of the barrier, the collision force with a front vehicle, the depth or height, length and width of a static barrier, the road surface unevenness and the speed and track of a dynamic barrier.

The prediction strategy based on the deep neural network learning specifically comprises the steps of taking traffic road condition information in a past time period tau, an estimated value of a front vehicle operation input U and an estimated value of a front vehicle interference input W as input of the deep neural network, predicting a road condition encountered by a front vehicle, and obtaining a prediction result which is an obstacle attribute theta in the past time period tau.

(8) An automatic driving decision and control module for controlling the automatic driving,

firstly, according to traffic road condition information Info _ traffic acquired by a sensing module, an automatic driving algorithm gives an initial vehicle decision control strategy Policy _ init which does not consider the encounter of a front vehicle, wherein the initial vehicle decision control strategy Policy _ init comprises the longitudinal speed Vx, the lateral speed Vy, the yaw angular speed Yawrate, a future driving path FuturePath and a suspension control mode SusMode of the vehicle within a period of time t _ pre in the future, namely the Policy _ init is [ Vx; vy; yawrate; FuturePath; SusMode ], wherein the suspension control modes include comfort mode (assigned "-1"), normal mode (assigned "0"), and maneuverability mode (assigned "+ 1").

Then, according to the attribute theta of the obstacle encountered by the front vehicle predicted by the front vehicle encounter prediction module and the traffic road condition information output by the sensing module, calculating the corrected delta Policy of the decision control strategy of the vehicle by using a correction strategy, wherein the corrected delta Policy is [ delta Vx; Δ Vy; Δ Yawrate; Δ FuturePath; Δ SusMode ].

Finally, according to Policy _ init and Δ Policy, the expected movement of the host vehicle is implemented, including the future travel track, linear speed, linear acceleration, angular speed, angular acceleration, and suspension control mode of the host vehicle.

The correction strategies comprise a rule-based correction strategy and a machine learning-based correction strategy.

The rule-based correction strategy outputs whether the obstacle influences the safety and the comfort of the future driving of the vehicle through a machine learning model according to the obstacle attribute theta encountered by the front vehicle, and determines whether to correct the vehicle decision control strategy according to the result, specifically:

the type and position of the obstacle in the obstacle attribute theta encountered by the front vehicle and the depth, height, length, width and road surface unevenness of the static obstacle are input into a machine learning model ML _ Comfort for judging the influence of Comfort, and whether the influence on the future travelling Comfort of the vehicle is output.

All dimensions in the obstacle attribute theta encountered by the front vehicle are input into the machine learning model ML _ Safe for judging the influence of safety, and whether the influence of safety on future driving of the vehicle is output.

If one of the Comfort judgment model ML _ Comfort and the safety judgment model ML _ Safe is influenced, the correction quantity of the decision control strategy of the vehicle is solved according to the attribute of the obstacle encountered by the vehicle ahead and the traffic road condition information:

ΔPolicy＝f_correct(θ,Info_traffic)

wherein f _ correct is a correction calculation function, and the construction rule is as follows: under the given traffic road condition information Info _ traffic, when the output of the Comfort judgment model ML _ Comfort is significant, the obstacle attribute θ indicates that the larger the road surface unevenness, the closer the obstacle position, and the larger the static obstacle size, the larger and negative the values of the vehicle longitudinal vehicle speed correction amount Δ Vx and the suspension mode correction amount Δ susMode, where the suspension mode correction indicates the switching of the corresponding suspension mode. When the output of the safety calculation model ML _ Safe has an influence, it is necessary to determine a future travel trajectory correction amount Δ future path to avoid an imminent collision, based on the position of the obstacle in the obstacle attribute θ, the depth, height, length, width of the static obstacle, or the speed and trajectory of the dynamic obstacle, and the correction amounts Δ Vx, Δ Vy, Δ Yawrate, and Δ SusMode are determined by an automatic driving control algorithm based on the corrected future travel trajectory (future path + Δ future path).

The correction strategy based on machine learning specifically comprises the steps of inputting the attribute theta of the obstacle encountered by the front vehicle in the past time period tau and the traffic road condition information output by the sensing module into a machine learning model, and outputting the information as the correction quantity delta Policy of the decision and control of the vehicle.

Example 1

The scenario 1 is applied by taking the deceleration strip encountered by the vehicle in front of the uphill turning road section as an embodiment, as shown in fig. 2. According to the traffic condition information provided by the perception module, the past time period tau is covered, the step length is recorded as ts, and the time tau is recorded as k ts, namely all perception data in the current time T0 and k sampling step lengths can be used for the system to implement.

The information of the road gradient, the road curvature, the lane boundary line, the lane width, and the like in the road condition ahead in the past time period τ is known.

According to the profile parameters of the front vehicle provided by the perception module, the type of the front vehicle is determined by searching a vehicle type database, and corresponding front vehicle human-vehicle-road dynamic models X (k +1) ═ F (X (k), Uk, Wk) and Y (k) ═ H (X (k), Uk, Wk) are established, wherein the forms of F and H model functions comprise a logic rule model, a transfer function model, a linear state space model, a nonlinear state space model and a neural network model.

In this example, the vehicle is a sedan of a certain size, and the estimated mass is 1600kg and the yaw moment of inertia is 2300kg · m²And the air resistance coefficient Cd is 0.34, and other parameters required by the corresponding model are extracted according to the database.

According to the traffic road condition information provided by the sensing module and the 'man-vehicle-road' dynamic model of the front vehicle, the non-emergency road condition front vehicle response calculation module can calculate the output response of the front vehicle when the front vehicle has no emergency road condition within the sensing capability range of the front vehicle: combining a 'human-vehicle-road' dynamic model of the front vehicle, and calculating the pitch angle of the front vehicle when encountering non-emergency road conditions according to the road gradient; calculating a course angle of a front vehicle when encountering non-emergency road conditions according to the curvature of the road; and calculating the running track of the front vehicle when encountering the non-emergency road condition according to the lane boundary line and the lane width information.

The actual road condition front vehicle response module regards the vehicle and the front vehicle as rigid bodies on the basis of the acquired six-degree-of-freedom dynamic state data Y _ ego of the vehicle body and the six-degree-of-freedom dynamic state data Y _ relative of the front vehicle body relative to the vehicle, generates a vehicle state Y _ ego by the motion of the vehicle relative to the current road surface, generates a state Y _ relative of the front vehicle relative to the vehicle by the relative motion of the front vehicle and the vehicle, and synthesizes the motion state Y _ ego and the relative motion state Y _ relative into an actual road condition front vehicle response Yreal generated by the absolute motion of the front vehicle relative to the current road surface under the actual road condition through the motion synthesis principle of the rigid bodies.

In this embodiment, when the front vehicle encounters the deceleration strip in the past time period τ, and the front wheels and the rear wheels pass through the deceleration strip, the motion response of the front vehicle under the actual road condition is represented by two significant pitch angle changes of the vehicle body, a reduction in longitudinal linear velocity, and two changes in vertical height of the vehicle head and the vehicle tail; the motion response of the front vehicle under the non-emergency road condition is calculated on the basis of the condition of a dry and flat road surface, and the pitch angle, the longitudinal linear velocity and the blowing height of the front vehicle and the rear vehicle are not obviously changed; therefore, the motion difference of the front vehicle under the actual road condition and the non-emergency road condition can be specifically characterized as a remarkable pitch angle difference value, a remarkable longitudinal linear velocity difference value and a remarkable vehicle body posture difference value; the front vehicle abnormal driving behavior judging module adopts a method based on a set threshold value, a pitch angle difference value, a longitudinal linear velocity difference value and a vehicle body posture difference value in the motion difference of the front vehicle are all larger than the set threshold value, the dynamic state of the front vehicle in three dimensions is abnormal, and the set rule judges that the front vehicle driving behavior is abnormal according to the severity and the number of the dynamic state abnormality.

The front vehicle control input and interference input estimation module adopts an estimation strategy of longitudinal-transverse-vertical three-way decoupling estimation, in the embodiment, the longitudinal linear velocity of the front vehicle is reduced, and the brake control input Com _ brk of the front vehicle is estimated through a vehicle longitudinal dynamic model; removing six-degree-of-freedom motion response Yhorizontal of a vehicle body caused by Com _ brk from the vehicle response Yreal before the actual road condition, wherein the difference between the two is Yreal-Yhorizontal, and estimating a vertical component force-component moment of interference input by adopting a vehicle vertical vibration model, wherein the vertical component force-component moment of interference input comprises an interference input vertical component force Fze, a side component moment Mxe and a pitching component moment Mye; and estimating road surface excitation input, aerodynamic acting force and obstacle collision force excitation by combining the weather condition obtained by the sensing module and the state data of the front vehicle.

In the embodiment, the preceding vehicle encounter prediction module adopts a rule-based strategy, and comprehensively judges the obstacle attribute theta encountered by the preceding vehicle within the past time period tau by using the estimation values of driving manipulation input and road surface excitation input within the past time period tau and combining the high-frequency noise of the tyre of the preceding vehicle provided by the sensing module; the obstacle attribute θ includes: the types of the obstacles are deceleration strips, positions, heights, lengths and widths of the deceleration strips.

The automatic driving decision and control module determines the emergency road condition of the front road section in the past time period tau according to the attribute of the obstacle encountered by the front vehicle; based on the traffic road condition information given by the sensing module, an initial decision control strategy Policy _ init is given in advance to the upper driving task of the vehicle, and the vehicle drives forwards along the vehicle. In this example, the front vehicle encounters deceleration during running, and besides the vehicle body movement caused by the uphill road segment, there is a significant abnormal expression, and this part of extra information can be used by the vehicle in advance. Furthermore, considering the predicted emergency road condition that the front vehicle is encountering, adopting a strategy based on rules to compare the attribute theta of the obstacle encountered by the front vehicle in the past time period tau with a set threshold theta to obtain a difference, and then deducing the urgency epsilon (epsilon is between 0 and 1) of the front road section based on the rules according to the size and distribution condition of each dimension in the difference. Finally, according to the attribute theta of the obstacle encountered by the front vehicle and the road condition urgency epsilon, calculating the correction quantity delta Policy of the decision control strategy of the vehicle, including the correction of longitudinal speed, transverse speed, yaw rate and suspension control mode, namely, the correction quantity delta Policy is [ delta Vx ]; Δ Vy; Δ Yawrate; Δ SusMode ]. When the road condition emergency degree epsilon is larger or the obstacle attribute theta indicates that the bumpiness of the road surface is more serious, the longitudinal vehicle speed correction amount delta Vx and the suspension mode correction amount delta SusMode of the vehicle are more negative, and when the obstacle (namely, a deceleration strip) is predicted to be in the position of the area scheduled to run by the vehicle initial decision control strategy, the longitudinal linear speed Vx needs to be reduced, the suspension control mode needs to be adjusted to be a comfort mode, and the influence of the vehicle on safety and comfort when the vehicle passes through the deceleration strip is further reduced. In this example, the predicted deceleration strip is located in the own lane, so the lateral velocity Vy and the yaw rate Yawrate are not adjusted, and only in decision making

Fig. 3, 4 and 5 show three other application scenarios of the present embodiment. In fig. 3, the front vehicle encounters a deceleration strip when following a vehicle on a horizontal road, so that the front vehicle has an obvious vertical acceleration, an obvious pitch angle and deceleration strip noise. In fig. 4 and 5, when the vehicle runs on a horizontal road, the front vehicle encounters one-side concave and convex road respectively, and obvious vehicle body rolling and left-right height difference of the vehicle body edge can be found from the front-view camera of the vehicle. According to the principle, for the three application scenarios, the present invention can also obtain the correction amount Δ of the decision and control of the vehicle, including the corrections of deceleration, steering, etc., based on all the sensing data from the current time T0 and within k sampling steps, according to the preceding vehicle "human-vehicle-road" dynamic model shown in fig. 1, the processes of motion difference calculation, input prediction, preceding vehicle encounter prediction, etc., so as to realize safe driving. Alternatively, the preceding vehicle encounter prediction module may input traffic road condition information (including average speed of traffic flow, noise reference level, front signal light information, and the like) in the past time period τ, a preceding vehicle manipulation input U, and a preceding vehicle interference input W as inputs into the deep neural network based on a deep neural network learning strategy, and predict a road condition that the preceding vehicle is encountering, where the prediction result is an obstacle attribute θ in the past time period τ, where the prediction result is specific attributes of a deceleration strip, a unilateral concavity, and a unilateral convexity of a road surface.

The embodiments of the present invention are not limited to any particular model, and the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made to the embodiments of the present invention by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. A system for automatic driving decision control considering a preceding vehicle encounter is characterized by comprising a sensing module, a leading vehicle and road dynamics modeling module, an actual road condition preceding vehicle response calculating module, a non-emergency road condition preceding vehicle response calculating module, a preceding vehicle abnormal driving behavior judging module, a preceding vehicle operation input and interference input estimating module, a preceding vehicle encounter predicting module and an automatic driving decision and control module;

the sensing module acquires traffic road condition information, vehicle state data and front vehicle state data through a sensor; the traffic condition information comprises obstacles, road surface adhesion coefficients, lane width, lane gradient, lane curvature, road marking, traffic signs, signal lamp states, traffic flow information and weather conditions; the vehicle state data comprises a vehicle body six-degree-of-freedom dynamic state Y _ ego of the vehicle relative to the road surface and vehicle manipulation input; the preceding vehicle state data comprise a six-degree-of-freedom dynamic state Y _ relative of the preceding vehicle relative to the vehicle body of the vehicle, vehicle appearance parameters, tail lamp states, steering indicator lamp states, warning horn sounds, tire noises, power transmission system noises and impact noises; the six-degree-of-freedom dynamic state of the vehicle body comprises linear velocity, angular velocity, linear acceleration, angular acceleration, position and attitude angle of six motion degrees of freedom of a rigid body of the vehicle body; the operation input comprises the operations of steering, braking, driving, indicating lamps and warning horns;

the front passenger and vehicle road dynamics modeling module comprises a vehicle type recognition sub-module and a passenger and vehicle road modeling sub-module; the vehicle type recognition sub-module determines the type of the front vehicle by combining a vehicle type database based on the vehicle appearance parameters in the front vehicle state data output by the sensing module; the man-vehicle-road modeling submodule establishes a front vehicle 'man-vehicle-road' dynamic model:

X(k+1)＝F(X(k),Uk,Wk)

Y(k)＝H(X(k),Uk,Wk)

wherein, X (k) is the state data of the front vehicle at the moment k; uk is the front vehicle operation input within the time of 0-k; wk is the interference input within the time range of 0-k; y (k) is the output response at time k; f is a state transfer function, and H is an output response function;

the actual road condition front vehicle response calculation module superposes Y _ ego and Y _ relative based on the vehicle state data and the front vehicle state data output by the sensing module to obtain a vehicle body six-degree-of-freedom dynamic state YReal of the front vehicle relative to the road surface under the actual road condition;

the non-emergency road condition front vehicle response calculation module calculates a six-degree-of-freedom dynamic state Ynormal of a front vehicle relative to a vehicle body of a road surface in a non-emergency road condition on the basis of traffic road condition information acquired by the sensing module and a front vehicle 'human-vehicle-road' dynamic model in a front vehicle and vehicle road dynamic modeling module; the non-emergency road condition refers to the condition that any obstacle outside the perception capability range of the vehicle is not encountered during the driving of the vehicle in front on a dry and flat road surface;

the front vehicle abnormal driving behavior judging module performs difference on dimensions of YReal and Ynormal to obtain multi-dimensional motion difference, and judges whether the front vehicle driving behavior is abnormal or not according to the motion difference;

the front vehicle operation input and interference input estimation module estimates a front vehicle operation input Uk and a front vehicle interference input Wk according to YReal on the basis of a front vehicle human-vehicle-road dynamic model in a front vehicle road dynamic modeling module and traffic road condition information obtained by a sensing module after the front vehicle abnormal driving behavior judgment module judges that the front vehicle driving behavior is abnormal;

when the preceding vehicle encounter prediction module judges that the driving behavior of the preceding vehicle is abnormal, the obstacle attribute theta of the preceding vehicle encounter is predicted based on the traffic road condition information and the preceding vehicle state data output by the sensing module, and the preceding vehicle operation input and the preceding vehicle interference input estimated by the preceding vehicle operation input and interference input estimation module, wherein the obstacle attribute theta comprises the type, the position and the collision force with the preceding vehicle, the depth or the height, the length and the width of a static obstacle, the road surface unevenness and the speed and the track of a dynamic obstacle;

the automatic driving decision and control module is specifically that firstly, according to the traffic road condition information Info _ traffic acquired by the sensing module, an automatic driving algorithm gives an initial vehicle decision control strategy Policy _ init ═ Vx not considering the encounter of a front vehicle; vy; yawrate; FuturePath; SusMode ] including a longitudinal speed Vx, a lateral speed Vy, a yaw rate, a future travel trajectory FuturePath, and a suspension control mode SusMode of the host vehicle for a future period of time; the suspension control modes comprise a comfort mode, a conventional mode and a controllability mode, and the corresponding assignments are-1, 0 and 1 in sequence; then, according to the attribute theta of the obstacle encountered by the front vehicle predicted by the front vehicle encountered prediction module and the traffic road condition information output by the sensing module, calculating the correction delta Policy of the decision control strategy of the vehicle as [ delta Vx by using a correction strategy; Δ Vy; Δ Yawrate; Δ FuturePath; Δ SusMode ], Δ Vx is a longitudinal vehicle speed correction amount, Δ Vy is a lateral speed correction amount, Δ Yawrate is a yaw rate correction amount, Δ FuturePath is a future travel trajectory correction amount, Δ SusMode is a suspension control mode correction amount; finally, Policy _ init is corrected in accordance with Δ Policy, and the desired motion of the vehicle is implemented.

2. The system for automated driving decision control in consideration of preceding vehicle encounter according to claim 1, wherein the determination method in the preceding vehicle abnormal driving behavior determination module is a threshold-setting-based method, a machine-learning-based classification method, or a machine-learning-based regression method;

the method based on the set threshold specifically includes that when the motion difference of a certain dimension is larger than the corresponding set threshold, the dynamic state of the corresponding dimension of the front vehicle is abnormal, and as long as the dynamic state of the front vehicle of one dimension is abnormal, the driving behavior of the front vehicle is abnormal; otherwise, the dynamic state of the front vehicle is not abnormal, and the decision control of the vehicle is not required to be corrected;

the classification method based on machine learning specifically comprises the steps of inputting motion differences into a machine learning classification model, and classifying the driving behaviors of the front vehicle, wherein the classification result comprises abnormity and normality;

the regression method based on machine learning specifically comprises the steps of taking motion difference as input of a machine learning regression model, outputting abnormal degree vectors lambda, wherein lambda corresponds to each dimension of a six-degree-of-freedom dynamic state of a vehicle body one by one, and elements of the vectors lambda are all in [0,1 ]; when the degree of abnormality of a certain dimension is greater than a correspondingly set threshold value, the dynamic state of the corresponding dimension is abnormal, and if the dynamic state of a front vehicle of one dimension is abnormal, the driving behavior of the front vehicle is abnormal; otherwise, the dynamic state of the front vehicle is not abnormal, and the decision control of the vehicle is not required to be corrected;

the training of the machine learning regression model is specifically to simulate the motion difference between a front vehicle and a non-emergency road condition when the front vehicle encounters different emergency road conditions through a vehicle dynamics model, and train the machine learning regression model based on abnormal degree vectors corresponding to different motion differences; defining dimensions of an abnormal degree vector lambda when an accident occurs as 1, and defining dimensions of the abnormal degree vector lambda when normal driving as 0; counting the six-degree-of-freedom dynamic state change of the vehicle body caused by corresponding deceleration and steering operations when the front vehicle encounters an emergency road condition and is successfully avoided, obtaining the maximum variable quantity of each dimension of the dynamic state, and forming a vector delta Ysimmax; defining each dimension lambda of the abnormal degree vector of the dynamic state_iComprises the following steps:

the system comprises a simulation system, a road surface simulation system and a road surface simulation system, wherein Ysimreal is a six-degree-of-freedom dynamic state of a vehicle body of a front vehicle relative to the road surface in the simulation system under actual road conditions; ysimnormal is a six-degree-of-freedom dynamic state of a vehicle body of a front vehicle relative to a road surface under a non-emergency road condition in simulation;

is a constant greater than zero to ensure λ_iBetween 0 and 1; subscript i denotes the ith dimension in the vector; the emergency road condition refers to the situation that a preceding vehicle encounters an obstacle outside the perception capability range of the vehicle during running.

3. The system for automated driving decision control taking into account preceding vehicle encounters according to claim 1, wherein the estimation strategy in the preceding vehicle maneuver and disturbance input estimation module is a vertical-horizontal-vertical decoupling estimation or a vertical-horizontal-vertical joint estimation;

the longitudinal-transverse-vertical joint estimation adopts a six-degree-of-freedom vehicle dynamic model to estimate the control input of braking, driving and steering of the front vehicle and the interference input;

the longitudinal-transverse-vertical three-direction decoupling estimation method specifically comprises the following steps: according to the longitudinal dynamics state in YReal and the longitudinal lane gradient in the traffic road condition information, a vehicle longitudinal dynamics model is adopted to estimate a brake control input Com _ brk, a drive control input Com _ drv and a longitudinal component force Fxe of interference input of a front vehicle; according to the transverse dynamic state in Yreal and the transverse lane gradient in the traffic road condition information, a vehicle steering control dynamic model is adopted to estimate a front vehicle steering control input Com _ str, a lateral component force Fye of an interference input and a yaw component moment Mze of the interference input; removing six-degree-of-freedom motion response Yhorizontal of a vehicle body caused by Com _ brk, Com _ drv, Com _ str, Fxe, Fye and Mze from Yreal to obtain a difference value Yreal-Yhorizontal, and estimating a vertical component force and a component force of interference input, including a vertical component force Fze, a roll component force Mxe and a pitch component force Mye of the interference input, as input of a vehicle vertical vibration model; the final estimated preceding vehicle maneuver inputs include a brake maneuver input Com _ brk, a drive maneuver input Com _ drv, and a steering maneuver input Com _ str; the estimated front vehicle disturbance inputs include the longitudinal component Fxe, the lateral component Fye, the yaw component moment Mze, the vertical component Fze, the roll component moment Mxe, and the pitch component moment Mye of the disturbance input.

4. The system for automated driving decision control taking into account preceding vehicle encounters according to claim 1, wherein the predictive strategy in the preceding vehicle encounter prediction module is a rule-based predictive strategy or a deep neural network learning-based predictive strategy;

the rule-based prediction strategy comprises rule base query and fuzzy rule reasoning, and comprehensively judges the barrier attribute theta encountered by the front vehicle in the past time period according to the estimated values of the control input and the interference input of the front vehicle in the past time period, tire noise and impact noise;

the prediction strategy based on the deep neural network learning specifically comprises the steps that traffic road condition information in the past time period, an estimated value of the previous vehicle operation input and an estimated value of the previous vehicle interference input are used as the input of the deep neural network, and the prediction result of the deep neural network is the obstacle attribute theta encountered by the previous vehicle in the past time period.

5. The system for automated driving decision control in view of preceding vehicle encounters according to claim 1, wherein the modification strategy in the automated driving decision and control module is a rule-based modification strategy or a machine learning-based modification strategy;

the rule-based correction strategy specifically comprises: inputting the type and position of the obstacle in the obstacle attribute theta encountered by the front vehicle and the depth or height, length and width of the static obstacle and the road surface unevenness into a machine learning model ML _ Comfort for judging the Comfort influence, and outputting whether the influence is caused on the future driving Comfort of the vehicle; inputting all dimensions in the attribute theta of the obstacle encountered by the front vehicle into a machine learning model ML _ Safe for judging the influence of safety, and outputting whether the influence is caused on the safety of future driving of the vehicle; as long as one of the Comfort judgment model ML _ Comfort and the safety judgment model ML _ Safe has an influence, the decision control strategy of the vehicle needs to be corrected, otherwise, the decision control strategy of the vehicle does not need to be corrected; according to the attribute of the obstacle encountered by the front vehicle and the traffic road condition information, solving the correction quantity of the vehicle decision control strategy:

ΔPolicy＝f_correct(θ,Info_traffic)

wherein f _ correct is a correction calculation function;

the correction strategy based on machine learning specifically comprises the steps that the attribute theta of a barrier encountered by a front vehicle in the past time period and traffic road condition information output by a sensing module are input into a machine learning model, and output is correction quantity delta Policy for decision and control of the vehicle.

6. The system for automated driving decision control considering preceding vehicle encounters as claimed in claim 5, wherein in the automated driving decision and control module, the f _ correct construction rule in the machine learning based correction strategy is: under given traffic road condition information Info _ traffic, when the output of the Comfort judgment model ML _ Comfort is influential, the larger the road surface unevenness, the closer the position of the obstacle and the larger the size of the static obstacle in the obstacle attribute theta are, the larger and negative the values of the vehicle longitudinal vehicle speed correction quantity DeltaVx and the suspension mode correction quantity DeltasMode are; when the output of the safety calculation model ML _ Safe has an influence, a future travel trajectory correction amount Δ FuturePath is determined to avoid an imminent collision based on the position of the obstacle in the obstacle attribute θ, the depth or height of the static obstacle, the length or width of the static obstacle, or the speed or trajectory of the dynamic obstacle, and the correction amounts Δ Vx, Δ Vy, Δ Yawrate, and Δ SusMode are determined by an automatic driving control algorithm based on the corrected future travel trajectory FuturePath + Δ FuturePath.

7. The system for automated driving decision control considering preceding vehicle encounters according to claim 1, wherein the categories of obstacles include traffic-forbidden signals, static obstacles, and dynamic obstacles.

8. The system for automated driving decision control considering preceding vehicle encounters according to claim 1 wherein the disturbance inputs include road surface excitation inputs, aerodynamic forces, and obstacle impact force excitations.

9. The system for automated driving decision control considering preceding vehicle encounters according to claim 1 wherein the forms of functions F and H include a logic rule model, a transfer function model, a linear state space model, a non-linear state space model, and a neural network model; the linear state space model comprises a vehicle six-degree-of-freedom dynamic model, a vehicle longitudinal dynamic model, a vehicle steering control dynamic model, a vehicle single-degree-of-freedom vertical vibration model, a vehicle two-degree-of-freedom vertical vibration model and a vehicle vertical-pitching-rolling vibration model.

10. The system for automated driving decision control taking into account preceding vehicle encounters according to claim 1, wherein the travel trajectory includes changes in position of various points of the vehicle body over time and changes in attitude of the vehicle body over time.

Images