System and method for creating scenarios

Abstract

The invention relates to a system for creating scenarios, designed to carry out a method comprising the following steps: a) receiving a textual description of a scenario, b) receiving creation instructions, c) generating a machine-readable description of the scenario by means of an algorithm, the algorithm comprising a trained machine learning model, d) outputting the machine-readable description of the scenario. The invention also relates to a corresponding method for creating scenarios.

Description

[0001] 23-022-EM - 1 - Description Title System and method for creating scenarios The invention relates to a system for creating scenarios according to the preamble of patent claim 1. Furthermore, the invention relates to a computer-implemented method for creating scenarios according to the preamble of patent claim 11. State of the art Scenarios which describe a traffic situation are known from the published German patent application DE 102021100395 A1. These are compared using the claimed method. Against this background, the object of the invention is to specify a system which further develops the state of the art and in particular to specify a method with which new scenarios can be easily generated. This object is achieved by a system for creating scenarios having the features of patent claim 1.The object is further achieved by a computer-implemented method for creating scenarios with the features of patent claim 11. 23-022-EM - 2 - The object is further achieved according to the invention by a computer program with the features of patent claim 14 and a computer-readable data carrier with the features of patent claim 15. Advantageous embodiments of the invention are the subject of dependent subclaims. According to the subject matter of the invention, a system for creating scenarios is set up to carry out a method comprising the following steps: a) receiving a textual description of a scenario, b) receiving creation instructions, c) generating a machine-readable description of the scenario by means of an algorithm, wherein the algorithm comprises a trained machine learning model, d) outputting the machine-readable description of the scenario.According to the subject matter of the invention, a computer-implemented method for creating scenarios is further claimed, comprising the following steps: a) receiving a textual description of a scenario, b) receiving creation instructions, c) generating a machine-readable description of the scenario using an algorithm, wherein the algorithm comprises a trained machine learning model, d) outputting the machine-readable description of the scenario. It should be noted that scenarios within the meaning of the claimed invention can be referred to as abstractions of a traffic situation. They are used or required in the test methodology of "scenario-based testing" of systems and system components for the autonomous control of a motor vehicle. A scenario describes the temporal development of scenes that begin with a start scene. In contrast to scenes, scenarios cover a certain period of time.A single scene can be viewed as a snapshot of the environment, comprising the scenery, dynamic elements, and all actors. The actors, and in particular the vehicles, in a scene. A fundamental distinguishing feature of traffic scenarios is the movements of the road users. This means that differentiation is not based on static parameters such as the environment, buildings, or road width, but rather on the driving behavior of the individual road users. The movements of the road users, and thus the driving behavior, are described by trajectories. Trajectories describe a path in both spatial and temporal directions. By describing the movements / trajectories of a road user, a movement profile can be created. An autonomous driving function is implemented by a system, for example a control unit.The 23-022-EM - 4 - control unit is traditionally tested in a real vehicle under real traffic situations and validated through hardware-in-the-loop tests or, alternatively, through completely virtual tests. Simulation-supported methods, such as scenario-based testing, are required to validate autonomous driving functions. The aim is to achieve the highest possible test coverage according to the driving function and the associated ODD (Operational Design Domain), especially for critical situations that only occur rarely, so-called "corner cases." Such validation procedures are based on scenarios. A scenario consists of a road description, the traffic on it, and, if required, a corresponding 3D environment. There are currently various formats for describing a scenario, such as the ASAM standards OpenRoad and OpenScenario.The description of a scenario takes place on three levels: abstract scenarios, logical scenarios, and concrete scenarios. The present invention therefore provides a very simple and efficient way to automatically generate an executable scenario file based on an informal textual description with keywords or desired content. 23-022-EM - 5 - Furthermore, it is possible to convert abstract scenarios, for example according to the OpenScenario 2 standard, into a large number of executable scenarios, which in turn can be used as the basis for scenario-based validation of an automated driving function of a motor vehicle. An abstract scenario definition is, for example, a combination of elements of an Operational Design Domain (ODD) for defining an area of ​​application for the automated driving function of the motor vehicle.The ODD specifies the area in which the driving function is valid, whereby any conditions for its execution can be defined. As an example, the ODD of a German motorway with two lanes and no construction sites can be seen. Based on this abstract information alone, a pre-trained machine learning model knows that no traffic lights, no stop signs and (typically) no oncoming traffic are to be expected, so that executable scenarios can be created automatically, such as a lane change on the motorway. Using the method according to the invention, it is also possible to create a so-called cut-in scenario, for example. The cut-in scenario can be described as a traffic situation in which a highly automated orAn autonomous vehicle is driving in a predetermined lane and another vehicle, with a lower speed than the ego vehicle, cuts in from another lane into the lane of the ego vehicle at a certain distance. The ego vehicle here refers to the vehicle being tested. The speed of the ego vehicle and the other vehicle, which is also referred to as the fellow vehicle, remains constant. Since the speed of the ego vehicle is higher than that of the fellow vehicle, the ego vehicle must be braked to avoid a collision between the two vehicles. However, a cut-in scenario can also occur in various forms, for example due to a difference in the speed of the road users.Furthermore, a scenario is possible in which the ego vehicle represents the overtaking vehicle, so that the ego vehicle is driving in a predetermined lane and overtaking another vehicle at a lower speed than the ego vehicle. In doing so, the ego vehicle changes / swipes out into another lane and drives past the fellow vehicle at a higher speed. The speed of the ego vehicle does not have to be constant in this scenario. After the ego vehicle has passed the fellow vehicle, the ego vehicle returns to the previous lane. An advantage of the device according to the invention is that the system for creating scenarios can be created very easily and with little knowledge of the formal description of scenarios.This allows the need for scenarios with certain properties necessary for the development of autonomous driving functions and corresponding control units to be created very quickly. In an alternative embodiment, the system is characterized in that the creation instructions specify a version of the machine-readable description of the scenario. One development is a system wherein the machine learning model at least partially has a transformer architecture. In one embodiment, it is preferred that the system is characterized in that the machine learning model has been provided and trained for processing natural language instructions. According to another development, the system is characterized in that the machine learning model has been trained with unsupervised learning and is in particular a large language model.In one development, the system further comprises a test system, wherein a test is carried out using the test system and the machine-readable description of the scenario. In another embodiment of the system, the test tests a control unit or a function of a control unit. 23-022-EM - 8 - According to a preferred development, the system is characterized in that the control unit or the function of the control unit is provided for controlling a highly automated or autonomous vehicle. According to another development, the system is characterized in that the system further comprises a simulation system, and a simulation is carried out using the simulation system and the machine-readable description of the scenario.According to a further preferred development of the invention, the textual description of the scenario is based on elements, in particular a combination of elements, of an operational design domain for defining an area of ​​application of an automated driving function of a motor vehicle. This advantageously allows for the automated creation of scenarios covering an operational design domain. According to a further preferred development of the invention, the algorithm is configured to generate a machine-readable description of at least one, in particular critical, scenario that leads to a predetermined event, in particular a vehicle collision, with a definable minimum probability. 23-022-EM - 9 - The minimum probability can be defined, for example, using at least one threshold value.According to a further preferred development of the invention, the minimum probability that the scenario will lead to a predetermined event is given by a key performance indicator included in the textual description of the scenario, in particular a time duration (TTC) until a vehicle collision. The predetermined event can be a vehicle collision or, alternatively, an environmental condition that makes driving more difficult. Examples include backlighting, precipitation, fog, night-time driving, etc. The key performance indicator is, for example, a numerical value that represents a specific condition, e.g., the environmental condition or the TTC value. The features of the system for creating scenarios described herein are equally applicable to the computer-implemented method for creating scenarios, and vice versa.The invention is explained in more detail below with reference to the drawings. Similar parts are labeled with identical designations. The illustrated embodiments are highly schematic 23-022-EM - 10 - and do not show all the details of the system necessary to implement the invention. These show: Fig. 1 shows a schematic structure of the system in an embodiment according to the invention; Fig. 2 shows a flowchart of a computer-implemented method for creating scenarios according to the preferred embodiment of the invention; and Fig. 3 shows a schematic representation of a turning scenario of a virtual test according to the preferred embodiment of the invention.Figure 1 shows a view of an embodiment comprising a textual description of the scenario (functional description) 10, creation instructions 20 containing additional generic, scenario-specific "instructions" for the algorithm 30, comprising a trained machine learning model. Furthermore, a machine-readable scenario description 40 (e.g., OpenSCENARIO versions 1 or 2) and a test system 50 for testing autonomous driving functions or corresponding control units. The system is further configured such that the textual description of the scenario is based on elements, in particular a combination of elements, of an operational design 23-022-EM - 11 - domain for defining an area of ​​application of an automated driving function of a motor vehicle. Figure 2 shows a flowchart of a computer-implemented method for creating scenarios according to the preferred embodiment of the invention.The method comprises receiving a textual description of a scenario and receiving creation instructions. Furthermore, the method comprises generating a machine-readable description of the scenario using an algorithm, wherein the algorithm comprises a trained machine learning model. The method further comprises outputting the machine-readable description of the scenario. The algorithm is configured to generate a machine-readable description of at least one, in particular critical, scenario that leads to a predetermined event, in particular a vehicle collision, with a definable minimum probability. The minimum probability that the scenario leads to a predetermined event is given by a key performance indicator included in the textual description of the scenario, in particular a time duration (TTC) until a vehicle collision. 23-022-EM - 12 - Fig.3 shows a schematic representation of a turning scenario of a virtual test according to the preferred embodiment of the invention. In order to improve the results of the algorithm's output, fine-tuning can be carried out, whereby the model, in particular a language model, is further trained with a larger number of functional scenario files or the associated documentation. In addition to the pure description of traffic, the approach can also generate complete road descriptions, e.g., according to the ASAM OpenDrive standard. Such a trained language model thus delivers different outputs based on a slightly modified input, which can be used to quickly generate randomized driving scenarios. The model can therefore also be adapted to critical driving situations, for example, in order to increase test coverage as efficiently as possible.A critical scenario exists, for example, when the proportion of possibilities leading to a crash situation or vehicle collision is high from a certain point in the scenario progression. For example, a critical scenario exists when an EGO vehicle 70 is turning in city traffic if almost no trajectories 60 can be found to prevent a crash with oncoming traffic, e.g. a FELLOW vehicle 80. 23-022-EM - 13 - A machine learning model can thus be trained to create only those executable scenarios from abstract scenarios that have a high proportion of possibilities leading to a crash. Precisely these critical scenarios are then used to validate an automated driving function. A critical scenario can also be viewed as a traffic situation that rarely occurs in reality, but if it does, would lead to a crash situation.For example, a KPI (Key Performance Indicator) can be defined for criticality, which indicates the probability with which a crash situation occurs during the course of the scenario. The machine learning model can be trained with this criticality measure, for example, using a backpropagation approach. The process chain is as follows: description of the scenario, e.g. turning, application of the machine learning model, executable scenario with high criticality. Although specific embodiments have been illustrated and described herein, it will be understood by those skilled in the art that numerous alternative and / or equivalent implementations exist. It should be noted that the exemplary embodiment or exemplary embodiments are only examples and are not intended to limit the scope, applicability, or configuration in any way.Rather, the foregoing summary and detailed description provide one skilled in the art with a convenient road map for implementing at least one exemplary embodiment, it being understood that various changes in functionality and arrangement of elements may be made without departing from the scope of the appended claims and their legal equivalents. Generally, this application is intended to cover modifications, adaptations, or variations of the embodiments presented herein. For example, an order of method steps may be altered. Furthermore, the method may be performed, at least in part, sequentially or in parallel.

[0002] 23-022-EM DSP051S - 15 - List of Reference Symbols 10 Scenario 20 Creation Instructions 30 Algorithm 40 Scenario Description 50 Test System 60 Trajectories 70 EGO Vehicle 80 FELLOW Vehicle S1-S4 Process Steps

Claims

23-022-EM DSP051S - 16 - Claims 1. A system for creating scenarios configured to execute a method comprising the following steps: a) receiving a textual description of a scenario, b) receiving creation instructions, c) generating a machine-readable description of the scenario using an algorithm, wherein the algorithm comprises a trained machine learning model, d) outputting the machine-readable description of the scenario.

2. The system according to claim 1, characterized in that the creation instructions specify a version of the machine-readable description of the scenario.

3. The system according to claim 1 or claim 2, characterized in that the machine learning model at least partially comprises a transformer architecture. 4.System according to one of the preceding claims, characterized in that the machine learning model has been provided and trained for processing natural language instructions.

5. System according to claim 4, characterized in that the machine learning model is provided with. -022-EM DSP051S - 17 - has been trained using unsupervised learning and is, in particular, a large language model.

6. System according to one of the preceding claims, characterized in that the system further comprises a test system, and a test is carried out using the test system and the machine-readable description of the scenario.

7. System according to claim 6, characterized in that the test tests a control unit or a function of a control unit.

8. System according to claim 7, characterized in that the control unit or the function of the control unit is intended for controlling a highly automated or autonomous vehicle.

9. System according to one of the preceding claims, characterized in that the system further comprises a simulation system, and a simulation is carried out using the simulation system and the machine-readable description of the scenario. 10.System according to one of the preceding claims, characterized in that the textual description of the scenario is based on elements, in particular a combination of elements, of an operational design domain for defining an area of ​​application of an automated driving function of a motor vehicle. -022-EM DSP051S - 18 - 11. A computer-implemented method for creating scenarios, comprising the following steps: a) receiving (S1) a textual description of a scenario, b) receiving (S2) creation instructions, c) generating (S3) a machine-readable description of the scenario using an algorithm, wherein the algorithm comprises a trained machine learning model, d) outputting (S4) the machine-readable description of the scenario.

12. The computer-implemented method according to claim 11, wherein the algorithm is configured to generate a machine-readable description of at least one, in particular critical, scenario that leads to a predetermined event, in particular a vehicle collision, with a definable minimum probability. 13.A computer-implemented method according to claim 12, wherein the minimum probability that the scenario leads to a predetermined event is given by a key performance indicator included in the textual description of the scenario, in particular a time duration (TTC) until a vehicle collision.

14. A computer program product comprising a computer program comprising software means for carrying out a method according to claim 11. -022-EM DSP051S - 19 - wherein the computer program is executed on a computer.

15. A computer-readable data carrier with program code of a computer program for executing at least parts of a method according to claim 11 when the computer program is executed on a computer.

Images