DE102019115967A1 - Hybrid vehicle model based on a single-track model and a neural network - Google Patents

Abstract

The invention relates to a method for the efficient and precise prediction of driving behavior. According to the invention, it is possible to upgrade or correct a simple prediction model in such a way that a reliable prediction is made possible despite low computing resources. This is particularly advantageous in autonomous driving, since this requires a reliable database, but the motor vehicle typically has few computing systems. The present invention also relates to a correspondingly configured system arrangement. In addition, a computer program product with control commands is proposed which implement the method or operate the proposed system arrangement.

Description

The invention relates to a method for the efficient and precise prediction of driving behavior. According to the invention, it is possible to upgrade or correct a simple prediction model in such a way that a reliable prediction is made possible despite low computing resources. This is particularly advantageous in autonomous driving, since this requires a reliable database, but the motor vehicle typically has few computing systems. The present invention also relates to a correspondingly configured system arrangement. In addition, a computer program product with control commands is proposed which implement the method or operate the proposed system arrangement.

WO 2007 060 134 A1 FIG. 10 shows a method for determining a value of a model parameter of a reference vehicle model. The reference value of the state variable is usually determined using a vehicle model. The vehicle models that can be used usually contain several parameters that must be adapted to a specific vehicle type so that the model correctly reproduces the reference behavior of a specific vehicle.

DE 10 2016 218 549 B3 shows a method for determining at least one prognosis result relating to a probable future collision of a motor vehicle with a collision object. To take into account the driving geometry, a two-lane model can preferably be used for four-wheeled motor vehicles and a single-lane model for two-wheelers, which can be parameterized, for example, by a wheel spacing.

EP 3 463 940 A1 shows a method for operating a control device for a chassis of a motor vehicle. The estimated value is determined by means of the single-track model, which has at least one model parameter that is dependent on the tire class. By changing the tire class on which the single-track model is based, the single-track model is adapted so that a different estimated value is obtained.

There is a trade-off between model accuracy and complexity / computing time. Depending on the application, either simple models are used, which do not exactly map the dynamics of the vehicle in all situations, but are not complex and enable quick simulation. Alternatively, a highly accurate analytical vehicle model is created, which consists of many individual components, has to be parameterized and validated in a complex manner and is very computationally intensive

Autonomous driving has its own requirements and therefore some known methods cannot be reused. Reliable data must be supplied in real time to ensure safe control of the vehicle. There is therefore a need for a new method which, in the underlying application scenario, creates a reliable prediction of driving behavior even when there are few system resources available.

It is an object of the present invention to propose an improved, in particular more reliable, method for the efficient and precise prediction of driving behavior, preferably of an automobile or motorcycle. The method should work with sufficient accuracy, but should have a low computational complexity. Furthermore, it is an object of the present invention to provide a correspondingly configured system arrangement and to propose a computer program product with control commands which implement the method or at least partially operate the proposed system arrangement.

The object is achieved by a method with the features according to claim 1. Further advantageous configurations are specified in the subclaims.

Accordingly, a method for efficient and accurate prediction of driving behavior is proposed for use in autonomous driving of a motor vehicle, comprising generating reference data which describe an actual driving behavior of the motor vehicle; an operation of a single-lane model for generating preliminary prediction data of the driving behavior of the motor vehicle; a detection of deviations of the preliminary prediction data from the reference data by means of artificial intelligence; and a customization 103 the preliminary prediction data as a function of the detected deviations for the generation of prediction data to be output.

The idea for modeling a new type of vehicle model arose from the requirement to create a highly accurate simulation model that delivers highly accurate simulation results in a minimum of computing time. The model is to be used to secure driving functions and to parameterize them on the basis of scenarios. There are approaches to model analytical models with black box models (such as neural networks). The idea of combining a simple analytical model (single-track model) with a neural network is, however, inventive.

The invention proposes combining a simple analytical model (single-track model) with a neural network and thus the accuracy of a complex analytical model, but to reduce the complexity and the modeling and parameterization effort to a minimum. The single-track model serves as the basis for this. The neural network in turn serves to learn the model inaccuracies of the single-track model and to compensate for the error in relation to reality by adding the outputs of the single-track model and the neural network. The advantages are, among other things, firstly, less modeling effort, secondly, less complexity and thus computing time, and thirdly, better traceability. a pure neural network.

The proposed method is efficient and accurate since it advantageously combines several methods. The so-called single-track model is used, for example, which enables the driving behavior of a motor vehicle to be predicted. The single-track model is kept very simple and does not provide sufficiently valid data for autonomous driving. There is always a trade-off between the complexity of the calculation and the accuracy of the resulting data. The single-lane model is therefore particularly suitable for autonomous driving, since a motor vehicle typically has very limited system resources and data is required in real time. In the case of autonomous driving, corresponding control data must always be available in real time at runtime, since otherwise safe control cannot be guaranteed.

The single-track model thus delivers pragmatic data, which can typically differ from real data. The single-track model, for example, has deviations or errors, since higher accuracy would also require higher complexity. According to the invention, this disadvantage is overcome in that a further model, which is based on artificial intelligence, is added to this model.

Thus, according to the invention, it is possible to continue to use the single-track model and thus to use the advantage of the low computational complexity, but in addition, corresponding deviations are corrected. Based on the large amounts of data that were collected empirically, for example, the artificial intelligence can determine where there are deviations or in which situations the single-track model is particularly prone to errors. These deviations can then be corrected using artificial intelligence.

In a preparatory method step, reference data are generated which describe an actual driving behavior of the motor vehicle. These are parameters that describe individual components and how they behave in certain situations. Driving behavior can relate, for example, to the behavior of the vehicle in curves, or it is also possible to describe a braking process. The generation of reference data can either take place in that sensors are read out during real test drives or a model is operated which is highly complex. In general, both approaches are to be combined in such a way that, in the end, actual driving behavior based on real test data and data from a model is used.

The collection of the reference data or the generation can take place at the factory in such a way that corresponding data only need to be stored in the vehicle and need not be generated in the field. Thus, an increased computational complexity of a corresponding model is used, but this can be done in advance and does not have to be done by the vehicle itself. This ensures that invalid reference data is available. The artificial intelligence can be trained on the basis of this reference data and the single-track model can be analyzed for corresponding errors or deviations.

In order to make this possible, according to the invention, the single-track model is operated in such a way that preliminary prediction data are produced which describe the driving behavior of the motor vehicle. This involves conventional data that is read out from this model or generated with this model. It describes the driving behavior, which generally requires a low level of computational complexity, but deviations are to be expected. These deviations can be recognized in that the reference data generated are compared with the preliminary prediction data generated and consequently it can be learned which errors the single-track model generates or these errors can then be compensated for. This is particularly advantageous because the low complexity of the single-track model can be used at runtime and the artificial intelligence can efficiently compensate for corresponding deviations. Thus, the creation of prediction data takes place in an efficient and yet precise manner.

According to the invention, it is advantageous that large amounts of reference data can be collected and that the single-track model can also be simulated before ongoing operation. It is thus possible to generate appropriate data at the factory and then use a trained artificial intelligence during the runtime. The process steps of generating reference data, operating the single-track model and recognizing deviations are thus carried out at the factory in such a way that parameterization at runtime is possible in such a way that corresponding errors are compensated for.

Thus the adaptation of the preliminary prediction data takes place at runtime, i. E. in other words, in the field when the vehicle has been delivered and is in operation. Thus, according to the invention, it is possible to split up conventional simulation processes and distribute them advantageously in such a way that complex computing processes can be carried out in the factory and corresponding parameters only have to be stored in the vehicle. The parameters that are used to adapt the preliminary prediction data are stored in the vehicle and can be used at runtime.

Computing complexity typically refers to the fact that it is described on the basis of necessary computing steps, or else required system resources are described. Computing complexity can demand a certain processor power or it can be particularly memory-intensive. A general measure of the computing complexity is the number of computing steps required.

The artificial intelligence can be implemented, for example, by providing a neural network which analyzes the reference data and the generated data and then determines deviations. In this way, situations can be described in which the single-track model works reliably and other situations which require adaptation. If these situations are known, then appropriate knowledge only needs to be applied at runtime and only those situations need to be adjusted which are faulty or do not have the required accuracy. It can happen that the single-track model works imprecisely in certain situations, which also involves very specific vehicle parameters. Depending on the type of vehicle, the suitability of this model can vary and corresponding deviations can be learned.

The preliminary prediction data are then adapted in such a way that the error is applied to the data in such a way that actual, correct data are again available. This can be done in that, for example, a detected discrepancy is added up inversely and the error is thus compensated. For example, if a parameter deviates by a value of five and is incorrectly five units higher than the expected value, these five units can be subtracted again and the correct value results. Corresponding data are typically represented as a function or, figuratively speaking, by a curve and consequently the deviation can be corrected by superimposing a correction curve or by applying a correction function.

According to one aspect of the present invention, the reference data are generated by means of a vehicle model and / or reading out sensors. This has the advantage that real sensor data can be used which are determined during test drives of the motor vehicle. In addition, it is possible to use another vehicle model that has sufficient accuracy. This vehicle model can be operated at the factory and corresponding data can then be used to parameterize the single-track model at runtime. In addition, it is also possible to combine both approaches in such a way that both sensor data and data from a vehicle model are used. In this way, reliable sensor data can be used if a deviation is to be expected and the database is particularly valid.

According to a further aspect of the present invention, the reference data are generated by means of a vehicle model which is more complex than the single-track model. This has the advantage that a more precise model can be used and thus the errors in the single-track model can be corrected in such a way that the complexity of the further model is used at the factory. Models of different complexity are available and it is possible to calculate the corresponding parameters in the factory with particular precision and then to use them again efficiently at runtime.

According to a further aspect of the present invention, the provisional prediction data are adapted by adding the deviations. This has the advantage that the data can be adjusted in a simple manner and consequently the errors only have to be adjusted by means of a summation. It is particularly preferable for the inverse of the error to be added up and consequently a valid value results.

According to a further aspect of the present invention, the prediction data to be output will describe driving parameters which are stored for further use. This has the advantage that the parameters can be determined empirically and a high level of computational complexity is only necessary in advance. In this way, the driving parameters can be calculated with computing capacities that are not available in the field and then these resulting parameters only have to be stored in a vehicle memory.

According to a further aspect of the present invention, the motor vehicle is under Use of the prediction data to be output is controlled automatically. This has the advantage that the data are not only output, but rather are used directly to control the vehicle. According to the invention, it is particularly advantageous that the model works particularly efficiently, since only the complexity of the single-track model is used and deviations are compensated for. The computing capacity of a vehicle is thus sufficient and the accuracy of the data is sufficient to ensure safe control of the vehicle.

According to a further aspect of the present invention, the prediction data to be output are used in the context of a safety function and / or a comfort function. This has the advantage that not only safe operation of the vehicle is possible, but rather comfort can also be guaranteed in that the driver can choose different driving modes and thus, for example, can drive in a particularly sporty or particularly relaxed manner. In this way, driving parameters can be set that correspond to the driver's wishes.

According to a further aspect of the present invention, the artificial intelligence is implemented using a neural network and / or a long short-term memory. This has the advantage that different techniques can be combined and, in particular, it is possible to process large data in such a way that patterns are recognized independently, resulting in algorithms that can react in a situation-specific manner. It is thus also possible to recognize patterns that remain hidden from the human observer and influencing variables are analyzed that only result from large amounts of data.

The object is also achieved by a system arrangement for efficient and precise prediction of driving behavior for use in autonomous driving of a motor vehicle, having a test device set up to generate reference data which describe actual driving behavior of the motor vehicle; a simulation device set up to operate a single-track model for generating preliminary prediction data of the driving behavior of the motor vehicle; a computing device set up to recognize deviations of the preliminary prediction data from the reference data by means of artificial intelligence; and a control device set up to adapt the preliminary prediction data as a function of the detected deviations in order to generate prediction data to be output.

The object is also achieved by a computer program product with control commands which execute the method and operate the proposed arrangement when they are executed on a computer.

According to the invention, it is particularly advantageous that the method can be used to operate the proposed devices and units or the system arrangement. The proposed devices and devices are also suitable for carrying out the method according to the invention. The device thus in each case implements structural features which are suitable for carrying out the corresponding method. However, the structural features can also be designed as method steps. The proposed method also provides steps for implementing the function of the structural features.

• 1: ein schematisches Blockschaltbild eines Aspekts der vorgeschlagenen Systemanordnung; und
• 2: ein schematisches Ablaufdiagramm eines Verfahrens zur effizienten und genauen Prädiktion eines Fahrverhaltens zur Verwendung bei einem autonomen Führen eines Kraftfahrzeugs gemäß einem Aspekt der vorliegenden Erfindung.

Further advantages, features and details of the invention emerge from the following description, in which aspects of the invention are described in detail with reference to the drawings. The features mentioned in the claims and in the description can each be essential to the invention individually or in any combination. The features mentioned above and those detailed here can also be used individually or collectively in any combination. Parts or components that are functionally similar or identical are in some cases provided with the same reference symbols. The terms “left”, “right”, “above” and “below” used in the description of the exemplary embodiments relate to the drawings in an orientation with normally legible figure designation or normally legible reference symbols. The embodiments shown and described are not to be understood as conclusive, but are exemplary for explaining the invention. The detailed description is provided for the information of those skilled in the art; therefore, in the description, known circuits, structures and methods are not shown or explained in detail in order not to make the present description more difficult to understand. In the figures show:

• 1 Figure 3 is a schematic block diagram of one aspect of the proposed system arrangement; and
• 2 : a schematic flow diagram of a method for efficient and precise prediction of driving behavior for use in autonomous driving of a motor vehicle according to one aspect of the present invention.

• • Tatsächliche Fahrparameter 1;
• • Modellparameter 2;
• • Einspurmodell 3;
• • Künstliche Intelligenz 4, diese kann einen sogenannten Long short-term memory LSTM (langes Kurzzeitgedächtnis) umfassen.

1 shows in a block diagram different components of the proposed system or the system arrangement. For example, there are individual devices in the center left arranged, wherein the devices generate the following data or perform the following functions:

• • Actual driving parameters 1 ;
• • Model parameters 2 ;
• • Single track model 3 ;
• • Artificial intelligence 4th , this can include a so-called long short-term memory LSTM (long short-term memory).

A hybrid model is proposed here, as the two lower units 3 and 4th cooperate in such a way that the single-track model 3 is operated, which is based on the component 4th is adjusted. For pattern recognition or for training the corresponding component 4th can the components 1 and / or 2 can be used. So the parameters can be taken from the component alone 2 can be used as a reference basis or this reference data from the component 2 are in turn based on the component 1 evaluated. With the component 1 it is a component that can generate sensor data and thus measures this component 1 Data from a real test drive.

In addition, a complex model can be operated 2, which in turn is used to train the component 4th can apply. The reference data is thus obtained from the component 1 and / or component 2 generated. The reference data can also be available alone as real data, that is to say data that is generated by the component 1 be measured. There is always an output on the right-hand side, which is the case with the hybrid model of the components 3 and 4th corresponds to the prediction data output.

The single-track model can also be adapted with regard to the parameters so that a so-called pre-processor, left side, and a post-processor, right side, are arranged. This is by means of the rectangles to the left and right of the component 3 displayed.

2 shows in a schematic flowchart a method for efficient and precise prediction of driving behavior for use in autonomous driving of a motor vehicle, including generating 100 of reference data which describe an actual driving behavior of the motor vehicle; an operating 101 a single lane model for generating preliminary prediction data of the driving behavior of the motor vehicle; a recognition 102 of deviations of the preliminary prediction data from the reference data by means of artificial intelligence; and a customization 103 the preliminary prediction data depending on the detected 102 Deviations from the generation of prediction data to be output.

The person skilled in the art recognizes here that the steps can have further substeps and in particular that the method steps can each be carried out iteratively and / or in a different order.

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• WO 2007060134 A1 [0002]
• DE 102016218549 B3 [0003]
• EP 3463940 A1 [0004]

Claims (10)

Method for efficient and precise prediction of driving behavior for use in autonomous driving of a motor vehicle, comprising: - Generating (100) reference data which describe an actual driving behavior of the motor vehicle; - Operating (101) a single-track model for generating preliminary prediction data of the driving behavior of the motor vehicle; - Detecting (102) deviations of the preliminary prediction data from the reference data by means of artificial intelligence; and - Adapting (103) the preliminary prediction data as a function of the detected (102) deviations in order to generate prediction data to be output. Procedure according to Claim 1 , characterized in that the reference data are generated by means of a vehicle model and / or reading out sensors. Procedure according to Claim 1 or 2 , characterized in that the reference data are generated by means of a vehicle model which is more complex than the single-track model. Method according to one of the preceding claims, characterized in that the adaptation (103) of the preliminary prediction data takes place by adding the deviations. Method according to one of the preceding claims, characterized in that the prediction data to be output describe driving parameters which are stored for further use. Method according to one of the preceding claims, characterized in that the motor vehicle is controlled automatically using the prediction data to be output. Method according to one of the preceding claims, characterized in that the prediction data to be output are used in the context of a safety function and / or a comfort function. Method according to one of the preceding claims, characterized in that the artificial intelligence is implemented using a neural network and / or a long short-term memory. System arrangement for efficient and accurate prediction of driving behavior for use in autonomous driving of a motor vehicle, comprising: - A test device set up to generate (100) reference data which describe an actual driving behavior of the motor vehicle; - A simulation device set up to operate (101) a single-track model for generating preliminary prediction data of the driving behavior of the motor vehicle; a computing device set up to recognize (102) deviations of the preliminary prediction data from the reference data by means of artificial intelligence; and a control device set up to adapt (103) the preliminary prediction data as a function of the detected (102) deviations in order to generate prediction data to be output. Computer program product with control commands, which the method according to one of the Claims 1 to 8th run when they are run on a computer.

Images