Optimization of the amplification factor of acceleration controllers for automobiles

Abstract

An aspect of the present invention relates to an apparatus for optimizing an amplification factor of an acceleration controller for a vehicle, the apparatus comprising: storing as information for at least two time steps a target speed of the vehicle, an actual speed of the vehicle, and a target acceleration of the vehicle set based thereon; selecting a first subset of information; training a model based on the first subset, the model being configured to predict an actual speed of a subsequent time step from the at least one stored actual speed and the at least one stored target acceleration; selecting a second subset of information; and optimizing an amplification factor based on the second subset, the model, and the acceleration controller.

Description

[Technical Field]

[0001] The present invention relates to an apparatus and method for optimizing the gain of an acceleration controller for a motor vehicle. [Background technology]

[0002] Within the scope of this specification, the term "automated driving" refers to driving with automatic forward or lateral steering, or autonomous driving with automatic forward and lateral steering. The term "automated driving" includes automated driving with any level of automation. Exemplary levels of automation are assisted driving, partially automated driving, highly automated driving, and fully automated driving. These levels of automation are defined by the Federal Association of Transport Systems (Bundesanstalt fuer Strassenwesen) (BASt). In assisted driving, the driver continuously performs forward or lateral steering while the system assumes other functions within certain limits. In partially automated driving (TAF), the system assumes forward and lateral steering for certain periods and / or under special circumstances, but the driver must constantly monitor the system, as in assisted driving. In highly automated driving (HAF), the system assumes forward and lateral steering for certain periods without the driver having to constantly monitor the system, but the driver must be able to take over control of the vehicle within a certain period of time. In fully automated driving (VAF), the system automatically manages driving in any situation for a specific use case, and no longer requires a driver for this use case. The four automation levels defined by BASt correspond to SAE Levels 1-4 in the SAE J3016 standard (SAE-Society of Automotive Engineering). For example, highly automated driving (HAF) corresponds to Level 3 in the SAE J3016 standard. SAE J3016 also defines SAE Level 5 as the highest automation level not included in the BASt definition. SAE Level 5 corresponds to driverless driving, in which the system can automatically handle any situation as if it were a human driver throughout the entire journey, essentially no longer requiring a driver.

[0003] All driving guidance systems are based on an acceleration controller that determines the target acceleration of the vehicle based on the target speed and the actual speed of the vehicle. The driving guidance of the vehicle is then controlled according to at least this target acceleration.

[0004] Parameterization of the acceleration controller is very time-consuming because it needs to be adapted to each vehicle model and various characteristics of the vehicle mass and drivetrain, for example, influence the parameterization. Summary of the Invention [Problem to be solved by the invention]

[0005] The object of the present invention is to simplify the parameterization of the acceleration controller. [Means for solving the problem]

[0006] This problem is solved by the features of the independent patent claims. Advantageous embodiments are set out in the dependent claims. It should be noted that additional features of claims dependent on an independent patent claim, either without the features of the independent patent claim or in combination with only some of the features of the independent patent claim, can form an independent invention independent of the combination of all the features of the independent patent claim, which can be the subject of an independent claim, a divisional application or a subsequent application. The same applies to technical teachings contained in the description, which can form an invention independent of the features of the independent patent claim.

[0007] A first aspect of the invention relates to a device for optimising the gain of an acceleration controller for a motor vehicle, in particular an automated motor vehicle.

[0008] An acceleration controller is provided for setting a target acceleration of the vehicle at a time step based on the target vehicle speed, the actual vehicle speed, and a gain factor (VF).

[0009] The vehicle's direction of travel is then guided based at least on this target acceleration. In particular, the target acceleration of the drive or prime mover control is set as the desired acceleration. Alternatively or additionally, the target acceleration is further processed before the target acceleration is set as the desired acceleration for the drive or prime mover control.

[0010] The device is configured to store information on the target velocity, the actual velocity, and the target acceleration set based on them for at least two time steps.

[0011] In particular, the device is configured to individually store the target speed, the actual speed, and the target acceleration set based on them as a tuple, so that it can be further determined from the stored information that the above-mentioned multiple data correspond to the same time step.

[0012] In particular, this device is configured to store information on the target speed, actual speed, target acceleration set based on them, and each time step for at least two time steps, so that the causal or chronological order of the above-mentioned data can also be determined from the stored information.

[0013] Additionally, the device is arranged to select a first subset of the above information, the first subset including, in particular, a maximum of 150 or 200 tuples of target speed, actual speed and / or target acceleration, the invention being based on the insight of selecting a number of tuples that allows processing under real-time conditions, i.e., processing under restrictive deadlines.

[0014] The apparatus is configured to train a model based on the first subset, the model being configured to predict an actual velocity at a later time step from at least one stored actual velocity and at least one stored target acceleration.

[0015] Here, the present invention is based on the finding that the actual velocity at a second time step can be predicted from the actual velocity and target acceleration at a first time step, taking into account the time difference between the first time step and a second time step following the first time step.

[0016] The actual acceleration of a vehicle will certainly deviate from the target acceleration of the vehicle, since it depends not only on influences that can be controlled by the vehicle, but also on, for example, the slope of the road, signal transmission time within the vehicle and / or system inertia, etc. However, since the actual speed of the vehicle has been stored for several time steps and is therefore known, a model can be trained retrospectively by supervised learning methods.

[0017] The apparatus is further arranged to select a second subset of said information, said second subset including, inter alia, up to 20, 50, 100 or 150 tuples of target velocity, actual velocity and / or target acceleration.

[0018] The apparatus is further configured to optimize the amplification factor based on the second subset, the model and the acceleration controller.

[0019] The invention is based on the finding that the choice of amplification factor has a strong influence on how quickly and with what quality the actual speed of the vehicle becomes equal to the target speed, which may deviate from the actual speed. For example, a very large amplification factor certainly helps the actual speed to become equal to the target speed quickly, but in the case of a very large amplification factor accompanied by a time delay there is a risk of oscillations.

[0020] The device is particularly adapted to train the model and optimize the acceleration controller multiple times, iteratively converging on the optimal gain factor, e.g., by appropriately selecting the frequency at which the model is trained and the acceleration controller is optimized, the optimum can be found with low computational power.

[0021] In an advantageous embodiment of the invention, the acceleration controller is arranged to determine the target acceleration from the product of the difference between the target speed and the actual speed and an amplification factor.

[0022] In yet another advantageous embodiment of the invention, the device is arranged to store the information in a ring buffer, the capacity of which is limited to storing information for a maximum of 5000 time steps.

[0023] The ring buffer continues to store data for a certain period of time and, after a predetermined time, overwrites these data again to free up storage space for new data again.

[0024] In particular, since the time difference between any two time steps is at most 20 ms, the ring buffer can store information for up to 100 seconds.

[0025] In yet another advantageous embodiment of the invention, the apparatus is arranged to train the model by optimizing the first weighting coefficients and the second weighting coefficients to minimize a prediction error of the model.

[0026] In particular, the optimization of the first weighting factor and the second weighting factor is performed using the Levenberg-Marquardt algorithm, where the present invention is based on the finding that the Levenberg-Marquardt algorithm converges significantly faster than other optimization algorithms in this problem setting, which, in combination with other configurations, allows the present invention to be used in an automobile (i.e., "online" as opposed to "offline" training in a data center).

[0027] The first weighting factor determines the influence of at least one stored actual speed on the prediction. In particular, if the at least one stored actual speed includes more than one actual speed, multiple first weighting factors can be used. Thus, for example, a unique first weighting factor can be used for each of multiple actual speeds.

[0028] The second weighting factor determines the influence of at least one stored target acceleration on the prediction. In particular, if the at least one stored target acceleration includes more than one target acceleration, multiple second weighting factors can be used. Thus, for example, a unique second weighting factor can be used for each of multiple target accelerations.

[0029] In yet another advantageous embodiment of the invention, the device is arranged to optimise the amplification factor by predicting the state of the vehicle based on the second subset, the model and the acceleration controller.

[0030] The vehicle state may, among other things, describe the actual dynamics of the vehicle and / or describe control or target criteria for vehicle systems that will affect the vehicle dynamics in the future. For example, the vehicle state may include the vehicle's target acceleration at the current time step, the vehicle's actual speed at the current time step, and the vehicle's target speed at the current time step. Additionally, the vehicle state may include the vehicle's actual speed at at least one past time step and / or the target acceleration at at least one past time step.

[0031] In particular, since the complete state of the vehicle can only be described with considerable effort, in this embodiment of the invention the state of the vehicle can only be partially described, for example, by at least one actual speed of the vehicle, at least one target speed of the vehicle and / or at least one target acceleration of the vehicle.

[0032] Additionally, the system is adapted to adjust the gain to minimize the degree of controller quality related to the state of the vehicle.

[0033] The quality measure of the controller here describes in particular the control deviation and / or the passenger comfort index.

[0034] In particular, the optimization of the amplification factor is performed using the Levenberg-Marquardt algorithm, where the invention is based on the finding that the Levenberg-Marquardt algorithm converges much faster than other optimization algorithms in this problem setting, which, together with other features, allows the invention to be used in automobiles.

[0035] The vehicle state includes, in particular, at least one actual velocity of the vehicle and / or at least one target acceleration of the vehicle and / or at least one target velocity of the vehicle in one time step.

[0036] Thus, for example, the model can be used to make predictions of how the target acceleration, target velocity and actual velocity of the vehicle will change in future time steps, starting from an initial state of the vehicle, assuming different values ​​for the acceleration controller amplification factor.

[0037] In yet another advantageous embodiment of the invention, the apparatus is configured to store information in a ring buffer, the ring buffer having a capacity limited to store information for up to 5000 time steps; train the model by optimizing first and second weighting factors using a Levenberg-Marquardt algorithm to minimize a prediction error of the model, the first weighting factor defining an influence of at least one stored actual speed on the prediction, and the second weighting factor defining an influence of at least one stored target acceleration on the prediction; predict a state of the vehicle based on the second subset, the model, and the acceleration controller; and optimize the gain by optimizing the gain using the Levenberg-Marquardt algorithm to minimize a controller quality measure related to the state of the vehicle.

[0038] This advantageous embodiment combines all the features that make the invention efficient enough to be readily used in automobiles, despite the limited resources of the automobile control unit.

[0039] A second aspect of the present invention relates to a method of optimizing a gain of an acceleration controller for a motor vehicle, the acceleration controller being arranged to set a target acceleration of the motor vehicle in a single time step based on a target speed of the motor vehicle, an actual speed of the motor vehicle and a gain.

[0040] One step is to store information on the target velocity, the actual velocity, and the target acceleration set based on them for at least two time steps.

[0041] A further step in the method is to select a first subset of information.

[0042] A further step of the method is training a model based on the first subset, the model being arranged to predict the actual velocity at a later time step from the at least one stored actual velocity and the at least one stored target acceleration.

[0043] A further step in this method is to select a second subset of information.

[0044] A further step in the method is to optimize the amplification factor based on the second subset, the model and the acceleration controller.

[0045] The above explanations for the device according to the invention according to the first aspect of the invention also apply correspondingly to the method according to the invention according to the second aspect of the invention. Advantageous embodiments of the method according to the invention not explicitly mentioned here or in the claims correspond to the advantageous embodiments of the device according to the invention mentioned above or claimed.

[0046] The present invention will now be described by way of an embodiment with reference to the accompanying drawings. [Brief explanation of the drawings]

[0047] [Figure 1] 1 shows a device according to the invention for optimising the amplification factor VF of an acceleration controller BR for a motor vehicle; DETAILED DESCRIPTION OF THE INVENTION

[0048] The acceleration controller BR is arranged to set, in one time step, a target acceleration SB of the vehicle based on the target vehicle speed SG, the actual vehicle speed IG and an amplification factor VF.

[0049] Furthermore, the acceleration controller BR is configured to determine the target acceleration SB from the product of the difference between the target speed SG and the actual speed IG and the amplification factor VF.

[0050] The present device is configured to store, as information, the target speed SG, the actual speed IG, and the target acceleration SB set based on them for at least two time steps.

[0051] In particular, the device is arranged to store said information in a ring buffer RS, the capacity of which is limited to storing information for a maximum of 5000 time steps.

[0052] The apparatus is further configured to select a first subset of information ET and train a model MU based on the first subset ET, the model MU being configured to predict the actual velocity IG at a later time step from at least one stored actual velocity IG and at least one stored target acceleration SB.

[0053] In particular, the apparatus is configured to train the model MU by optimizing a first weighting coefficient and a second weighting coefficient to minimize a prediction error of the model MU, wherein the first weighting coefficient defines the influence of at least one stored actual speed IG on the prediction, and the second weighting coefficient defines the influence of at least one stored target acceleration SB on the prediction.

[0054] Furthermore, the device is arranged to select a second subset of information ZT and to optimise the amplification factor VF based on the second subset ZT, the model MU and the acceleration controller BR, for example by using optimisation means CU.

[0055] In particular, the apparatus is configured to optimize the gain VF by providing a device for predicting the state of the vehicle based on the second subset ZT, the model MU and the acceleration controller BR and optimizing the gain VF so as to minimize a degree of controller quality related to the state of the vehicle.

[0056] In this case, the state of the vehicle includes at least one actual velocity IG of the vehicle and / or at least one target acceleration SB of the vehicle in one time step.

Claims

1. 1. A device for optimizing a gain (VF) of an acceleration controller (BR) for a motor vehicle, comprising: said acceleration controller (BR) is arranged to set, during one time step, a target acceleration (SB) of said vehicle on the basis of a target speed (SG) of said vehicle, an actual speed (IG) of said vehicle and said amplification factor (VF); - storing the target speed (SG), the actual speed (IG) and the target acceleration (SB) set based thereon as information for at least two time steps; - selecting a first subset of said information (ET), training a model (MU) on said first subset (ET), said model (MU) being adapted to predict an actual velocity (IG) of a subsequent time step from at least one stored actual velocity (IG) and at least one stored target acceleration (SB); - selecting a second subset (ZT) of said information, - Optimizing the amplification factor (VF) based on the second subset (ZT), the model (MU) and the acceleration controller (BR). A device that is provided so as to:

2. 2. The device according to claim 1, wherein the acceleration controller (BR) is configured to calculate the target acceleration (SB) from the product of the difference between the target speed (SG) and the actual speed (IG) and the amplification factor (VF).

3. 3. A device according to claim 1 or 2, arranged to store said information in a ring buffer (RS), the capacity of said ring buffer (RS) being limited to storing information for a maximum of 5000 time steps.

4. 4. An apparatus according to any one of claims 1 to 3, - train the model (MU) by optimizing the first and second weighting coefficients so as to minimize the prediction error of the model (MU); said first weighting factor defines the influence of at least one stored said actual speed (IG) on said prediction; said second weighting factor defines the influence of at least one stored target acceleration (SB) on said prediction; Device.

5. 5. An apparatus according to any one of claims 1 to 4, The device, - predicting the state of the vehicle based on said second subset (ZT), said model (MU) and said acceleration controller (BR); - Optimizing the gain (VF) to minimize the degree of controller quality related to the state of the vehicle. The device is configured to optimize the amplification factor (VF) by providing the amplifier as described above.

6. 6. The apparatus of claim 5, wherein the state of the vehicle includes the actual velocity (IG) of the vehicle and / or the target acceleration (SB) of the vehicle at least one time step.

7. 7. An apparatus according to any one of claims 1 to 6, - it is provided to store said information in a ring buffer (RS), the capacity of which is limited to storing said information for a maximum of 5000 time steps; - training said model (MU) by optimizing first and second weighting factors using a Levenberg-Marquardt algorithm so as to minimize a prediction error of said model (MU), said first weighting factor defining the influence of at least one stored said actual speed (IG) on said prediction, and said second weighting factor defining the influence of at least one stored said target acceleration (SB) on said prediction; - predicting the state of the vehicle based on said second subset (ZT), said model (MU) and said acceleration controller (BR); - by optimizing said amplification factor (VF) using the Levenberg-Marquardt algorithm so as to minimize a degree of controller quality related to said state of said vehicle, To optimize the amplification factor (VF) The device provided.

8. 1. A method for optimizing a gain (VF) of an acceleration controller (BR) for a motor vehicle, comprising: said acceleration controller (BR) is arranged to set, during one time step, a target acceleration (SB) of said vehicle on the basis of a target speed (SG) of said vehicle, an actual speed (IG) of said vehicle and said amplification factor (VF); - below; - storing the target speed (SG), the actual speed (IG) and the target acceleration (SB) set based thereon as information for at least two time steps; - selecting a first subset of said information (ET), training a model (MU) on said first subset (ET), said model (MU) being adapted to predict an actual velocity (IG) of a subsequent time step from at least one stored actual velocity (IG) and at least one stored target acceleration (SB); - selecting a second subset (ZT) of said information, - Optimizing the amplification factor (VF) based on the second subset (ZT), the model (MU) and the acceleration controller (BR). A method having the steps.

Images