CONTROL SYSTEM FOR A MOTOR VEHICLE AND METHOD FOR ADAPTING THE CONTROL SYSTEM

Abstract

A control system for a motor vehicle, for outputting a controlled variable, with the aid of which a directly controlled variable of a motor vehicle is adjustable via suitable control operations, in order to adapt the directly controlled variable to a reference variable of the control system. The control system includes a controller, which is configured to output a first output variable on the basis of the directly controlled variable of the motor vehicle, and on the basis of the reference variable of the control system. The control system further includes a predictive model, which may be trained to output a second output variable that reflects a deviation of a driving behavior of a driver of the motor vehicle from the first output variable of the controller. The controlled variable of the control system encompasses an addition of the first output variable and the second output variable.

Claims

1. A control system for a motor vehicle configured to output a controlled variable, using which a directly controlled variable of a motor vehicle is adjustable via control operations, to adapt the directly controlled variable to a reference variable of the control system, the control system comprising: a controller configured to output a first output variable based on the directly controlled variable of the motor vehicle, and based on the reference variable of the control system; and a predictive model trained to output a second output variable that reflects a deviation of a driving behavior of a driver of the motor vehicle from the first output variable of the controller; wherein the controlled variable of the control system encompasses an addition of the first output variable and the second output variable.

2. The control system as recited in claim 1, wherein the directly controlled variable of the motor vehicle reflects a distance of the motor vehicle from a reference object in a surrounding area of the motor vehicle.

3. The control system as recited in claim 1, wherein: (i) the controller includes a PID-type controller, and/or (ii) the predictive model includes a Gaussian process model or a neural network.

4. The control system as recited in claim 1, wherein the predictive model is trained to output the second output variable as a function of at least one input variable, and wherein the at least one of the input variables includes one of the following variables: the reference variable of the control system, and/or the directly controlled variable of the motor vehicle, and/or a variable that represents operating data of the motor vehicle, and/or surrounding-area data of the motor vehicle.

5. A computer-implemented method for training a predictive model for a control system for a motor vehicle, the control system configured to output a controlled variable, using which a directly controlled variable of a motor vehicle is adjustable via control operations, to adapt the directly controlled variable to a reference variable of the control system, the control system including a controller and the predictive model, the method comprising the following steps: a first training phase including: in a deactivated state of the control system, ascertaining a deviation of a driving behavior of a driver of the motor vehicle from a first output variable of a controller of the control system; and training the predictive model, using the ascertained deviation of the driving behavior.

6. The method as recited in claim 5, wherein the first training phase further includes: ascertaining the driving behavior of the driver as a function of the directly controlled variable of the motor vehicle, and computing the first output variable of the controller.

7. The method as recited in claim 5, wherein the ascertaining of the driving behavior includes ascertaining at least one variable which represents: an accelerator pedal action and/or a braking action and/or a steering action.

8. The method as recited in claim 5, wherein the training takes place as a function of at least one further variable, which represents operating data of the motor vehicle and/or surrounding-area data of the motor vehicle.

9. The method as recited in claim 5, the method further comprising: a second training phase including optimizing the predictive model as a function of at least one further variable, which is associated with a reference object in a surrounding area of the motor vehicle.

10. The method as recited in claim 9, wherein the optimizing of the predictive model includes: ascertaining a state of the motor vehicle at one time, including at least one variable, which is associated with the motor vehicle; ascertaining a state of the reference object at the time, including at least one variable, which is associated with the reference object; ascertaining a distribution over future states; and identifying at least one model parameter, which minimizes an expected value of an error in the distribution over the future states.

11. The method as recited in claim 9, the method further comprising: a third training phase including: in an activated state of the control system, testing the predictive model in comparison with an action of the driver.

12. The method as recited in claim 11, wherein: (i) the first and/or the second training phase are repeated, and/or (ii) further steps including deactivating the control system and/or outputting a warning, are executed as a function of the testing of the predictive model.

13. A non-transitory machine-readable storage medium on which is stored a computer program for training a predictive model for a control system for a motor vehicle, the control system configured to output a controlled variable, using which a directly controlled variable of a motor vehicle is adjustable via control operations, to adapt the directly controlled variable to a reference variable of the control system, the control system including a controller and the predictive model, the computer program, when executed by a computer, causing the computer to perform the following steps: a first training phase including: in a deactivated state of the control system, ascertaining a deviation of a driving behavior of a driver of the motor vehicle from a first output variable of a controller of the control system; and training the predictive model, using the ascertained deviation of the driving behavior.

14. A control unit configured to train a predictive model for a control system for a motor vehicle, the control system configured to output a controlled variable, using which a directly controlled variable of a motor vehicle is adjustable via control operations, to adapt the directly controlled variable to a reference variable of the control system, the control system including a controller and the predictive model, the control unit configured to, in a first training phase: in a deactivated state of the control system, ascertain a deviation of a driving behavior of a driver of the motor vehicle from a first output variable of a controller of the control system; and train the predictive model, using the ascertained deviation of the driving behavior.

15. The control system as recited in claim 1, wherein the control system is used for adapting a system of the motor vehicle to an individual driving behavior of the driver.

16. The control system as recited in claim 1, wherein the control system is used in a driving assistance system of a motor vehicle for adaptive cruise control (ACC).

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0031] FIG. 1 shows a schematic representation of a control system for a motor vehicle in accordance with an example embodiment of the present invention.

[0032] FIG. 2 shows a schematic representation of steps of a first training phase of a computer-implemented method for training a predictive model in accordance with an example embodiment of the present invention.

[0033] FIG. 3 shows a schematic representation of steps of a second training phase of a computer-implemented method for training a predictive model in accordance with an example embodiment of the present invention.

[0034] FIG. 4 shows a schematic representation of steps of a third training phase of a computer-implemented method for training a predictive model in accordance with an example embodiment of the present invention.

[0035] FIG. 5 shows a schematic overall view of the second training phase according to FIG. 3 in accordance with an example embodiment of the present invention.

[0036] FIG. 6 shows a schematic view of a control unit for a motor vehicle in accordance with an example embodiment of the present invention.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

[0037] FIG. 1 schematically shows a control system 100 for a motor vehicle (not shown), for outputting a controlled variable u, with the aid of which a directly controlled variable y of a motor vehicle is adjustable via suitable control operations, in order to adapt directly controlled variable y to a reference variable w of the control system.

[0038] To implement these control operations, the control system is preferably connected to an electrical system of the motor vehicle, using, preferably, at least one bus, preferably, the CAN bus (not shown), so that by actively intervening in on-board systems, such as, in particular, a steering system, brake system, power train and warning systems, directly controlled variable y may be adapted to a reference variable w of the control system.

[0039] The control system includes a controller 110, which is configured to output a first output variable u1 on the basis of directly controlled variable y of the motor vehicle, and on the basis of reference variable w of the control system. Controller 110 of control system 100 includes, for example, a conventional control algorithm, for example, a PID-type controller.

[0040] Control system 100 further includes a predictive model 120, which may be trained to output a second output variable u2 that reflects a deviation of a driving behavior of a driver of the motor vehicle from first output variable u1 of the controller. According to the specific embodiment shown, controlled variable u of control system 100 encompasses an addition of first output variable u1 and second output variable u2.

[0041] In order to adapt control system 100 to the driving behavior of an individual driver, then, with the aid of predictive model 120, the difference of the driving behavior from current controller 110 is modeled, and control system 100 is adapted to the driving behavior of an individual driver, by adding second output variable u2 of predictive model 120, which reflects the deviation of the driving behavior of a driver of the motor vehicle from first output variable u1 of controller 110, to first output variable u1 of controller 110.

[0042] Control system 100 is, for example, a driving assistance system, which may be used in a motor vehicle, in order to assist and/or relieve the stress on the driver in certain driving situations, for example, for regulating the distance from a reference object, in particular, a ranging assistance system or a parking assistance system or an assistance system for integrating a vehicle driving at least partially autonomously into a flow of traffic.

[0043] To control spacing, a distance of the motor vehicle from the reference object is normally adapted to a desired setpoint value, that is, to the reference variable of the control system, using suitable control operations, such as acceleration and/or braking and/or steering actions. By adjusting the controlled variable to the driving behavior of an individual driver, the control operations may be adjusted to the driving behavior, as well. This advantageously increases the acceptance of such systems.

[0044] In one further preferred specific embodiment of the present invention, the directly controlled variable of the motor vehicle reflects a distance of the motor vehicle from a reference object in a surrounding area of the motor vehicle.

[0045] The reference object in the surrounding area of the motor vehicle is, for example, a third motor vehicle, in particular, one driving ahead, a pedestrian, an animal or another road user. Alternatively, the reference object may also be a stationary object in the surrounding area, for example, a guardrail, a tree, a pole, a building, or the like. In the same way, a road marking, such as a lane boundary, broken white line, or the like, may also be understood as a reference object, as well.

[0046] In order to measure the distance of the motor vehicle from the reference object, the motor vehicle preferably includes surround sensors (not shown), such as radar sensors, lidar sensors, laser scanners, video sensors and ultrasonic sensors. If the motor vehicle is equipped with a navigation system, then data of this system may also be accessed.

[0047] In one further preferred specific embodiment of the present invention, controller 110 includes a conventional type of controller, in particular, a PID-type controller, and/or predictive model 120 includes a Gaussian process model or a neural network.

[0048] In a further preferred specific embodiment of the present invention, predictive model 120 may be trained to output second output variable u2 as a function of at least one input variable; an input variable including one of the following variables: reference variable w of the control system, directly controlled variable y of the motor vehicle, a variable that represents operating data of the motor vehicle and/or surrounding-area data of the motor vehicle. Reference variable w of control system 100 is the desired setpoint value, to which directly controlled variable y is intended to be adapted. Operating data of the motor vehicle include, for example, speed, acceleration, steering angle, inclination. Surrounding-area data of the motor vehicle include, for example, information about the road condition, weather, grade of the road, course of the road, etc. By utilizing the above-mentioned variables as input variables for predictive model 120, second output variable u2 may be outputted advantageously as a function of these variables. These variables are advantageously measured by suitable sensors, such as surround sensors, and/or provided to the control system by suitable devices for transmitting data.

[0049] FIG. 2 schematically shows steps of a first training phase of a computer-implemented method 200 for training a predictive model 120 for a control system 100 for a motor vehicle, according to the specific embodiments of the present invention; the first training phase including the following steps: in a deactivated state of control system 100, ascertaining 220 a deviation of a driving behavior of a driver of the motor vehicle from first output variable u1 of controller 110 of control system 100; and training 230 predictive model 120, using the ascertained deviation of the driving behavior.

[0050] A deactivated state of control system 100 is understood to mean that control system 100 is not used for controlling a driving assistance function, but that the driver of the motor vehicle controls this.

[0051] In further preferred specific embodiments of the present invention, the first training phase of method 200 further includes the following steps: ascertaining 210a the driving behavior of the driver as a function of directly controlled variable y of the motor vehicle; and computing 210b first output variable u1 of controller 110. In light of computed, first output variable u1 of controller 110 and the ascertained driving behavior with a deactivated control system 100, the deviation of the driving behavior from first output variable u1 of the controller may be ascertained. Predictive model 120 is advantageously trained, using the ascertained deviation of the driving behavior as a function of directly controlled variable y of the motor vehicle.

[0052] In further preferred specific embodiments of the present invention, the ascertaining 210a of the driving behavior includes the ascertaining of at least one variable, which represents an accelerator pedal action and/or a braking action and/or a steering action.

[0053] In further preferred specific embodiments of the present invention, the training of predictive model 120 takes place as a function of at least one further variable, which represents operating data of the motor vehicle and/or surrounding-area data of the motor vehicle. Operating data of the motor vehicle include, for example, speed, acceleration, steering angle, inclination. Surrounding-area data of the motor vehicle include, for example, information about the road condition, weather, grade of the road, course of the road, etc.

[0054] In one further preferred specific embodiment of the present invention, a second training phase of the method includes: optimizing the predictive model as a function of at least one further variable, which is associated with a reference object in a surrounding area of the motor vehicle. The reference object is, for example, a third vehicle, in particular, one driving ahead. By optimizing predictive model 120 with regard to the reference object, predictive model 120 may be optimized advantageously with regard to a future position of the reference object.

[0055] In one further preferred specific embodiment (FIG. 3) for the present invention, the optimizing (240) of predictive model 120 further includes: ascertaining (240a) a state of the motor vehicle at a time t, including at least one variable, which is associated with the motor vehicle; ascertaining (240b) a state of the reference object at time t, including at least one variable, which is associated with the reference object; and ascertaining (240c) a distribution over future states and identifying (240d) at least one model parameter, which minimizes the expected value of an error in the distribution over the future states. The model parameter characterizes an association between input variables and output variables of predictive model 120.

[0056] In this manner, the formation of a prediction error that accumulates in the long term may be advantageously prevented. In particular, an error that accumulates long-term may be formed, if predictive model 120 is not able to reflect the deviation of the driving behavior accurately.

[0057] FIG. 5 shows a schematic overall view of the second training phase for optimizing predictive model 120 as a function of at least one further variable, which is associated with a reference object in a surrounding area of the motor vehicle. The variable, which is associated with a reference object in a surrounding area of the motor vehicle, is given by a further, second predictive model 130, which is suitable for predicting a state of the reference object. A further, third predictive model 140 combines controller 110 and predictive model 120 and, thus, is suitable for predicting the state of the motor vehicle.

[0058] x.sub.t.sup.own represents the state of the motor vehicle at time t. x.sub.t.sup.own advantageously includes all of the variables, which are made available to predictive model 120 and controller 110. x.sub.t.sup.lead represents the state, in particular, information about the position and/or speed, of the reference object, for example, a third vehicle driving ahead, at time t. The distance from this reference object at time t is also supplied to predictive model 120 and controller 110.

[0059] If at least one of the predictive models 120, 130, 140 or controller 110 is a stochastic model, then a distribution over future states may be derived from it; the distribution being given by

p(x.sub.t+1.sup.own,x.sub.t+1.sup.lead,x.sub.t+2.sup.own,x.sub.t+2.sup.lead, . . . |x.sub.t.sup.own,x.sub.t.sup.lead,θ).

[0060] An error in the future states at time t+δ is given by L(x.sub.t+δ.sup.own,x.sub.t+δ.sup.lead). An error measures, for example, a difference from the reference variable and/or an exceedance and/or undershooting of maximum or minimum allowable differences. A model parameter, which minimizes the expected value of the error, solves the following optimization problem

θ=argmin.sub.θ[Σ.sub.δ=1.sup.T.sup.maxL(x.sub.t+δ.sup.own,x.sub.t+δ.sup.lead)|p(x.sub.t+1.sup.own,x.sub.t+1.sup.lead,x.sub.t+2.sup.own,x.sub.t+2.sup.lead, . . . |x.sub.t.sup.own,x.sub.t.sup.lead,θ)],

where T.sub.max describes the maximum prediction horizon. The identified model parameter minimizes the accumulated error of time step T.sub.max. Predictive model 120 is advantageously optimized on this basis.

[0061] In one further preferred specific embodiment of the present invention, a third training phase of method 200 includes: in the activated state of the control system, testing 250 the predictive model in comparison with an action of the driver. A schematic depiction of steps of the third training phase of computer-implemented method 200 is shown in FIG. 4.

[0062] In one further preferred specific embodiment of the present invention, the first and/or the second training phase are repeated, and/or further steps, in particular, deactivation 260a of control system 100 and/or outputting 260b of a warning, are executed as a function of the testing 250 of predictive model 120.

[0063] Further preferred specific embodiments of the present invention relate to a computer program, which is configured to execute the steps of the method 200 according to the specific embodiments.

[0064] Further preferred specific embodiments of the present invention relate to a machine-readable storage medium, in which the computer program according to the specific embodiments is stored.

[0065] Further preferred specific embodiments of the present invention relate to a control unit 300, which is configured to execute the steps of a method 200 according to the specific embodiments of the present invention. Control unit 300 includes a computing device 310 and at least one storage device 320, in which control system 100 is stored. In addition, control unit 300 includes an input 330 for receiving information about variables of the control system, such as a reference variable and directly controlled variable, and additional variables, which represent the operating data of the motor vehicle and/or surrounding-area data of the motor vehicle. These variables are advantageously measured by suitable sensors, such as surround sensors, and/or provided to the control system by suitable devices for transmitting data. Furthermore, control unit 300 includes an output 340 for controlling actuators of on-board systems of the motor vehicle, in particular, a steering system, brake system, the power train, and warning systems.

[0066] Further preferred specific embodiments of the present invention relate to use of a control system 100 according to the specific embodiments, and/or of a predictive model 120 that is trained by a method 200 according to the specific embodiments, and/or of a method according to the specific embodiments, and/or of a computer program according to the specific embodiments, and/or of a machine-readable storage medium according to the specific embodiments, and/or of a control unit 300 according to the specific embodiments, for adapting a control system 100 for a motor vehicle to an individual driving behavior of a driver.

[0067] Further preferred specific embodiments of the present invention relate to use of a control system 100 according to the specific embodiments of the present invention, and/or of a predictive model 120 that is trained by a method 200 according to the specific embodiments of the present invention, and/or of a method 200 according to the specific embodiments of the present invention, and/or of a computer program according to the specific embodiments of the present invention, and/or of a machine-readable storage medium according to the specific embodiments of the present invention, and/or of a control unit 300 according to the specific embodiments of the present invention, in a driving assistance system of a motor vehicle, in particular, for adaptive cruise control (ACC).