CN114407902B

CN114407902B - Driving decision system based on road water layer depth estimation

Info

Publication number: CN114407902B
Application number: CN202210059775.0A
Authority: CN
Inventors: 李道飞; 张家杰; 潘豪; 肖斌; 陈林辉; 蒋鑫
Original assignee: Zhejiang University ZJU
Current assignee: Zhejiang University ZJU
Priority date: 2022-01-19
Filing date: 2022-01-19
Publication date: 2023-11-28
Anticipated expiration: 2042-01-19
Also published as: CN114407902A

Abstract

The invention discloses a driving decision system based on road water layer depth estimation, which comprises the steps of calculating actual motion response of a front vehicle according to state information of the vehicle and state information of the front vehicle, estimating front vehicle parameters by utilizing a vehicle type database and estimating front vehicle input based on the actual response; judging the attribute of the water layer medium according to the splashing information of the front water layer and identifying the splashing characteristic parameter; based on the actual response of the front vehicle, the front vehicle parameters and input estimation, the water layer attribute and splash characteristic parameter discrimination results, firstly, the water layer size and position are judged, finally, the front water layer depth is predicted, and the automatic driving decision, track planning and control are corrected according to the prediction results. According to the invention, the depth of the water layer is predicted based on the driving behavior, the movement state and the splashing characteristics of the water layer when the front vehicle drives through the water layer, the defect of the sensing range of the existing sensor is overcome, the capability boundary of automatic driving sensing and prediction is expanded, the automatic driving decision and control strategy is corrected based on the capability boundary, and the driving safety, the comfort and the politicity are improved.

Description

Driving decision system based on road water layer depth estimation

Technical Field

The invention belongs to the technical field of intelligent network vehicles and automatic driving, and particularly relates to a driving decision system based on road water layer depth estimation.

Background

The water layer and puddle generated by road ponding are called as "invisible killers" on the road, and a vehicle can generate a "water slipping phenomenon" when passing through the water layer at a higher vehicle speed. That is, the wheels cannot be in direct contact with the ground due to the fact that the wheels generate extremely large acting force with the water layer due to the high rotating speed, so that the vehicle cannot obtain enough ground braking force, and finally serious consequences of out-of-control of the vehicle are extremely easy to occur. Therefore, whether the driver can obtain instant perception on the water layer information on the driving path in the driving process is a key for ensuring the safe driving of the vehicle. For example, the expressway generates a water accumulation layer with a certain depth due to heavy rainfall, unsmooth road drainage and the like, and the rear vehicle is extremely easy to cause a water sliding phenomenon due to the fact that a driver of the rear vehicle cannot find a road water layer in time and incorrectly estimates the water layer depth, so that serious traffic accidents are caused.

In addition to affecting the driving safety of the host vehicle, water layer splatter generated when the host vehicle drives across the water layer can have varying degrees of negative impact on surrounding traffic participants. For example, the vehicle chooses to drive through the water layer at a faster speed due to misestimation of the front water layer depth, resulting in large-scale splash liquid falling onto the windshield of the side-rear vehicle; this will greatly affect the steering control of the sidecar. Therefore, the splash characteristic of the water layer generated when the front vehicle passes through the water layer is directly reflected on the depth and the size range of the water layer in front, and the splash characteristic has important reference significance for decision control of the vehicle.

In recent years, with the development of automatic driving technologies, various types of advanced driving assistance systems (ADAS, advanced Driving Assistance System) such as adaptive cruise (ACC, adaptive Cruise Control), lane keeping assistance, automatic emergency braking (AEB, autonomous Emergency Braking) and the like are applied to common vehicles on a large scale. In the design scheme of most of ADAS systems at present, the high-precision sensors of the vehicle or other vehicle sharing information is obtained through the vehicle networking technology to sense the surrounding environment of the vehicle, and the planning decision of the vehicle and the real-time control of steering, braking, driving and the like are realized through high-speed operation, so that safe running is realized. Among them, an adaptive cruise system (ACC) refers to a system that detects a front vehicle using a radar or a camera, and maintains a distance from the front vehicle by controlling a brake and a throttle according to a driver's setting of a following speed and a following distance.

However, in the presently disclosed ADAS vehicle-mounted sensor, neither the lidar nor the high-definition camera can obtain better perception of the road water layer depth, and especially under the night rainfall condition, the perception capability of the road water layer depth is lacking. Thus, if the autonomous vehicle is unable to obtain a forward water layer depth estimate, decision control is made based solely on existing perception algorithms, possibly with serious consequences. For example, the ACC system makes a braking or accelerating instruction according to the speed of the front vehicle and the distance between the front vehicle and the front vehicle, if the front vehicle suddenly accelerates after driving through the water layer, the vehicle still makes decision control for accelerating and following the vehicle due to neglecting the front water layer depth estimation, so that the vehicle drives through the water layer too fast to generate a water slipping phenomenon. For another example, activating the AEB function of automatic emergency braking deeper in the water layer may cause a traffic accident caused by insufficient braking.

In summary, no report is available at present on the prediction and estimation of the depth of the front water layer according to the liquid splashing generated when the front vehicle passes through the water layer, and the decision control correction of the vehicle is carried out on the basis of the prediction and estimation.

Disclosure of Invention

The invention aims to provide a driving decision system based on road water layer depth estimation, aiming at the defects of the prior art. The invention considers the driving behavior, the movement state and the water layer splashing characteristic of the front vehicle when the front vehicle passes through the water layer, and is used for decision control of the vehicle.

The aim of the invention is realized by the following technical scheme: a driving decision system based on road water layer depth estimation comprises a sensing module, a front vehicle actual response calculation module, a front vehicle parameter and input estimation module, a water layer medium attribute discrimination module, a water layer splash characteristic parameter identification module, a front water layer judgment and water layer depth prediction module, a self-driving decision module and the like.

The sensing module acquires road traffic information, host vehicle state information, front vehicle state information and front water layer splashing information. The road traffic information comprises road adhesion coefficient, lane width, lane number, road gradient, lane curvature, lane marking, obstacles, traffic signs, traffic light states, pedestrian information, traffic flow information and weather states; the vehicle state information comprises a vehicle parameter delta, a vehicle body six-degree-of-freedom motion state Y_ ego of the vehicle relative to the ground and operation input; the front vehicle state information comprises a six-degree-of-freedom motion state Y_relative of a front vehicle relative to a vehicle body of the vehicle, vehicle appearance parameters, license plate information, brake indicator lamp states, steering indicator lamp states, tire rolling marks, horn warning sounds, tire road noise and power transmission system noise; the front water layer splash information includes splash liquid color, splash liquid reflectivity, splash liquid refractive index, liquid splash trajectory, liquid splash direction, liquid splash time, and splash liquid landing sound. The six-degree-of-freedom dynamic state of the vehicle body comprises linear speeds, angular speeds, linear accelerations, angular accelerations, positions and attitude angles of six degrees of freedom of motion of the rigid body of the vehicle body; the steering inputs include steering, braking, driving, indicator lights and warning horn operations.

The front vehicle actual response calculation module is used for superposing the six-degree-of-freedom motion state Y_ ego of the vehicle body relative to the ground and the six-degree-of-freedom motion state Y_relative of the front vehicle relative to the vehicle body based on the vehicle state information and the front vehicle state information output by the sensing module according to the vehicle relative motion synthesis principle, so as to obtain the six-degree-of-freedom motion state Y_real of the front vehicle relative to the road surface under the current road condition.

The front vehicle parameter and input estimation module comprises a front vehicle parameter estimation sub-module and a front vehicle operation interference input estimation sub-module. The front vehicle parameter estimation sub-module is used for estimating the front vehicle parameter gamma based on the vehicle appearance parameter, license plate information, tire rolling marks and tire road noise in the front vehicle state information output by the sensing module and determining the front vehicle type by combining a vehicle type database. The front vehicle operation interference input sub-module is used for estimating front vehicle input I through a vehicle dynamics model based on the six-degree-of-freedom motion state Y_real of the front vehicle relative to the vehicle body of the road surface, which is output by the front vehicle actual response calculation module, and the front vehicle parameter gamma which is output by the front vehicle parameter estimation sub-module, wherein the front vehicle input I comprises operation input and interference input.

The water layer splashing characteristic parameter identification module is used for identifying the characteristic parameter alpha of the front water layer splashing motion based on the liquid splashing track, the liquid splashing direction and the liquid splashing time output by the sensing module.

The water layer medium attribute judging module is used for identifying the type of liquid in the water layer encountered by the front vehicle based on the splash liquid color, the splash liquid reflectivity, the splash liquid refractive index and the splash liquid landing sound output by the sensing module by utilizing a liquid judging method, and further obtaining the inherent attribute beta of the liquid in the water layer encountered by the front vehicle.

The front water layer judging and water layer depth predicting module is used for judging the size and the position POS of a front water layer by utilizing an image recognition technology and predicting the water layer depth d by utilizing a depth predicting method based on the six-degree-of-freedom motion state Y_real of the front vehicle relative to the vehicle body of the road surface, the front vehicle parameters and the front vehicle steering input and interference input output by the front vehicle actual response calculating module, the front water layer splash motion characteristic parameter alpha output by the water layer splash characteristic parameter recognizing module and the inherent attribute beta of water layer medium property judging module. The size of the front water layer comprises the length L and the width W of the water layer.

The automatic driving decision module of the host vehicle firstly calculates an initial host vehicle decision control strategy strategy_ini= [ Vx ] which does not consider the geometric size and the position of a front water layer, the depth d of the water layer and the attribute beta of the water layer by adopting an automatic driving algorithm of the host vehicle based on road traffic information info_traffic output by the perception module; vy; yawrate; futurepath; susmode ], wherein the decision control strategy includes a longitudinal speed Vx, a lateral speed Vy, a yaw rate Yawrate, a future travel trajectory Futurepath, and a suspension control mode Susmode for a period of time in the future of the host vehicle. Next, outputting a dangerous grade early warning to a driver based on the vehicle state information and the road traffic information info_traffic output by the sensing module, the water layer attribute beta output by the water layer medium attribute judging module, the water layer size and the geometric position POS and the water layer depth d output by the front water layer judging and water layer depth predicting module, and outputting a vehicle decision control strategy correction quantity delta strategy_scp= [ delta Vx_scp by utilizing a correction strategy; Δvy_scp; Δyawrate_scp; Δfuturepath_scp; Δsusmode_scp ]. And finally, correcting an initial vehicle decision control strategy strategy_ini according to the correction quantity deltastrategy_scp, and controlling the future motion state of the vehicle, including future motion trail, linear speed, linear acceleration, angular speed, angular acceleration and suspension control mode.

Further, the estimation strategy in the front truck maneuver interference input sub-module includes a combined longitudinal-transverse-vertical estimation or a decoupled longitudinal-transverse-vertical estimation.

The longitudinal-transverse-vertical joint estimation is specifically as follows: based on the six-degree-of-freedom dynamics model of the vehicle, driving, braking and steering control inputs and disturbance inputs of the front vehicle are estimated according to the six-degree-of-freedom motion state Y_real of the front vehicle relative to the road surface and the road gradient in the road traffic information.

The decoupling estimation of the longitudinal-transverse-vertical directions is specifically as follows: estimating a brake control input input_brk and a drive control input input_drv of a front vehicle by utilizing a vehicle longitudinal dynamics model according to the longitudinal dynamics state in Y_real and the longitudinal road gradient in road traffic information, and interfering the longitudinal component Fxe of the input; estimating a steering operation input_str, a lateral component force Fye of disturbance input and a yaw moment Mze of disturbance input of a front vehicle by utilizing a vehicle transverse dynamics model according to the transverse dynamics state in Y_real and the transverse road gradient in road traffic information; next, the six-degree-of-freedom motion response y_horizontal of the vehicle body caused by input_brk, input_drv, input_ str, fxe, fye, and Mze is subtracted from y_real, and the difference y_real-y_horizontal is used as an input of the vehicle vertical vibration model to estimate the vertical component Fze of the disturbance input, the roll component moment Mxe of the disturbance input, and the pitch component moment Mye of the disturbance input. The final estimated front vehicle steering inputs include a brake steering input input_brk, a drive steering input_drv, and a steering input input_str; the estimated front truck disturbance inputs include the disturbance input longitudinal component Fxe, lateral component Fye, vertical component Fze, yaw moment Mze, roll moment Mxe, and pitch moment Mye.

Further, the identification method in the water layer splash characteristic parameter identification module specifically comprises the following steps:

according to the liquid splashing direction and the liquid splashing track output by the perception module in the past time period, the transverse distance L_side, the maximum vertical height H_side and the longitudinal distance X_side of the front water layer side wave splashing are obtained by utilizing a computer vision recognition algorithm, and the maximum vertical height H_pickup and the longitudinal distance X_pickup of the tread splashing are obtained. Next, according to the liquid splashing time output by the sensing module and based on a physical motion law, deducing and calculating to obtain a side wave velocity V_side, wherein the angle between the projection of a side wave velocity vector in an xz plane and an x-axis is theta 1, the angle between the projection of the side wave velocity vector in a yz plane and a z-axis is theta 2, the angle between the projection of the side wave velocity vector in the yz plane and a y-axis is theta 3, the tread splashing velocity V_pick up and the tread splashing velocity vector form an angle theta 4 with the ground; the water layer splash characteristic attribute alpha is obtained and comprises L_side, H_side, X_side, V_side, V_jackup, theta 1, theta 2, theta 3 and theta 4.

Further, the liquid discriminating method in the water layer medium attribute discriminating module comprises a table look-up method based on an off-line database or a discriminating method based on machine learning.

The table look-up method based on the off-line database specifically comprises the following steps: road liquids of different categories and concentrations are prepared, basic parameters of the prepared liquids in a splashing state are recorded, wherein the basic parameters comprise splashing liquid color LQ_color, splashing liquid reflectivity LQ_reflection, splashing liquid refractive index LQ_reflection and splashing liquid landing sound LQ_sound, and a liquid database is formed. And then, according to the color LQ_color, the reflectivity LQ_reflection, the refractive index LQ_reflection and the landing sound LQ_sound of the splashing liquid output by the sensing module, searching the attribute beta of the splashing liquid in front of the current database by comparing with the established liquid database, wherein the attribute beta comprises the category and the corresponding concentration of the liquid.

The machine learning based discrimination method specifically comprises the following steps: and taking the splash liquid color, the splash liquid reflectivity, the splash liquid refractive index and the splash liquid landing sound obtained by the perception module in the past time period as inputs of a machine learning judgment model, and outputting the input as the attribute beta of the front splash liquid, wherein the attribute beta comprises the type and the corresponding concentration of the liquid.

Further, the prediction method in the front water layer judgment and water layer depth prediction module comprises a prediction method based on a mechanism model or a prediction method based on deep neural network learning.

The prediction method based on the mechanism model specifically comprises the steps of predicting the depth d of a front water layer according to a front vehicle six-degree-of-freedom motion state Y_real output by a front vehicle actual response calculation module, a front vehicle parameter gamma output by a front vehicle input I and a front water layer medium attribute judgment module, a front splashing liquid attribute beta output by a water layer splashing characteristic parameter identification module, a front water layer splashing movement characteristic parameter alpha output by the water layer splashing characteristic parameter identification module, the length L, the width W and the geometric position POS of the front water layer which are judged previously by the front vehicle actual response calculation module, and a front vehicle parameter gamma and an input estimation module of the front vehicle, wherein the front water layer is calculated according to a dynamics model of water layer splashing:

s is a water layer depth calculation function when the wheels on one side of the front vehicle drive through the water layer to generate splashing, and D is a water layer depth calculation function when the wheels on two sides of the front vehicle drive through the water layer to generate splashing.

The prediction method based on deep neural network learning specifically comprises the following steps: the six-degree-of-freedom motion state Y_real of the front vehicle, the vehicle parameter gamma of the front vehicle, the input I of the front vehicle, the attribute beta of front splashing liquid, the characteristic parameter alpha of front water layer splashing motion, the length L, the width W and the geometric position POS of the front water layer which are judged by the module in advance are taken as inputs of a depth neural network, and the prediction result of the depth neural network is the depth d of the front water layer in the past time period.

Further, the correction strategy in the automatic driving decision module of the host vehicle comprises a correction strategy based on rules or a correction strategy based on machine learning.

The rule-based correction strategy specifically comprises: based on an automatic driving algorithm of the vehicle, the length L, the width W, the geometric position POS and the water layer depth d of the front water layer output by a front water layer judging and water layer depth predicting module are compared with six-degree-of-freedom motion state Y_ ego of the vehicle body relative to the ground, vehicle parameter delta and road traffic information info_traffic, whether the vehicle is driven through the front water layer or not is judged, driving safety, comfortableness and politics are comprehensively considered, and the vehicle decision control strategy correction quantity delta strategy_scp is calculated and obtained.

The correction strategy based on machine learning is specifically as follows: by setting different known conditions including the length L and the width W of a front water layer, the geometrical position POS and the depth d of the water layer, the six-degree-of-freedom motion state Y_ ego of the vehicle body, the vehicle parameter delta and the road traffic information info_traffic, and utilizing the rule-based correction strategy, the optimal vehicle decision control strategy correction amount comprehensively considering the driving safety, the comfort and the politicity under different known conditions is calculated offline. And training a machine learning model according to the corresponding relation between different known conditions and the calculated optimal vehicle decision control strategy correction amount, and finally obtaining a model fcorrect. L, W, POS, d, Y _ ego, δ, info_traffic in the past period are input to the machine learning model fcorrect, and the correction amount Δstrategy_scp of the driving decision control of the host vehicle is output.

Further, the rule-based correction strategy includes:

(A) If the host vehicle initial decision control strategy strategy_ini would result in driving through the front water layer, then:

(A1) L, W, POS, d, Y _ ego, delta and info_ traffic, strategy _ini are input into a safety-based correction function fcorrect_safe, and a safety correction quantity delta strategy_safe is solved; each variable Δstrategy_safe is the amount of area.

(A2) Inputting the Δstrategy_safe calculated in the step (A1), L, W, POS, d, Y _ ego, δ and info_ traffic, strategy _ini into a comfort-based correction function fcorrect_scomfort to obtain a comfort correction amount Δstrategy_scomfort. The range of the variables Δstrategy_scomfort are all within or equal to the range of Δstrategy_safe.

(A3) Polite correction amount deltastrategy_scitate is calculated.

(A31) If step (A1) cannot avoid directly driving through the water layer by Δstrategy_safe, Δstrategy_ scomfort, L, W, POS, d, Y _ ego, δ, info_ traffic, strategy _ini are input into polite-based correction function fcorrect_scplite to obtain polite correction amount Δstrategy_scplite. The range of the deltastrategy_scitate variables is within or equal to the range of deltastrategy_scimfort.

(A32) If step (A1) can avoid directly driving through the water layer by Δstrategy_safe, after step (A2) calculates the comfort correction amount Δstrategy_scomfort, no consideration is required for correction for polite driving, and the range of each variable of Δstrategy_scomfort is equal to Δstrategy_scomfort.

(A4) Outputting the decision control strategy correction quantity of the vehicle: the optimal value of each correction amount is selected from the Δstrategy_scite interval amount as the own-vehicle decision control strategy correction amount Δstrategy_scp. Wherein each dimension Δvx_scp, Δvy_scp, Δyawrate_scp, Δfuturepath_scp, Δsusmode_scp of Δstrategy_scp is obtained by maximizing the objective function J:

J＝w1*g ₁ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)+w2*g ₂ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)+w3*g ₃ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)

Wherein w1, w2 and w3 respectively represent weights of safety, comfort and politicity, and a function g ₁ 、g ₂ 、g ₃ And respectively calculating safety, comfort and politicity indexes of the vehicle under the decision control strategy correction quantity delta strategy_scp of the vehicle, wherein the constraint condition of each variable of delta strategy_scp is a section corresponding to each dimension of delta strategy_scplite.

(B) If the host vehicle initial decision control strategy strategy_ini does not result in driving through the front water layer, then:

(B1) Only the comfort correction amount Δstrategy_comfort ' is calculated, and y_ ego, δ, info_ traffic, strategy _ini are input to the comfort correction function fcorrect_comfort ' to obtain Δstrategy_comfort '. Each variable Δstrategy_comfort' is an interval amount.

(B2) Outputting the decision control strategy correction quantity of the vehicle: the optimum value Δstrategy_scp of each correction amount is selected from the interval amounts of Δstrategy_comfort'.

Further, in the step (A1), the construction rule of the fcorrect_safe function is: on the premise of adhering to the road traffic rules, whether the aim of avoiding directly driving through a water layer can be achieved is judged according to the automatic driving algorithm of the vehicle. If the automatic driving algorithm can be realized, delta strategy-safe is calculated according to the automatic driving algorithm of the vehicle, wherein the delta strategy-safe comprises correction amounts delta Vx-safe, delta Vy-safe, delta Yawrate-safe, delta futurePath-safe and delta SusMode-safe, and the vehicle is prevented from directly driving through a water layer. If the correction cannot be realized, under the premise of ensuring the safety of the vehicle when driving through the water layer, calculating delta strategy_safe including correction amounts delta Vx_safe, delta Vy_safe, delta Yawrate_safe, delta futurePath_safe and delta SusMode_safe; specifically, the calculation rule is that the longer the longitudinal length L, the wider the transverse width W, and the deeper the depth d of the water layer, the greater the potential danger of the water layer in front, the greater and negative the correction amounts Δvx_safe, Δvy_safe, Δyawrate_safe, Δfuturepath_safe, Δsusmode_safe of the host vehicle.

Further, in the step (A2), the construction rule of the fcorrect_scomfort function is: on the premise that deltastrategy_safe ensures the safety of the vehicle, deltastrategy_scofort, which comprises deltavx_scofort, deltavy_scofort, deltayawrate_scofort, deltafuturepath_scofort and deltasusmode_scofort, is calculated with the riding comfort of the vehicle passenger as a target; the calculation rule is specifically as follows:

if the Δstrategy_safe calculated in the step (A1) cannot avoid directly driving through the water layer, the deeper the depth d of the front water layer is, which indicates that the road is uneven, the smaller the change rate of the correction amounts Δvx_scomfort, Δvy_scomfort, Δyawrate_scomfort, Δfuturepath_scomfort, Δsusmode_scomfort along with time is, so as to improve the comfort of the vehicle;

if Δstrategy_safe calculated in step (A1) can avoid directly driving through the front water layer, the rate of change of correction amounts Δvx_scomfort, Δvy_scomfort, Δyawrate_scomfort, Δfuturepath_scomfort, Δsusmode_scomfort over time is made as small as possible.

Further, in the step (B1), the construction rule of the fcorrect_compact' function is: the correction amounts Δstrategy_comfort' are calculated for the driving comfort of the host vehicle, including Δvx_comfort, Δvy_comfort, Δyawrate_comfort, Δfuturepath_comfort, and Δsusmode_comfort. The more uneven the road on the path planned by the initial decision control strategy strategy_ini of the host vehicle, the larger the host vehicles Vx, vy, Δvx_compact, Δvy_compact, Δyawrate_compact, Δfuturepath_compact, and Δsusmode_compact are, the smaller the rate of change over time.

Further, in the step (a 31), the construction rule of the fcorrect_scpolite function is: on the premise that deltastrategy_scomfort ensures the safety and the comfort of the vehicle driving through the water layer, deltastrategy_scipline, including deltavx_scipline, deltavy_scipline, deltayawrate_scipline, deltafuturepath_scipline and deltasusmode_scipline, is calculated with the goal of polite driving. The calculation rule is specifically as follows: when the longitudinal length L of the water layer is longer, the transverse width W is wider, the depth d is deeper, the longitudinal speed Vx and the transverse speed Vy of the vehicle in a six-degree-of-freedom motion state of the vehicle body relative to the horizontal ground are larger, which indicates that the negative influence on surrounding traffic vehicles, non-motor vehicles and pedestrians caused by liquid splashing generated when the vehicle passes through the front water layer is larger, the numerical values of correction amounts DeltaVx_scitate, deltaVy_scitate, deltaYawrate_scitate, deltaFuturePath_scitate are larger and negative, and DeltaSusMode_scinate is equal to DeltaSusMode_scifor improving the driving polishness of the vehicle.

The beneficial effects of the invention are as follows:

1. the influence of driving behavior, movement state and water layer splashing characteristics on the decision control of the vehicle when the front vehicle drives through the water layer is considered, and the decision, track planning and control of automatic driving are corrected by predicting the water layer attribute, the geometric position and the water layer depth existing on the front road, so that the safety in the driving process of the vehicle is improved;

2. The method has the advantages that the geometric position and depth of the front water layer are predicted by comprehensively considering the movement state of the front vehicle, the parameters of the front vehicle and the splashing characteristics of the front water layer, so that the information acquisition and predictability of automatic driving decision control are effectively increased, the effectiveness of the decision control is improved, and the driving safety can be effectively improved; especially under the conditions of night, rainy days and the like, the capability of a driver and a vehicle-mounted sensor platform for acquiring road information is limited, and the invention can obviously expand the capability boundary of automatic driving perception and prediction;

3. the method estimates the front water layer attribute, the geometric position and the water layer depth by predicting, corrects the decision control of the vehicle based on the front water layer attribute, the geometric position and the water layer depth, ensures the driving safety of the vehicle and improves the comfort of passengers; and the influence of splash on the side vehicles, non-motor vehicles and pedestrians is reduced to the greatest extent, so that polite driving is realized. Besides full-automatic driving vehicles, the invention can also be applied to manual driving vehicles and semi-automatic driving vehicles based on the form of reminding or driving active intervention, and improves the safety, comfort and politicity of vehicle driving.

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 and together with the description serve to explain the invention.

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

FIG. 2 is a schematic illustration of a traffic light intersection turning right and driving through a water layer;

FIG. 3 is a schematic view of the front truck after traveling through the water layer and producing splash;

FIG. 4 is a schematic illustration of a side view of splash generated when a front truck is driven through a water layer;

FIG. 5 is a schematic diagram of side wave (left) versus tread splash (right) velocity vectors as a front truck travels through a water layer;

FIG. 6 is a schematic top view of a front truck traveling straight on both sides while traveling through a water layer;

FIG. 7 is a schematic top view of a straight-going right single-sided drive across a wide water layer for a side-to-front vehicle.

Detailed Description

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

As shown in FIG. 1, the system for driving decision based on road water layer depth estimation comprises a sensing module, a front vehicle actual response calculation module, a front vehicle parameter and input estimation module, a water layer medium attribute discrimination module, a water layer splash characteristic parameter identification module, a front water layer judgment and water layer depth prediction module and a self-driving decision module.

(1) The sensing module acquires road traffic information, host vehicle state information, front vehicle state information and front water layer splashing information through the sensor. The information is obtained through a vehicle-mounted high-definition camera, a laser radar, a vehicle-mounted noise sensor, a high-precision map and vehicle networking information.

The road traffic information includes road surface attachment coefficient, lane width, number of lanes, road gradient, lane curvature, lane markings, obstacles, traffic signs, traffic light status, pedestrian information, traffic flow information, and weather status.

The vehicle state information comprises a vehicle parameter delta, a vehicle body six-degree-of-freedom motion state Y_ ego of the vehicle relative to the horizontal ground and a manipulation input. The steering inputs include steering, braking, driving, indicator lights and warning horn operations.

The front vehicle state information comprises a six-degree-of-freedom motion state Y_relative of a front vehicle relative to a vehicle body of the vehicle, vehicle appearance parameters, license plate information, brake indicator lamp states, turn indicator lamp states, tire rolling marks, loudspeaker warning sounds, tire road noise and power transmission system noise.

The front water layer represents a road pit generated by the road due to daily weathering erosion, heavy-duty vehicle pressure loss, road construction and the like, and the road water layer generated by the reasons of larger precipitation, small road gradient, unsmooth drainage and the like in rainfall weather has the properties of rainwater ponding, slurry and the like.

The front water layer splash information includes splash liquid color, splash liquid reflectivity, splash liquid refractive index, liquid splash trajectory, liquid splash direction, liquid splash time, and splash liquid landing sound.

Wherein the six degrees of freedom motion state of the vehicle body comprises six degrees of motion of the rigid body of the vehicle body: linear velocity, angular velocity, linear acceleration, angular acceleration, position, and attitude angle.

(2) And the front vehicle actual response calculation module is used for superposing the six-degree-of-freedom motion state Y_ ego of the vehicle body relative to the horizontal ground and the six-degree-of-freedom motion state Y_relative of the front vehicle relative to the vehicle body according to the principle of rigid relative motion synthesis by taking the front vehicle as a rigid body based on the vehicle state information and the front vehicle state information output by the perception module, so as to obtain the six-degree-of-freedom motion state Y_real of the vehicle body of the front vehicle relative to the horizontal road surface under the current road condition at the moment. Where Y_real is a multidimensional vector that varies over time.

(3) The front vehicle parameter and input estimation module is responsible for estimating front vehicle parameters and front vehicle driver manipulation inputs in the past time period according to the front vehicle appearance and the six-degree-of-freedom motion state Y_real. The module comprises a front vehicle parameter estimation sub-module and a front vehicle steering interference input estimation sub-module.

And (3.1) a front vehicle parameter estimation submodule, which is used for determining the type of the front vehicle based on the vehicle appearance parameter, license plate information, tire rolling marks and tire road noise in the front vehicle state information output by the perception module and combining a vehicle type database, so as to estimate the front vehicle parameter gamma, wherein the front vehicle parameter gamma comprises the front vehicle mass m, the wheel track d_w, the wheel width b, the yaw moment of inertia I_r, the air resistance coefficient Cd and the like. The vehicle types include sedans, sport utility vehicles SUVs, pickup trucks, medium-sized trucks, and heavy goods vehicles.

And (3.2) a front vehicle operation interference input sub-module for estimating front vehicle input I through a vehicle dynamics model based on the six-degree-of-freedom motion state Y_real of the front vehicle relative to the horizontal road surface, which is output by the front vehicle actual response calculation module, and the front vehicle parameter gamma which is output by the front vehicle parameter estimation sub-module, wherein the front vehicle input I comprises operation input and interference input. The vehicle dynamics model comprises a six-degree-of-freedom dynamics model, a longitudinal dynamics model, a transverse dynamics model and a vertical vibration model (comprising a single-degree-of-freedom vertical vibration model, a two-degree-of-freedom vertical vibration model and a vertical-pitching-rolling vibration model).

The estimation strategy in the front vehicle steering interference input sub-module comprises longitudinal-transverse-vertical joint estimation and longitudinal-transverse-vertical decoupling estimation.

The combined estimation of the longitudinal direction, the transverse direction and the vertical direction is specifically as follows: based on the six-degree-of-freedom dynamics model of the vehicle, driving, braking and steering control inputs and disturbance inputs of the front vehicle are estimated according to the six-degree-of-freedom motion state Y_real of the front vehicle relative to the horizontal road surface and the road gradient in the road traffic information.

The vertical-horizontal-vertical decoupling estimation is specifically:

based on the longitudinal dynamics state in y_real and the longitudinal road gradient in the road traffic information, the longitudinal dynamics model of the vehicle is used to estimate the brake steering input_brk, the drive steering input_drv, and the longitudinal component Fxe of the disturbance input of the preceding vehicle.

Based on the lateral dynamics state in y_real and the lateral road gradient in the road traffic information, the steering input input_str, the lateral component Fye of the disturbance input, and the yaw moment Mze of the disturbance input of the preceding vehicle are estimated using the vehicle lateral dynamics model.

Next, subtracting from y_real, the six degrees of freedom motion response of the vehicle body y_horizontal caused by input_brk, input_drv, input_ str, fxe, fye, and Mze; the difference y_real-y_horizontal is used as an input of the vehicle vertical vibration model, and the vertical component Fze, the roll component moment Mxe and the pitch component moment Mye of the disturbance input of the preceding vehicle are estimated.

Finally, estimated front vehicle steering inputs, including brake steering input input_brk, drive steering input input_drv, steering input input_str; estimated front car disturbance inputs include a longitudinal component Fxe, a lateral component Fye, a vertical component Fze, a yaw moment Mze, a roll moment Mxe, and a pitch moment Mye of the disturbance input.

(4) And the water layer splash characteristic parameter identification module is responsible for calculating and deducing key motion parameters of the front water layer. Based on the liquid splashing track, the liquid splashing direction and the liquid splashing time output by the sensing module, deriving based on image recognition and a motion principle, and recognizing a characteristic parameter alpha of the front water layer splashing motion. The method comprises the following steps:

firstly, outputting a liquid splashing direction and a liquid splashing track generated when a front vehicle drives through a water layer in a past time period by using a vehicle-mounted high-definition camera, and obtaining a transverse distance L_side, a maximum vertical height H_side and a longitudinal distance X_side of front water layer side wave splashing by using a calibrated computer vision recognition algorithm, wherein the maximum vertical height H_pick up and the longitudinal distance X_pick up of tread splashing are obtained.

Next, according to the liquid splashing time t output by the sensing module, and based on a physical motion law, deducing and calculating to obtain a side wave speed V_side, a projection and an x-axis clamping angle theta 1 of a side wave speed vector in an xz plane, a projection and a z-axis clamping angle theta 2 in a yz plane, an included angle theta 3 between the side wave speed vector and a y-axis in the xy plane, a tread splashing speed V_jackup and a tread splashing speed vector and a horizontal ground included angle theta 4.

And combining the parameters to obtain a water layer splashing characteristic alpha, wherein the water layer splashing characteristic alpha comprises L_side, H_side, X_side, V_side, V_pickup, theta 1, theta 2, theta 3 and theta 4.

The physical motion law is an oblique throwing motion law, and the initial motion state of object oblique throwing can be obtained by deduction according to the time of the object falling to the horizontal ground, the transverse distance and the maximum longitudinal height.

(5) The water layer medium attribute distinguishing module is responsible for distinguishing the specific attribute of the front water layer, including the liquid category and the corresponding concentration. Based on the splash liquid color, the splash liquid reflectivity, the splash liquid refractive index and the splash liquid landing sound output by the sensing module, the liquid type of the water layer encountered by the front vehicle is identified by utilizing a liquid distinguishing method, and then the inherent attribute beta of the water layer encountered liquid is obtained, wherein the inherent attribute comprises the type and the corresponding concentration of the liquid.

The liquid discriminating method in the water layer medium attribute discriminating module comprises a table look-up method based on an off-line database and a discriminating method based on machine learning.

The table look-up method based on the off-line database specifically comprises the following steps: the method comprises the steps of artificially preparing common road liquids with different types and concentrations, and recording basic parameters of the prepared liquids in a splashing state by using precision equipment, wherein the basic parameters comprise splashing liquid colors LQ_color, splashing liquid reflectivity LQ_reflection, splashing liquid refractive index LQ_reflection and splashing liquid landing sound LQ_sound, and a liquid database with higher sensitivity is formed, wherein the types and the concentrations of the different liquids correspond to the splashing liquid colors LQ_color, the splashing liquid reflectivity LQ_reflection, the splashing liquid refractive index LQ_reflection and the splashing liquid landing sound LQ_sound. And then, according to the splashing liquid color LQ_color, the splashing liquid reflectivity LQ_reflection, the splashing liquid refractive index LQ_reflection and the splashing liquid landing sound LQ_sound output by the sensing module, searching the attribute beta of the splashing liquid in front of the current database by comparing with the established liquid database, wherein the attribute beta comprises the category and the corresponding concentration of the liquid.

The machine learning based discrimination method specifically comprises the following steps: and outputting the attribute beta of the front splash liquid, including the type and the corresponding concentration of the liquid, which is obtained by the sensing module in the past time period, as the input of the machine learning judgment model. The training of the machine learning discrimination model is to obtain colors, reflectivity, refractive indexes and landing sounds of different types and concentrations of liquid splashing through experiments, and train the machine learning discrimination model based on the corresponding relations of the colors, reflectivity, refractive indexes and landing sounds of different liquid splashing, the types and the concentrations of the liquid.

(6) The front water layer judging and water layer depth predicting module is used for judging the size and the geometric position POS of a front water layer based on the six-degree-of-freedom motion state Y_real of the front vehicle relative to a horizontal road surface, the front vehicle parameters and the front vehicle steering input and interference input output by the front vehicle actual response calculating module, the front water layer splash motion characteristic parameters alpha output by the water layer splash characteristic parameter identifying module and the inherent attribute beta of water layer liquid encountered by the front vehicle output by the water layer medium property judging module, and predicting the water layer depth d by utilizing an image identifying technology. Wherein the dimensions of the front aqueous layer include the longitudinal length L and the transverse width W of the aqueous layer.

The prediction method in the front water layer judgment and water layer depth prediction module comprises a prediction method based on a mechanism model and a prediction method based on deep neural network learning.

The prediction method based on the mechanism model specifically comprises the following steps: the front vehicle six-degree-of-freedom motion state Y_real output by the front vehicle actual response calculation module, the front vehicle parameter gamma and the front vehicle input I output by the front vehicle parameter input estimation module, the front splash liquid attribute beta output by the water layer medium attribute judgment module, the front water layer splash motion characteristic parameter alpha output by the water layer splash characteristic parameter identification module, the front water layer longitudinal length L, the transverse width W and the geometric position POS which are previously judged by the front vehicle actual response calculation module, and the depth d of the front water layer is predicted according to a dynamics model of water layer splash:

i is front vehicle input estimation; s is a water layer depth calculation function when the wheels on one side of the front vehicle drive through the water layer to generate splashing, and D is a water layer depth calculation function when the wheels on two sides of the front vehicle drive through the water layer to generate splashing. The calculation function S, D is obtained by regression fitting of experimental data, specifically, under the combination of different experimental parameters, the splash result of the corresponding water layer is observed and recorded, and the calculation function S, D is obtained based on the splash result.

The prediction method based on deep neural network learning specifically comprises the following steps: the six-degree-of-freedom motion state Y_real of the front vehicle, the vehicle parameter gamma of the front vehicle, the input I of the front vehicle, the attribute beta of front splashing liquid, the characteristic parameter alpha of front water layer splashing motion, the longitudinal length L, the transverse width W and the geometric position POS of the front water layer which are judged by the module previously are used as the input of a depth neural network, and the prediction result of the depth neural network is the depth d of the front water layer in the past time period. Training of deep neural network learning is specifically as follows: different experimental parameter combinations, namely (Y_real, beta, gamma, I, L, W, POS and d), are simulated by using fluid mechanics simulation software, corresponding to the generated characteristic parameter alpha of the front water layer splashing motion, corresponding simulation experimental conditions and results are used as training samples, and the deep neural network model is trained.

(7) The automatic driving decision module of the vehicle:

firstly, based on road traffic information info_traffic output by a sensing module, an automatic driving algorithm of the vehicle calculates, and an initial vehicle decision control strategy structyIi= [ Vx ] without considering a front water layer geometric position POS, a water layer depth d and a water layer attribute beta; vy; yawrate; futurepath; susmode ], including a longitudinal speed Vx, a lateral speed Vy, a yaw rate Yawrate, a future travel track Futurepath, and a suspension control mode Susmode for a period of time in the future of the host vehicle.

Then, based on the host vehicle state information and the road traffic information info_traffic output by the sensing module, the water layer attribute beta output by the water layer medium attribute judging module, the water layer size and position POS and the water layer depth d output by the front water layer judging and water layer depth predicting module, outputting a danger early warning to a driver, and outputting a host vehicle decision control strategy correction quantity delta strategy_scp= [ delta Vx by utilizing a correction strategy; Δvy; Δyawrate; Δfuturepath; Δsusmode ]. The danger early warning is specifically output by providing information of the geometrical position POS, the longitudinal length L, the transverse width W and the water layer depth d of the front water layer for a driver in a voice broadcasting mode. Δstrategy_scp is a correction amount considering the safety, comfort, and politicity of the vehicle.

And finally, correcting the initial vehicle decision control strategy strategy_ini according to the vehicle decision control strategy correction quantity delta strategy_scp, and controlling the future motion state of the vehicle, including the future motion trail, linear velocity, angular velocity and suspension control mode of the vehicle.

The correction strategy in the automatic driving decision module of the vehicle comprises a correction strategy based on rules and a correction strategy based on machine learning.

The rule-based correction strategy specifically comprises: based on an automatic driving algorithm of the vehicle, the longitudinal length L, the transverse width W, the geometric position POS and the water layer depth d of the front water layer output by the front water layer judging and water layer depth predicting module are compared with six-degree-of-freedom motion state Y_ ego of the vehicle body of the vehicle relative to the horizontal ground, the vehicle parameter delta and road traffic information info_traffic, and whether the vehicle can drive through the front water layer or not is judged.

(a) L, W, POS, d, Y _ ego, δ and info_ traffic, strategy _ini are input into a safety-based correction function fcorrect_safe, and the safety correction amount Δstrategy_safe is solved:

Δstrategy_safe＝fcorrect_safe(L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini)

the construction rule of the fcorrect_safe function is as follows: on the premise of adhering to the road traffic rules, whether the aim of avoiding directly driving through a water layer can be achieved is judged according to the automatic driving algorithm of the vehicle. If the automatic driving algorithm can be realized, delta strategy-safe is calculated according to the automatic driving algorithm of the vehicle, wherein the delta strategy-safe comprises correction amounts delta Vx-safe, delta Vy-safe, delta Yawrate-safe, delta futurePath-safe and delta SusMode-safe, and the vehicle is prevented from directly driving through a water layer. If the correction cannot be realized, under the premise of ensuring the safety of the vehicle when driving through the water layer, calculating delta strategy_safe including correction amounts delta Vx_safe, delta Vy_safe, delta Yawrate_safe, delta futurePath_safe and delta SusMode_safe; specifically, the calculation rule is that the longer the longitudinal length L, the wider the transverse width W, and the deeper the depth d of the water layer, the greater the potential danger of the water layer in front, the greater and negative the correction amounts Δvx_safe, Δvy_safe, Δyawrate_safe, Δfuturepath_safe, Δsusmode_safe of the host vehicle. The safety correction amount Δstrategy_safe calculated in this step is the amount of the region.

(b) Inputting the Δstrategy_safe calculated in the step (a), L, W, POS, d, Y _ ego, delta and info_ traffic, strategy _ini into a comfort-based correction function fcorrect_scomfort to obtain a comfort correction amount Δstrategy_scomfort:

Δstrategy_scomfort＝

fcorrect_scomfort(L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_safe)

the construction rule of the fcorrect_scomfort function is as follows: on the premise that deltastrategy_safe ensures the safety of the vehicle, deltastrategy_scofort, which comprises deltavx_scofort, deltavy_scofort, deltayawrate_scofort, deltafuturepath_scofort and deltasusmode_scofort, is calculated with the riding comfort of the vehicle passenger as a target; the calculation rule is specifically as follows:

if the Δstrategy_safe calculated in the step (a) cannot avoid directly driving through the water layer, the deeper the front water layer depth d is, which indicates that the road is uneven, the smaller the change rate of the correction amounts Δvx_scomfort, Δvy_scomfort, Δyawrate_scomfort, Δfuturepath_scomfort, Δsusmode_scomfort along with time is, so as to improve the comfort of the vehicle;

if Δstrategy_safe calculated in step (a) can avoid directly driving through the front water layer, the rate of change of correction amounts Δvx_scomfort, Δvy_scomfort, Δyawrate_scomfort, Δfuturepath_scomfort, Δsusmode_scomfort over time is made as small as possible.

The comfort correction amount Δstrategy_scomfort calculated in this step is a smaller amount of cells obtained based on the safety correction amount Δstrategy_safe of step (a), that is, the ranges of the variables Δstrategy_scomfort are all within or equal to the ranges of the corresponding variables in Δstrategy_safe.

(c) Polite correction amount deltastrategy_scitate is calculated.

(c1) If step (a) cannot avoid directly driving through the water layer by Δstrategy_safe, then Δstrategy_ scomfort, L, W, POS, d, Y _ ego, δ, info_ traffic, strategy _ini are input into polite-based correction function fcorrect_scplite to obtain polite correction Δstrategy_scplite:

Δstrategy_scpolite＝

fcorrect_scpolite(L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scomfort)

the construction rule of the fcorrect_scpolite function is as follows: on the premise that deltastrategy_scomfort ensures the safety and the comfort of the vehicle driving through the water layer, deltastrategy_scipline, including deltavx_scipline, deltavy_scipline, deltayawrate_scipline, deltafuturepath_scipline and deltasusmode_scipline, is calculated with the goal of polite driving. The calculation rule is specifically as follows: when the longitudinal length L of the water layer is longer, the transverse width W is wider, the depth d is deeper, the longitudinal speed Vx and the transverse speed Vy of the vehicle in a six-degree-of-freedom motion state of the vehicle body relative to the horizontal ground are larger, which indicates that the negative influence on surrounding traffic vehicles, non-motor vehicles and pedestrians caused by liquid splashing generated when the vehicle passes through the front water layer is larger, the numerical values of correction amounts DeltaVx_scitate, deltaVy_scitate, deltaYawrate_scitate, deltaFuturePath_scitate are larger and negative, and DeltaSusMode_scinate is equal to DeltaSusMode_scifor improving the driving polishness of the vehicle.

The polite modifier Δstrategy_scitate calculated in this step is the minimum amount of cells obtained based on the modifier Δstrategy_sciabout in step (b), i.e., the range of each variable of Δstrategy_scitate is within or equal to the range of the corresponding variable in Δstrategy_sciabout.

(c2) If step (a) can avoid directly driving through the water layer by Δstrategy_safe, then step (b) calculates the comfort correction amount Δstrategy_scomfort without taking the polite correction amount into consideration, i.e., Δstrategy_scinit=Δstrategy_scomfort.

(d) And finally outputting the correction quantity of the decision control strategy of the vehicle, and selecting the optimal value of each correction quantity from the delta strategy_scitate interval quantity during execution. The dimensions Δvx_scp, Δvy_scp, Δyawrate_scp, Δfuturepath_scp, Δsusmode_scp of the optimal correction Δstrategy_scp are obtained by maximizing the objective function J:

J＝w1*g ₁ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)

+w2*g ₂ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)

+w3*g ₃ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)

wherein w1, w2 and w3 respectively represent weights of safety, comfort and politicity, and a function g ₁ 、g ₂ 、g ₃ Respectively calculating decision control of own vehicleThe constraint condition of each variable of the deltastrategy_scp is a section corresponding to each dimension of deltastrategy_scplite.

(e) Only the comfort correction amount Δstrategy_comfort 'is calculated, and y_ ego, δ, info_traffic, strategy_ini is input to the comfort correction function fcorrect_comfort':

Δstrategy_comfort’＝fcorrect_comfort’(Y_ego,δ,info_traffic,strategy_ini)

the construction rule of the fcorrect_compact' function is: the correction amounts Δstrategy_comfort' are calculated for the driving comfort of the host vehicle, including Δvx_comfort, Δvy_comfort, Δyawrate_comfort, Δfuturepath_comfort, and Δsusmode_comfort. The more uneven the road on the path planned by the initial decision control strategy strategy_ini of the host vehicle, the larger the host vehicles Vx, vy, Δvx_compact, Δvy_compact, Δyawrate_compact, Δfuturepath_compact, and Δsusmode_compact are, the smaller the rate of change over time.

The safety correction amount Δstrategy_comfort' calculated in this step is a block amount for each variable.

(f) And finally outputting the correction quantity of the decision control strategy of the vehicle, and selecting the optimal value delta strategy_scp of each correction quantity from the interval quantity delta strategy_comfort' when the decision control strategy of the vehicle is executed.

The correction strategy based on machine learning is specifically as follows: the front water layer longitudinal length L, the transverse width W, the geometric position POS and the water layer depth d which are obtained in the past time period, the six-degree-of-freedom motion state Y_ ego of the vehicle body, the vehicle parameter delta and the road traffic information info_traffic are input into a machine learning model fcorrect, and the vehicle decision control strategy correction quantity delta strategy_scp=fcorrect (L, W, POS, d, Y_ ego, delta, info_traffic) is output; the correction amount is an optimal value comprehensively considering the safety, comfort and politicity of the automatic driving of the vehicle.

The training process of the machine learning model fcorrect is as follows: different known conditions including the longitudinal length L of a front water layer, the transverse width W, the geometric position POS and the water layer depth d, the six-degree-of-freedom motion state Y_ ego of the body of the vehicle, the vehicle parameter delta and the road traffic information info_traffic are manually set, and the optimal vehicle decision control strategy correction quantity comprehensively considering the driving safety, the comfort and the politeness under different known conditions is calculated offline by utilizing the rule-based correction strategy. And then training a machine learning model according to the corresponding relation between different known conditions and the calculated optimal vehicle decision control strategy correction amount, and finally obtaining a model fcorrect.

Examples

The application scenario 1 of the embodiment is shown in fig. 2, wherein the right turn water-encountering layer of the front vehicle at the traffic light crossroad in the rainy day is shown, and the straight-going signal lamp is a red light, so that the vehicles at the right turn lane are allowed to turn right, and the non-motor vehicles and pedestrians wait for the traffic light at the crossroad. The sensing module provides traffic road condition information, covers the past time period tau, adopts the step length ts, and the time tau=k ts, namely all sensing data in the current time T0 and k sampling step lengths can be used for implementation of the system.

The perception module has obtained road gradient, road curvature, lane boundary line, lane width, traffic light status, pedestrian information, traffic flow information, etc. in the road traffic information over the past period of time τ.

Based on the obtained state information of the vehicle and the state information of the front vehicle, the front vehicle is regarded as a rigid body, and the six-degree-of-freedom motion state Y_ ego of the vehicle body of the vehicle relative to the horizontal ground and the six-degree-of-freedom motion state Y_relative of the vehicle body of the front vehicle relative to the vehicle are overlapped according to the principle of rigid body relative motion synthesis, so that the six-degree-of-freedom motion state Y_real of the vehicle body of the front vehicle relative to the horizontal ground is obtained.

The front vehicle parameters and input estimation module firstly determines the type of the front vehicle by searching a vehicle type database according to the shape of the front vehicle, license plate information and tire rolling marks, and estimates front vehicle parameters gamma; in this example, parameters required for extracting the corresponding model according to the vehicle type database include 1600kg in mass, 2300kg·m2 in yaw moment of inertia, 0.34 in air resistance coefficient, 1785mm in track width d_w, 195mm in wheel width b, and the like.

The next step adopts a longitudinal-transverse-vertical decoupling estimation strategy for the input estimation strategy of the front vehicle, in the embodiment, the longitudinal speed of the front vehicle is reduced, the course angle of the vehicle deflects rightwards, and firstly, the brake operation input_brk of the front vehicle is estimated through a longitudinal dynamics model of the vehicle; estimating steering control input_str of the front vehicle by using the transverse dynamics model of the vehicle; the six-degree-of-freedom motion response Y_horizontal of the vehicle body, which is caused by input_brk and input_str, is removed from the actual six-degree-of-freedom motion state Y_real of the front vehicle, that is, the difference Y_real-Y_horizontal is obtained, and finally, the vertical component Fze, the roll component moment Mxe and the pitch component moment Mye of the disturbance input are estimated according to the vertical vibration model of the vehicle. And finally comprehensively estimating to obtain a front vehicle input I, wherein the front vehicle input I comprises a manipulation input and an interference input.

Information such as liquid splashing direction and track is obtained according to an image generated when a front vehicle obtained by the vehicle-mounted high-definition camera passes through the water layer, and in the embodiment, a front vehicle right single-side wheel passes through the water layer, and a rear view and a side view are shown in fig. 3 and 4. Estimating a lateral wave splashing transverse distance L_side, a maximum vertical height H_side and a longitudinal distance X_side of a front truck when the front truck passes through a water layer by using a calibrated computer image recognition algorithm, wherein the maximum vertical height H_pickup and the longitudinal distance X_pickup of the tire surface splashing are calculated in a real-time reasoning mode according to the splashing time t of corresponding liquid to obtain a lateral wave velocity V_side, the projection of a lateral wave velocity vector in an xz plane and an X-axis clamping angle are theta 1, the projection of the lateral wave velocity vector in a yz plane and a z-axis included angle are theta 2, and the y-axis included angle in the xy plane and the y-axis included angle is theta 3; the tread splash speed magnitude V_pickup, the tread splash speed vector, and the horizontal ground angle θ4, are shown in FIG. 5. And combining the parameters to obtain the splashing characteristic alpha of the water layer.

The front water layer type and the corresponding concentration are judged by the water layer medium attribute judging module by utilizing a judging method based on machine learning while the front water layer splashing characteristic alpha is identified. The color, reflectivity, refractive index and floor sound of the liquid splashing obtained by the sensing module in the past time period are used as inputs of a machine learning judgment model, and the inputs are output as the attribute beta of the liquid splashing in front, wherein the attribute beta comprises the type and the corresponding concentration of the liquid. In this embodiment, the front water layer is ordinary road rainfall ponding, as heavy vehicles often turn right at the intersection causing road pit breakage.

The next step of front water layer judgment and water layer depth prediction module firstly utilizes an image obtained by a vehicle-mounted high-definition camera to estimate the geometric dimension and the position of a front water layer, wherein in the embodiment, the water layer is a water layer with the length of L=1m, the width of W=1m and POS (point of sale) which is specifically positioned on the right wheel of a front vehicle and is 5m away from the vehicle. In the embodiment, a prediction method based on a mechanism model is adopted in a front water layer depth estimation strategy, and a depth prediction function is input into a real six-degree-of-freedom motion state Y_real of a front vehicle, a water layer splash motion parameter alpha, a water layer inherent attribute beta, a front vehicle parameter gamma, a front vehicle input I, a water layer length and width L, W and a geometric position POS:

in this embodiment, since the front vehicle is driven through the water layer on the right side, the depth prediction function selects S, and the front water layer depth d is calculated.

Finally, the automatic driving decision module of the host vehicle calculates an initial host vehicle decision control strategy strategy_ini which does not consider the geometric position of a front water layer, the depth d of the water layer and the attribute beta of the water layer based on the road traffic information info_traffic output by the sensing module, and in the embodiment, the host vehicle automatically driving algorithm turns right to pass through a traffic light intersection. However, considering that the front vehicle passes through to generate splash, the fact that an unknown depth water layer exists in front is explained, so that the control strategy of the vehicle needs to be modified, and in the embodiment, the modification strategy adopts a modification strategy based on rules. Firstly, according to an initial decision control strategy strategy_ini of the host vehicle, whether the host vehicle drives through a water layer is judged, and in the embodiment, the host vehicle is also in a right turn lane, so that the host vehicle must drive through a front water layer.

(a) The information L, W, POS, d, Y _ ego, delta, info_ traffic, strategy _ini is first input to the safety-based correction function fcorrect_safe, and the safety correction Δstrategy_safe is solved for, in this embodiment, because the water layer is located in the intersection solid lane, Δstrategy_safe cannot prevent the host vehicle from driving directly over the front water layer.

Δstrategy_safe＝fcorrect_safe(L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini)

(b) Inputting the Δstrategy_safe calculated in (a), L, W, POS, d, Y _ ego, delta, info_ traffic, strategy _ini, and the comfort-based correction function fcorrect_scomfort to obtain a comfort correction amount Δstrategy_scomfort:

Δstrategy_scomfort＝

fcorrect_scomfort(L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_safe)

(c) Inputting the Δstrategy_scomfort calculated in (b), L, W, POS, d, Y _ ego, δ, info_ traffic, strategy _ini into a polite-based correction function fcorrect_scite to obtain a polite correction amount Δstrategy_scite:

Δstrategy_scpolite＝

(d) And selecting the optimal value delta strategy_scp of each correction quantity from the interval quantity delta strategy_scplite, and finally outputting the vehicle decision control strategy correction quantity, wherein the correction quantity comprehensively considers the safety, the comfort and the politicity of the vehicle.

Fig. 6 and 7 show two other application scenarios 2 and 3 of the embodiment, wherein wheels on two sides of a straight running front vehicle in fig. 6 simultaneously drive through a water layer with unknown depth, and water layer splashing tracks, splashing liquid landing sounds and the like are generated on two sides of the front vehicle simultaneously; the right wheels of the vehicle in front of the left lane in fig. 7 drive through a wide position and depth water layer, and generate water layer splash track, splash liquid landing sound and the like to the right. For the two application scenes 2 and 3, the invention can also obtain the correction quantity deltastrategy_scp for comprehensively considering the decision and control of safety, comfort and politicity of the vehicle, including the correction of deceleration, steering and the like from all the perception data in the current time T0 and k sampling step sizes according to the processes of actual response calculation of the front vehicle, front vehicle parameter and input estimation, water layer splash characteristic parameter identification, water layer medium attribute discrimination, water layer position discrimination, depth prediction and the like shown in figure 1, thereby realizing the safe driving of the vehicle and reducing the influence of water layer splash on other traffic participants to the greatest extent.

The embodiments of the present invention are not limited to any particular model, but only the preferred embodiments of the present invention are described above, and are not intended to limit the present invention, and various modifications and variations can be made to the embodiments of the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The driving decision system based on the road water layer depth estimation is characterized by comprising a sensing module, a front vehicle actual response calculation module, a front vehicle parameter and input estimation module, a water layer medium attribute discrimination module, a water layer splash characteristic parameter identification module, a front water layer judgment and water layer depth prediction module and a self-driving decision module;

the sensing module is used for acquiring road traffic information, host vehicle state information, front vehicle state information and front water layer splashing information; the road traffic information comprises road adhesion coefficient, lane width, lane number, road gradient, lane curvature, lane marking, obstacles, traffic signs, traffic light states, pedestrian information, traffic flow information and weather states; the vehicle state information comprises a vehicle parameter delta, a vehicle body six-degree-of-freedom motion state Y_ ego of the vehicle relative to the ground and operation input; the front vehicle state information comprises a six-degree-of-freedom motion state Y_relative of a front vehicle relative to a vehicle body of the vehicle, vehicle appearance parameters, license plate information, brake indicator lamp states, steering indicator lamp states, tire rolling marks, horn warning sounds, tire road noise and power transmission system noise; the front water layer splashing information comprises splashing liquid color, splashing liquid reflectivity, splashing liquid refractive index, liquid splashing track, liquid splashing direction, liquid splashing time and splashing liquid landing sound; the six-degree-of-freedom dynamic state of the vehicle body comprises linear speeds, angular speeds, linear accelerations, angular accelerations, positions and attitude angles of six degrees of freedom of motion of the rigid body of the vehicle body; the control inputs comprise steering, braking, driving, indicator lights and warning horn operations;

The front vehicle actual response calculation module is used for superposing a six-degree-of-freedom motion state Y_ ego of the vehicle body relative to the ground and a six-degree-of-freedom motion state Y_relative of the front vehicle relative to the vehicle body based on the vehicle state information and the front vehicle state information output by the sensing module according to the vehicle relative motion synthesis principle, so as to obtain a six-degree-of-freedom motion state Y_real of the front vehicle relative to the road surface under the current road condition;

the front vehicle parameter and input estimation module comprises a front vehicle parameter estimation sub-module and a front vehicle operation interference input estimation sub-module; the front vehicle parameter estimation sub-module is used for estimating a front vehicle parameter gamma based on vehicle appearance parameters, license plate information, tire rolling marks and tire road noise in the front vehicle state information output by the sensing module and determining a front vehicle type by combining a vehicle type database; the front vehicle operation interference input sub-module is used for estimating front vehicle input I through a vehicle dynamics model based on the six-degree-of-freedom motion state Y_real of the front vehicle relative to the vehicle body of the road surface, which is output by the front vehicle actual response calculation module, and the front vehicle parameter gamma which is output by the front vehicle parameter estimation sub-module, wherein the front vehicle input I comprises operation input and interference input;

The water layer splashing characteristic parameter identification module is used for identifying characteristic parameters alpha of front water layer splashing motion based on the liquid splashing track, the liquid splashing direction and the liquid splashing time output by the sensing module;

the water layer medium attribute judging module is used for identifying the type of liquid in the water layer encountered by the front vehicle by utilizing a liquid judging method based on the splash liquid color, the splash liquid reflectivity, the splash liquid refractive index and the splash liquid landing sound output by the sensing module, so as to obtain the inherent attribute beta of the liquid in the water layer;

the front water layer judging and water layer depth predicting module is used for judging the size and the position POS of a front water layer by utilizing an image recognition technology and predicting the water layer depth d by utilizing a depth predicting method based on the six-degree-of-freedom motion state Y_real of the front vehicle relative to the vehicle body of the road surface, the front vehicle parameters and the front vehicle steering input and interference input output by the front vehicle actual response calculating module, the front water layer splashing motion characteristic parameter alpha output by the water layer splashing characteristic parameter recognizing module and the inherent attribute beta of water layer liquid encountered by the front vehicle output by the water layer medium property judging module; wherein the size of the front water layer comprises the length L and the width W of the water layer;

The automatic driving decision module of the host vehicle firstly calculates an initial host vehicle decision control strategy strategy_ini= [ Vx; vy; yawrate; futurepath; susmode ] which does not consider the geometric size and position of a front water layer, the depth d of the water layer and the attribute beta of the water layer by adopting an automatic driving algorithm of the host vehicle based on road traffic information info_traffic output by the perception module, wherein the decision control strategy comprises longitudinal speed Vx, transverse speed Vy, yaw rate Yawrate, future driving track Futurepath and suspension control mode Susmode in future of the host vehicle; next, outputting a vehicle decision control strategy correction amount deltastrategy_scp= [ deltavx_scp; deltavy_scp; deltayawrate_scp; deltafuturepath_scp; deltasusmode_scp ] by utilizing a correction strategy while outputting a danger level early warning to a driver based on the vehicle state information and road traffic information info_traffic output by the perception module, the water layer attribute beta output by the water layer medium attribute discrimination module, the water layer size and the geometric position POS and the water layer depth d output by the front water layer judgment and water layer depth prediction module; and finally, correcting an initial vehicle decision control strategy strategy_ini according to the correction quantity deltastrategy_scp, and controlling the future motion state of the vehicle, including future motion trail, linear speed, linear acceleration, angular speed, angular acceleration and suspension control mode.

2. The system for driving decision based on road water depth estimation according to claim 1, wherein the estimation strategy in the front car steering disturbance input sub-module comprises a combined longitudinal-lateral-vertical estimation or a decoupled longitudinal-lateral-vertical estimation;

the longitudinal-transverse-vertical joint estimation is specifically as follows: estimating driving, braking and steering control input and disturbance input of the front vehicle according to the six-degree-of-freedom motion state Y_real of the front vehicle relative to the vehicle body of the road surface and the road gradient in road traffic information based on the six-degree-of-freedom dynamics model of the vehicle;

the decoupling estimation of the longitudinal-transverse-vertical directions is specifically as follows: estimating a brake control input input_brk and a drive control input input_drv of a front vehicle by utilizing a vehicle longitudinal dynamics model according to the longitudinal dynamics state in Y_real and the longitudinal road gradient in road traffic information, and interfering the longitudinal component Fxe of the input; estimating a steering operation input_str, a lateral component force Fye of disturbance input and a yaw moment Mze of disturbance input of a front vehicle by utilizing a vehicle transverse dynamics model according to the transverse dynamics state in Y_real and the transverse road gradient in road traffic information; next, subtracting the six-degree-of-freedom motion response y_horizontal of the vehicle body caused by input_brk, input_drv, input_ str, fxe, fye and Mze from y_real, and estimating a vertical component Fze of the disturbance input, a roll component moment Mxe of the disturbance input and a pitch component moment Mye of the disturbance input by taking the difference y_real-y_horizontal as an input of the vehicle vertical vibration model; the final estimated front vehicle steering inputs include a brake steering input input_brk, a drive steering input_drv, and a steering input input_str; the estimated front truck disturbance inputs include the disturbance input longitudinal component Fxe, lateral component Fye, vertical component Fze, yaw moment Mze, roll moment Mxe, and pitch moment Mye.

3. The system for driving decision based on road water layer depth estimation according to claim 1, wherein the recognition method in the water layer splash characteristic parameter recognition module specifically comprises:

firstly, according to the liquid splashing direction and the liquid splashing track output by a perception module in the past time period, a computer vision recognition algorithm is utilized to obtain the transverse distance L_side, the maximum vertical height H_side and the longitudinal distance X_side of the side wave splashing of a front water layer, and the maximum vertical height H_pickup and the longitudinal distance X_pickup of tread splashing are obtained; next, according to the liquid splashing time output by the sensing module and based on a physical motion law, deducing and calculating to obtain a side wave velocity V_side, wherein the angle between the projection of a side wave velocity vector in an xz plane and an x-axis is theta 1, the angle between the projection of the side wave velocity vector in a yz plane and a z-axis is theta 2, the angle between the projection of the side wave velocity vector in the yz plane and a y-axis is theta 3, the tread splashing velocity V_pick up and the tread splashing velocity vector form an angle theta 4 with the ground; the water layer splash characteristic attribute alpha is obtained and comprises L_side, H_side, X_side, V_side, V_jackup, theta 1, theta 2, theta 3 and theta 4.

4. The system for driving decision based on road water layer depth estimation according to claim 1, wherein the liquid discrimination method in the water layer medium attribute discrimination module comprises a table look-up method based on an offline database or a discrimination method based on machine learning;

The table look-up method based on the off-line database specifically comprises the following steps: preparing road liquids with different types and concentrations, recording basic parameters of the prepared liquids in a splashing state, wherein the basic parameters comprise splashing liquid color LQ_color, splashing liquid reflectivity LQ_reflection, splashing liquid refractive index LQ_reflection and splashing liquid landing sound LQ_sound, and forming a liquid database; next, according to the splashing liquid color LQ_color, the splashing liquid reflectivity LQ_reflection, the splashing liquid refractive index LQ_reflection and the splashing liquid landing sound LQ_sound output by the sensing module, searching the attribute beta of the splashing liquid in front of the current according to the established liquid database, wherein the attribute beta comprises the category and the corresponding concentration of the liquid;

5. The system for driving decision based on road water layer depth estimation according to claim 1, wherein the prediction method in the front water layer judgment and water layer depth prediction module comprises a prediction method based on a mechanism model or a prediction method based on deep neural network learning;

s is a water layer depth calculation function when the wheels on one side of the front vehicle drive through the water layer to generate splashing, and D is a water layer depth calculation function when the wheels on two sides of the front vehicle drive through the water layer to generate splashing;

6. The system for driving decision based on road water layer depth estimation according to claim 1, wherein the correction strategy in the host vehicle automatic driving decision module comprises a rule-based correction strategy or a machine-learning-based correction strategy;

the rule-based correction strategy specifically comprises: firstly, based on an automatic driving algorithm of the vehicle, judging whether the initial decision control strategy strategy_ini of the vehicle can cause the vehicle to drive through a front water layer or not by comparing the six-degree-of-freedom motion state Y_ ego of the vehicle relative to the vehicle body on the ground, the vehicle parameter delta and the road traffic information info_traffic of the vehicle, and then comprehensively considering driving safety, comfort and politics, and calculating to obtain the decision control strategy correction delta strategy_scp of the vehicle;

the correction strategy based on machine learning is specifically as follows: by setting different known conditions including the length L and the width W of a front water layer, the geometrical position POS and the depth d of the water layer, the six-degree-of-freedom motion state Y_ ego of a vehicle body of the vehicle, the vehicle parameter delta and the road traffic information info_traffic, and utilizing the rule-based correction strategy, the optimal vehicle decision control strategy correction amount comprehensively considering the driving safety, the comfort and the politicity under different known conditions is calculated offline; training a machine learning model according to the corresponding relation between different known conditions and the calculated optimal host vehicle decision control strategy correction amount, and finally obtaining a model fcorrect; l, W, POS, d, Y _ ego, δ, info_traffic in the past period are input to the machine learning model fcorrect, and the correction amount Δstrategy_scp of the driving decision control of the host vehicle is output.

7. The system for driving decision based on road water layer depth estimation of claim 6, wherein the rule-based correction strategy comprises:

(A1) L, W, POS, d, Y _ ego, delta and info_ traffic, strategy _ini are input into a safety-based correction function fcorrect_safe, and a safety correction quantity delta strategy_safe is solved; each variable of deltastrategy_safe is interval quantity;

(A2) Inputting the deltastrategy_safe calculated in the step (A1), L, W, POS, d, Y _ ego, delta and info_ traffic, strategy _ini into a comfort-based correction function fcorrect_scomfort to obtain a comfort correction quantity deltastrategy_scomfort; the range of each variable Δstrategy_scomfort is within or equal to the range of Δstrategy_safe;

(A3) Calculating polite correction amount deltastrategy_scitate;

(A31) If step (A1) cannot avoid directly driving through the water layer by Δstrategy_safe, inputting Δstrategy_ scomfort, L, W, POS, d, Y _ ego, δ, info_ traffic, strategy _ini into polite-based correction function fcorrect_scplite to obtain polite correction amount Δstrategy_scplite; the range of the variables Δstrategy_scitate is within or equal to the range of Δstrategy_scinfort;

(A32) If the step (A1) can avoid directly driving through the water layer by Δstrategy_safe, after the step (A2) calculates the comfort correction amount Δstrategy_scomfort, no consideration is required for correction for polite driving, and the range of each variable of Δstrategy_scomfort is equal to Δstrategy_scomfort;

(A4) Outputting the decision control strategy correction quantity of the vehicle: selecting an optimal value of each correction amount from the delta strategy_scite interval amount as a host vehicle decision control strategy correction amount delta strategy_scp; wherein each dimension Δvx_scp, Δvy_scp, Δyawrate_scp, Δfuturepath_scp, Δsusmode_scp of Δstrategy_scp is obtained by maximizing the objective function J:

J=w1*g ₁ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)

+w2*g ₂ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)

+w3*g ₃ (L,W,POS,d,Y_ego,δ,info_traffic,strategy_ini,Δstrategy_scp)

wherein w1, w2 and w3 respectively represent weights of safety, comfort and politicity, and a function g ₁ 、g ₂ 、g ₃ Respectively calculating safety, comfort and polite indexes of the vehicle under the decision control strategy correction quantity delta strategy_scp of the vehicle, wherein the constraint condition of each variable of delta strategy_scp is a section corresponding to each dimension of delta strategy_scplite;

(B1) Only calculating a comfort correction quantity delta strategy_comfort ', and inputting Y_ ego, delta and info_ traffic, strategy _ini into a comfort correction function fcorrect_comfort ' to obtain delta strategy_comfort '; each variable of deltastrategy_compact' is an interval quantity;

8. The system for driving decision based on road water layer depth estimation according to claim 7, wherein in step (A1), the construction rule of fcorrect_safe function is: on the premise of adhering to the road traffic rules, judging whether the aim of avoiding directly driving through a water layer can be fulfilled according to the automatic driving algorithm of the vehicle; if the automatic driving algorithm can be realized, delta strategy-safe is calculated according to the automatic driving algorithm of the vehicle, wherein the delta strategy-safe comprises correction amounts delta Vx-safe, delta Vy-safe, delta Yawrate-safe, delta futurePath-safe and delta SusMode-safe, and the vehicle is prevented from directly driving through a water layer; if the correction cannot be realized, under the premise of ensuring the safety of the vehicle when driving through the water layer, calculating delta strategy_safe including correction amounts delta Vx_safe, delta Vy_safe, delta Yawrate_safe, delta futurePath_safe and delta SusMode_safe; specifically, the calculation rule is that the longer the longitudinal length L, the wider the transverse width W, and the deeper the depth d of the water layer, the greater the potential danger of the water layer in front, the greater and negative the correction amounts Δvx_safe, Δvy_safe, Δyawrate_safe, Δfuturepath_safe, Δsusmode_safe of the host vehicle.

9. The system for driving decision based on road water layer depth estimation of claim 7, wherein:

in step (A2), the construction rule of the fcorrect_scomfort function is: on the premise that deltastrategy_safe ensures the safety of the vehicle, deltastrategy_scofort, which comprises deltavx_scofort, deltavy_scofort, deltayawrate_scofort, deltafuturepath_scofort and deltasusmode_scofort, is calculated with the riding comfort of the vehicle passenger as a target; the calculation rule is specifically as follows:

if Δstrategy_safe calculated in step (A1) can avoid directly driving through the front water layer, making the rate of change of correction amounts Δvx_scomfort, Δvy_scomfort, Δyawrate_scomfort, Δfuturepath_scomfort, Δsusmode_scomfort over time as small as possible;

in step (B1), the construction rule of the fcorrect_compact' function is: calculating correction amounts deltastrategy_comfort' including deltavx_comfort, deltavy_comfort, deltayawrate_comfort, deltafuturepath_comfort, deltasusmode_comfort, targeting the driving comfort of the host vehicle; the more uneven the road on the path planned by the initial decision control strategy strategy_ini of the host vehicle, the larger the host vehicles Vx, vy, Δvx_compact, Δvy_compact, Δyawrate_compact, Δfuturepath_compact, and Δsusmode_compact are, the smaller the rate of change over time.

10. The system for driving decision based on road water layer depth estimation according to claim 7, wherein in step (a 31), the construction rule of fcorrect_scite function is: on the premise that deltastrategy_scomfort ensures the safety and the comfort of the vehicle driving through a water layer, deltastrategy_scipline is calculated with polite driving as a target, and deltavx_scipline, deltavy_scipline, deltayawrate_scipline, deltafuturepath_scipline and deltasusmode_scipline are included; the calculation rule is specifically as follows: when the longitudinal length L of the water layer is longer, the transverse width W is wider, the depth d is deeper, the longitudinal speed Vx and the transverse speed Vy of the vehicle in a six-degree-of-freedom motion state of the vehicle body relative to the horizontal ground are larger, which indicates that the negative influence on surrounding traffic vehicles, non-motor vehicles and pedestrians caused by liquid splashing generated when the vehicle passes through the front water layer is larger, the numerical values of correction amounts DeltaVx_scitate, deltaVy_scitate, deltaYawrate_scitate, deltaFuturePath_scitate are larger and negative, and DeltaSusMode_scinate is equal to DeltaSusMode_scifor improving the driving polishness of the vehicle.