Method for predicting the travel path of a motor vehicle and prediction apparatus

Abstract

A method for predicting the expected travel path of a moving vehicle by numerical integration of a dynamic vehicle model using at least one rotational travel state and at least one longitudinal travel state is disclosed. To provide a prediction of the expected travel path, which does not depend on map information and provides the maximum possible accuracy even for non-steady-state travel states of a vehicle, time-related function rules for the rotational travel state .sub.Pre(t) and/or for the longitudinal travel state .sub.Pre(t) are determined, and values for the travel state concerned .sub.Pre, .sub.Pre are predicted at specific points in time by integration using said function rule .sub.Pre(t), .sub.Pre(t). In this process, the time-related function rule of the travel state concerned is determined by obtaining respective (rotational or longitudinal) input variables for at least two time-derivatives of the travel state concerned from measured values.

Claims

1. A method for controlling a moving vehicle by numerical integration of a dynamic vehicle model using at least one rotational travel state, which is affected by angular motion of the moving vehicle, and at least one longitudinal travel state, which is affected by linear motion of the moving vehicle, the method comprising: generating, using at least one sensor of the moving vehicle, measured values corresponding to input variables of the at least one rotational travel state and the at least one longitudinal travel state; determining time-related function rules for a predicted rotational travel state and/or for a predicted longitudinal travel state; generating prediction values for a travel state concerned at specific points in time by numerical integration using said time-related function rule; and facilitating control of the moving vehicle based on the prediction values, wherein the time-related function rule of a predicted travel state is determined by: obtaining respective rotational or longitudinal input variables for at least two time-derivatives of the travel state concerned from the measured values, and relating the input variables in linear dynamic models of order equal to the number of input variables of the travel state concerned using specified time constants, by determining from the linear model a time-related function rule for prediction values for the input variable concerned, and by analytically integrating the function rule for the prediction values of the input variable.

2. The method according to claim 1, wherein variables obtained using the input variables are derived from the linear dynamic model and taken into account as weighting factors in determining the time-related function rules, wherein the weighting factors are updated whenever a new measured value is acquired.

3. The method according to claim 1, wherein measured values are acquired for each input variable.

4. The method according to claim 1, wherein a speed of the vehicle is used as the longitudinal travel state for the numerical integration, wherein as the input variables for the travel-path prediction, the acceleration and a sudden linear movement are determined from the measured values or estimated from the measured values.

5. The method according to claim 1, wherein a bearing angle of the vehicle is used as the rotational travel state for the numerical integration, wherein as the input variables for the travel-path prediction, yaw rate and yaw acceleration are determined from the measured values or estimated from the measured values.

6. The method according to claim 1, wherein the vehicle specified time constants lie in a time range of 0.3 s to 15.0 s.

7. A prediction apparatus for implementing the method as claimed in claim 1, wherein the prediction apparatus comprises a measured-value input for connecting to a vehicle-sensor unit of the vehicle.

8. The prediction apparatus according to claim 7, wherein the prediction apparatus is allocated to a camera-based or optoelectronic driver assistance system or is integrated in such a driver assistance system.

Description

(1) Exemplary embodiments of the invention are explained in more detail below, where:

(2) FIG. 1 is a wiring diagram of an exemplary embodiment of a prediction apparatus in a vehicle;

(3) FIG. 2 is a flow diagram of an exemplary embodiment of a method for travel-path prediction;

(4) FIG. 3 is a graph of the variation over time of the yaw rate;

(5) FIG. 4 shows a detail for the interval IV-IV in FIG. 3;

(6) FIG. 5 is a graph showing the predicted travel path of a vehicle.

(7) FIG. 1 shows a wiring diagram of a motor vehicle 1 equipped with one or more driver assistance systems. The driver assistance systems comprise a central control unit 2, to which measured information is constantly input from a vehicle-sensor unit 3. The control unit 2 controls those driver assistance systems that rely on the prediction of the expected travel path of the vehicle. For this purpose, the control unit 2 is allocated a prediction apparatus 4, which informs the control unit 2 of its prediction result X.sub.pre(t). In the exemplary embodiment shown, the prediction apparatus 4 is a module of the control unit 2 inside a central electronics unit 5.

(8) The control unit 2 uses the input information to analyse the driving behaviour of the vehicle. If the driving behaviour of the vehicle 1 deviates from a target behaviour, a control intervention is made to the brakes of one or more wheels 6 and/or to the engine management system, thereby affecting the drive and, for instance, changing the torque. The driver assistance system that uses the prediction of the expected travel path of the vehicle 1 is an adaptive cruise control (ACC) system, for example. The ACC system also takes into account information transmitted from the vehicle-sensor unit 3 relating to the drive of the vehicle 1 or information from an electronic stability control system, for instance an ESP control system, which is transmitted via a CAN data bus. The control unit 2 converts the accelerations required by the driver assistance system, in this case by the ACC controller, into drive torques and/or braking torques.

(9) The vehicle-sensor unit 3 uses a yaw-rate sensor 7 on the vehicle 1 to detect rotational quantities, which are influenced by the angular motion of the vehicle 1. The yaw-rate sensor 7 here responds to angular movements in the direction of the arrow 8 about a vertical axis 9 of the vehicle 1. Various measured quantities that provide information about the angular motion of the vehicle about the vertical axis 9, i.e. the yaw motion, can be generated from the measured signal from the yaw-rate sensor 7 by time-related analysis.

(10) The vehicle-sensor unit 3 also captures longitudinal measured quantities, which relate to the linear motion of the vehicle 1. Suitable sensors are provided for this purpose. In the exemplary embodiment, revolution counters 10 are arranged on the wheels 6 as sensors for capturing longitudinal measured quantities. The wheel rotation rate or wheel speed can be used to infer the travel speed and other longitudinal values.

(11) The vehicle-sensor unit 3 also comprises an optoelectronic detection device, specifically a laser scanner 11, which in the exemplary embodiment shown is installed in the front region of the vehicle 1. The laser scanner 11 is used to obtain information about the surroundings of the moving motor vehicle 1, for instance obstacles or moving objects such as other vehicles on the road. Different sensor technology can be used instead of a laser scanner 11 to capture visual information, for example as part of a camera-based driver assistance system. If the prediction apparatus 4 is allocated to such a camera-based or optoelectronic driver assistance system, then the camera-based or optoelectronic driver assistance system can take into account the expected travel path of the vehicle in which it is fitted and also the movement of other vehicles to compute suitable control scenarios and trajectories, and respond by making control interventions if necessary.

(12) A measured-value input 40 of the prediction apparatus 4 is connected to the vehicle-sensor unit 3 and receives longitudinal input variables .sub.0, .Math..sub.0 and rotational input variables .sub.0, {acute over ()}.sub.0, as explained in greater detail below with reference to FIG. 2.

(13) In the exemplary embodiment shown, the vehicle-sensor unit 3 obtains further measured values, which provide information about the driving behaviour of the vehicle 1, specifically, for example, about those actions coming from the driver of the vehicle. Thus a steering-angle sensor 12 and a braking-signal transducer 13 are part of the vehicle-sensor unit 3. A steering-angle signal from the steering-angle sensor 12, or a brake request via the braking-signal transducer 13, is processed in certain driver assistance systems and can also be included in the travel-path prediction to refine the measured values.

(14) In predicting the expected travel path of a moving motor vehicle, discrete position values 15 are determined for a prediction horizon of, for example, 3 to 5 seconds at respective points in time lying in the order in which they occur on the expected travel path 14 (FIG. 5).

(15) The calculation of individual prediction positions 15 on the travel path 14 is explained in greater detail below with reference to the flow diagram in FIG. 2. The discrete position values 15 at specific points in time are the prediction result X.sub.pre(t) of a numerical integration 16 of a dynamic vehicle model. Both the speed v.sub.Pre of the vehicle 1 as the longitudinal travel state and the bearing angle .sub.Pre of the vehicle 1 as the rotational travel state are included in this process as given by the following integral:

(16) $X_{\mathrm{pre}}\left(t\right)={}_0^t{}_{\mathrm{pre}}\left(\right)\left(\begin{array}{c}\cos \left({}_{\mathrm{pre}}\left(\right)\right)\\ \sin \left({}_{\mathrm{pre}}\left(\right)\right)\end{array}\right)d$

(17) This integral yields the following approximation for the prediction result as a sum of the positional values 15:

(18) $X_{\mathrm{pre}}\left(t\right)\underset{=0}{\overset{T\text{/}t}{.Math.}}\left[{}_{\mathrm{pre}}\left(t\right)\left(\begin{array}{c}\cos \left({}_{\mathrm{pre}}\left(t\right)\right)\\ \sin \left({}_{\mathrm{pre}}\left(t\right)\right)\end{array}\right)\right]$

(19) In this equation, t denotes time-discrete integration steps and T denotes the prediction horizon of preferably 3 to 5 seconds.

(20) In order to determine both the rotational travel state, i.e. the bearing angle .sub.Pre of the vehicle 1, and the longitudinal travel state, i.e. the speed v.sub.Pre, a time-related function rule .sub.pre(t) is determined for the bearing angle and a time-related function rule .sub.pre(t) for the speed.

(21) To determine the function rules for the (respective) travel states, initially respective input variables (rotational or longitudinal) .sub.0, {acute over ()}.sub.0, .sub.0, .Math..sub.0 are obtained from measured values for at least two time-derivatives of the travel state concerned, and related in a time-discrete manner by time constants .sub.,1, .sub.,2, .sub.,1, .sub.,2 in the linear model of order equal to the number of input variables of the travel state concerned.

(22) To determine the continuous function rule for the rotational travel state, the instantaneous yaw rate .sub.0 and the yaw acceleration {acute over ()}.sub.0 are found from measured values from the vehicle-sensor unit 3. In all the equations mentioned below, the index 0 denotes an instantaneous measured value for the corresponding variable. In the exemplary embodiment shown, the instantaneous yaw rate .sub.0 is provided as a measured value from the vehicle-sensor unit 3. The instantaneous yaw acceleration {acute over ()}.sub.0 is determined in an estimation stage 41 from the measured values for the yaw rate .sub.0. The yaw rate is here the first time-derivative of the bearing angle of the vehicle, and the yaw acceleration is the second time-derivative of the bearing angle, and, in combining step 17, as input variables for the travel-path prediction, are related in a second-order linear model according to the following equation:

(23) $\left(\begin{array}{c}{}_0\\ {\stackrel{.}{}}_0\end{array}\right)=\left[\begin{array}{cc}1& 1\\ -1\text{/}{}_{,1}& -1\text{/}{}_{,2}\end{array}\right]*\left(\begin{array}{c}{A}_{}\\ B_{}\end{array}\right)$

(24) Time constants .sub.,1, .sub.,2 are here provided from a specification 18, for instance from a memory element. Said time constants .sub.,1, .sub.,2 each lie in a time range of 0.3 s to 15.0 s, preferably approximately 0.5 s.

(25) The dependent variables obtained using the instantaneous input variables, i.e. the instantaneous yaw rate and yaw acceleration, which are measured directly or determined from measured values, are derived from the time-discrete linear model and are taken into account as weighting factors A.sub., B.sub. in determining the time-related function rules:

(26) $\left(\begin{array}{c}{A}_{}\\ B_{}\end{array}\right)=\frac{1}{{}_{,2}-{}_{,1}}*\left[\begin{array}{cc}-{}_{,1}& -{}_{,1}{}_{,2}\\ {}_{,2}& {}_{,1}{}_{,2}\end{array}\right]*\left(\begin{array}{c}{}_0\\ {\stackrel{.}{}}_0\end{array}\right).$

(27) The weighting factors A.sub., B.sub. are updated whenever a new measured value is acquired.

(28) In a determination step 19, the instantaneous weighting factors A.sub., B.sub. are used to determine a time-related function rule for the expected variation of an input variable of the linear model. Assuming the linear model, the expected yaw rate .sub.pre(t) can be determined as follows as a continuous function rule:
.sub.pre(t)=A.sub.exp(t/.sub.,1)+B.sub.exp(t/.sub.,2)

(29) The function rule .sub.pre(t) for predicting the yaw rate is analytically integrated in an integration step 20 in order to obtain the time-related function rule .sub.pre(t) for the bearing angle:
.sub.pre(t)=A.sub..sub.,1(1exp(t/.sub.,1))+B.sub.,2(1exp(t/.sub.,2))

(30) The function rule .sub.pre(t) for the expected speed of the vehicle is determined in a corresponding manner to determining the function rule for the bearing angle. In the combining step 17, the instantaneous acceleration .sub.0 and the value of an instantaneous sudden linear movement .Math..sub.0, which are available from the information from the vehicle-sensor unit 3, are taken into account for this purpose. The input variables of the instantaneous acceleration .sub.0 and of the sudden linear movement .Math..sub.0 are determined from the measured values of the instantaneous speed v0 by means of an estimation stage 41. A relationship between the acceleration and the sudden linear movement, which is the rate of change of the acceleration and hence the second derivative of the speed, is established in a linear model taking into account time constants .sub.,1, .sub.,2. The specified time constants .sub.,1, .sub.,2 are taken from a specification 18 as explained above with regard to the rotational travel state. From the second-order linear model for the longitudinal travel state is obtained the following equation for calculating the longitudinal weighting factors A.sub., B.sub. for determining the longitudinal travel state:

(31) $\left(\begin{array}{c}{}_0\\ {\mathrm{.Math.}}_0\end{array}\right)=\left[\begin{array}{cc}1& 1\\ -1\text{/}{}_{,1}& -1\text{/}{}_{,2}\end{array}\right]*\left(\begin{array}{c}{A}_{}\\ B_{}\end{array}\right)$

(32) The following weighting factors are obtained by rearranging the above equation in a linear system:

(33) $\left(\begin{array}{c}{A}_{}\\ B_{}\end{array}\right)=\frac{1}{{}_{,2}-{}_{,1}}*\left[\begin{array}{cc}-{}_{,1}& -{}_{,1}{}_{,2}\\ {}_{,2}& {}_{,1}{}_{,2}\end{array}\right]*\left(\begin{array}{c}{}_0\\ {\mathrm{.Math.}}_0\end{array}\right).$

(34) The weighting factors A.sub., B.sub. are updated whenever a new measured value is captured.

(35) In a subsequent determination step 19, the weighting factors for the longitudinal travel state are taken into account to obtain a time-related function rule for the expected acceleration according to the following equation:
.sub.pre(t)=A.sub.exp(t/.sub.,1)+B.sub.exp(t/.sub.,2).

(36) The time-related function rule for the expected variation of the linear acceleration is analytically integrated in a subsequent integration step 20, from which is obtained the following equation for the expected speed .sub.pre(t):
.sub.pre(t)=A.sub..sub.,1(1exp(t/.sub.,1))+B.sub..sub.,2(1exp(t/.sub.,2))+.sub.0

(37) In said equation, the initial speed .sub.0 of the vehicle is added to the predicted increases or decreases in speed at the respective measurement times.

(38) Using the function rules obtained for the expected speed and/or for the expected bearing angle, the numerical integration 16 is performed in accordance with the above-mentioned integral for the prediction result X.sub.pre(t). This integration takes into account both rotational and longitudinal changes for non steady-state travel states. In exemplary embodiments that are not shown, according to the invention, function rules are determined in linear models for specific travel states, while other travel states are assumed to be constant. This can reduce the time and cost involved in acquiring measured values.

(39) FIG. 3 shows a variation over time of the yaw rate . It is clear from this figure that the yaw rate for travel in a straight line is substantially constant and almost no angular movements occur in the vehicle. Changes in the yaw rate arise, for example, from travelling around a bend, which manifests itself as a positive peak 21. In the time interval IV-IV, which is shown magnified FIG. 4, the vehicle is travelling around a right-hand bend, which manoeuvre manifests itself as a negative peak in the yaw rate .

(40) The peaks in the yaw-rate curve are considered as isolated from one another in time. Over prolonged time periods, the yaw-rate signal always decays back to zero. The shape of the peaks can be approximated by a bell-shaped signal 22. The fact that the yaw rate of the vehicle increases when entering the turn and decreases when leaving the turn produces the common bell-shaped variation of the yaw-rate signal during cornering. Owing to the deflection in the yaw rate being only temporary and following a bell-shaped curve, the response of a second-order dynamic system can be used for predicting the expected travel path.

(41) The described approach for predicting the expected travel path that assumes a linear model of order at least two yields significantly more accurate prediction results than conventional prediction algorithms. FIG. 5 shows the expected travel path 14 compared with a conventional travel-path prediction 23 (shown dashed). The conventional travel-path prediction 23 results in a course for the predicted travel path that has a considerably narrower radius of curvature because it does not take into account the effect of changes in the travel state during non steady-state phases. The predicted travel path 14 according to the invention, which assumes a linear dynamic model of order two or higher, takes into account dynamic changes in the travel state during the prediction interval. Thereby in particular even those driver assistance systems that work cooperatively, i.e. take into account the movement of destination vehicles 24, can determine more accurate and realistic evasion scenarios.