EP4364027A1 - Auslegen eines automatisierungsalgorithmus eines fahrzeugs - Google Patents
Auslegen eines automatisierungsalgorithmus eines fahrzeugsInfo
- Publication number
- EP4364027A1 EP4364027A1 EP22718724.2A EP22718724A EP4364027A1 EP 4364027 A1 EP4364027 A1 EP 4364027A1 EP 22718724 A EP22718724 A EP 22718724A EP 4364027 A1 EP4364027 A1 EP 4364027A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- parameters
- automation algorithm
- algorithm
- optimization
- cost function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W60/00—Drive control systems specially adapted for autonomous road vehicles
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
Definitions
- the invention relates to a method for designing an automation algorithm for a vehicle by simulating the automation algorithm and a virtual scenario, and a simulator device for executing the method.
- Scenario- and simulation-based testing is becoming increasingly important for the development and validation of automated vehicles, especially to take critical scenarios into account.
- Algorithms of the automated vehicle whether they are applications of control technology to control the vehicle status or whether they are higher-level decision-making algorithms, must be designed in such a way that these critical scenarios can also be mastered.
- the challenge here is that it is often not the rigid environment but the behavior of other road users that is the cause of critical scenarios.
- Methods for configuring virtual environments and for designing and training self-learning algorithms are known in the prior art.
- EP 3690 754 A1 relates to a method for designing a traffic scenario in a virtual environment.
- - freely reproduced - in particular optically recorded data from driving tests and a video thereof are used in order to gain data about the environment with a situation analysis and to obtain information about the respective vehicle status, as well as to take traffic information into account.
- the traffic information is fed into a "scenario augmentation network”.
- a traffic simulator is configured based on a critical scenario of the data.
- DE 102017 007 136 A1 also relates to a method for training self-learning algorithms for an automated vehicle with a specified automation module by generating learning situations, the learning situations being generated as follows: Carrying out a traffic simulation in which a virtual ego drive vehicle is placed in a virtual scenario with the automation module of the real vehicle, the scenario comprising a route structure with a specified route, further comprehensively automatically generated further virtual moving objects with individually specifiable object properties and behavior models, the objects in the course of the progressive simulation with each other independently and adaptively on the Based on the respective object properties and behavior models interact; Carrying out a driving dynamics simulation on the basis of the automation module and of virtual sensor signals of the moving objects of a virtual sensor system assigned to the ego vehicle, which correspond to a sensor system of the actually existing vehicle, in which reactions of the ego vehicle are generated; and identification of a relevant learning situation using selection criteria that are determined on the basis of predeterminable metrics.
- the object of the invention is to make the design of an automation algorithm for a vehicle more efficient and effective in order to enable more robust automation algorithms for vehicles.
- a first aspect of the invention relates to a method for designing an automation algorithm of a vehicle by simulating the automation algorithm and a virtual scenario, having the steps:
- a function value of the cost function depends on the parameters of the automation algorithm as the first group of changeable optimization variables and on the parameters of the virtual scenario as the second group of changeable optimization variables and correlates with a quality of execution of the automation algorithm in the virtual scenario
- the automation algorithm of a real vehicle serves - in particular as a part a more complex driving control system - to operate the vehicle in an at least partially automated up to a fully automated mode.
- the automation algorithm can therefore be the basic controller of a driver assistance system in order to influence a vehicle state in a partially automated or fully automated manner.
- a vehicle state is, for example, a speed, an acceleration of the vehicle, a steering angle or a state of another control element of the vehicle such as a handbrake.
- the automation algorithm can also include higher-order control loops of a cascade controller up to, in particular, discrete decision-making algorithms. The latter are typically used to enable highly automated operation up to and including fully autonomous operation of the vehicle.
- the higher the degree of authority of the automation algorithm in its impact on the vehicle the more critical its decisions are in safety-relevant situations.
- the virtual scenario represents what creates the prerequisites and boundary conditions on the basis of which the automation algorithm runs. Since the virtual scenario is part of the overall simulation in which the automation algorithm is also simulated, the behavior of the automation algorithm in the virtual scenario can be analyzed as if real existing physical sensors had an actual situation in front of them instead of the virtual scenario. A corresponding sensor dynamic is also advantageously simulated, but for the sake of simplicity a data set of the virtual scenario can also be transmitted directly to the automation algorithm, with real influences such as sensor noise or other inadequacies in the detection by a sensor being neglected in the latter case.
- the security and repeatability of the simulator can be exploited and any scenarios, particularly critical ones, can be loaded in order to check the behavior of the automation algorithm in response thereto.
- a cost function is used according to the invention, the function value of which correlates with a quality of the execution of the automation algorithm.
- a positive correlation is preferably used, so that the function value increases with a higher quality.
- a negative correlation can also be used by using the inverse sign convention of the cost function. If a positively chosen cost function is to be minimized, can equivalently, the negatively chosen cost function can be maximized and vice versa. This example makes it clear that the signs of the cost function lead to the same result as long as they are consistently taken into account in the further process steps.
- the cost function has a large number of optimization variables that can be freely modified in the simulator. While the first group of optimization variables includes parameters of the automation algorithm, a second group of optimization variables includes parameters of the virtual scenario.
- the term group and parameter can also be understood to mean an individual value in each case, for example a scalar gain in a regulator of the automation algorithm. Optimization variables of the first group are therefore in particular not also part of the second group and optimization variables of the second group are therefore in particular also not part of the first group.
- the value of the cost function can be determined in particular by running a simulation, with the cost function preferably indicating the quality of the execution of the automation algorithm in a physically tangible quantity, for example the time required to carry out emergency braking to a standstill, the lateral acceleration of a vehicle evasive manoeuvres, correct decision-making, or the like; a combination of the above terms can also be used to define the cost function.
- Binary values like the above decision making can also be mapped into a numeric value, for example +100 for the wrong decision and -100 for the correct decision when a decision leaves exactly two options open.
- At least two related optimization methods are carried out:
- a respective starting value for the parameters of the automation algorithm is used in order to adapt the virtual scenario.
- a provisional design can be used for the respective starting value of these parameters.
- the virtual scenario with its parameters is changed in such a way that the value of the cost function changes in such a way that the quality of the execution of the automation algorithm decreases.
- the cost function is therefore used in a first optimization method in such a way that the quality of the execution of the automation algorithm decreases intentionally in order to artificially generate a particularly critical scenario for the automation algorithm.
- the automation algorithm is optimized precisely for this critical scenario, so that the critical scenario can also be reliably managed.
- Agent A the optimization process with regard to the automation algorithm as Agent A and the optimization process with regard to the virtual scenario as Agent B work against each other.
- Agent A's goal is to optimize its mission success given a frozen status of the traffic simulation, and an updated status A_new is created.
- Agent A can only act within certain limits, which are specified, for example, and without restricting the general public, by the German road traffic regulations.
- Agent B's goal is to minimize Agent A's mission success if Agent A's status is frozen.
- An updated status B_new is created. Agent B can only proceed within certain limits, since traffic scenarios that are not too extreme and unrealistic are generated.
- the adversarial traffic planner B can use the parameters of the different levels L work, with a particular focus for Agent B on the following parameters:
- Selected L4 level parameters such as traffic flow, are defined in the mission descriptions and can be varied across different missions.
- the method steps of the execution of the optimization method are repeated iteratively until a predefined convergence criterion of the optimization method, or a predefined robustness of agent A, is reached.
- a convergence criterion is, for example, the remaining of a relative or absolute change in the function value of the cost function within a specified limit due to a change in one or more optimization variables.
- Another possible convergence criterion is reaching a predefined absolute value of the cost function, reaching a physically realistic or otherwise specified limit at least for one of the optimization variables, or the like.
- the method steps of the execution of the optimization methods for different missions of the vehicle are repeated in a virtual scenario.
- a mission of the vehicle essentially indicates the intention, for example a planned left turn at an intersection. By including several possible missions, it is easier to create critical scenarios be identified.
- the method steps of the execution of the optimization method are repeated for different virtual scenarios.
- Different virtual scenarios are, for example, different intersections that the vehicle crosses in the simulation. In this way, a wide range of scenarios that are actually possible can advantageously be analysed.
- one of the following is used as the respective optimization method:
- the parameters of the virtual scenario include at least one of the following:
- the parameters of the automation algorithm include at least one of the following:
- the function value of the cost function correlates - positively or negatively - with at least one of the following terms or with a sum based on the following terms:
- the cost function correlates with a sum of at least two squares of the terms. This can be used advantageously if the cost function can be evaluated analytically. Typically, however, it will lead numerically as a simulation result to the function value.
- limits are specified for the parameters of the virtual scenario and for the parameters of the automation algorithm, within which the parameters and thus the optimization variables can be adjusted by the respective optimization method.
- the limits for the parameters are determined, for example, by a required robustness of a control loop, a maximum permissible amplification of noise, or in relation to the virtual scenario by parameters of a real scenario that can still be expected realistically.
- a further aspect of the invention relates to a simulator device that is set up and designed to carry out the following steps for designing an automation algorithm of a vehicle by simulating the automation algorithm and a virtual scenario:
- a function value of the cost function depends on the parameters of the automation algorithm as the first group of changeable optimization variables and on the parameters of the virtual scenario as the second group of changeable optimization variables and correlates with a quality of execution of the automation algorithm in the virtual scenario
- a cost function is set up S1, the cost function depending on the parameters of the automation algorithm as well as on the parameters of the virtual scenario.
- the functional value of the cost function is made up of a sum of the time required to complete a braking process and the required acceleration of a combined evasive and braking maneuver.
- the individual terms of the cost function do not have to be determined explicitly analytically as optimization variables depending on the parameters mentioned above, since the respective term is determined by the simulation itself. Of the The function value of the cost function is therefore given after execution of the simulation software code and comes from numerical integrations of the simulation.
- the function value of the cost function also decreases as the quality of the execution of the automation algorithm in the virtual scenario increases.
- the second step while keeping the parameters of the automation algorithm constant, it follows the execution S2 of an optimization procedure with an evolution algorithm, which repeatedly executes the simulation under random but convergent variation of the parameters of the virtual scenario in order to find such parameters of the virtual scenario, that the cost function assumes a local maximum accompanied by a low quality of the automation algorithm.
- an optimization method is executed S3 using an evolution algorithm and again repeated simulation for adapting the parameters of the automation algorithm such that the cost function assumes a local minimum accompanied by a high quality of the automation algorithm.
- the adjustment of the parameters of the automation algorithm relates to the weights of artificial neural networks, while the adjustment of the parameters of the virtual scenario relates to geometric framework conditions, a road infrastructure, movement sequences of pedestrians who cross a traffic light in its red phase and thus the vehicle to avoid and/or force brakes.
- the adjustment of the parameters of the virtual scenario relates to geometric framework conditions, a road infrastructure, movement sequences of pedestrians who cross a traffic light in its red phase and thus the vehicle to avoid and/or force brakes.
- these procedural steps of the execution of the optimization method are repeated iteratively for several missions of the vehicle until the predefined convergence criterion of the optimization method is reached, so that the respective optimization method is a barrier of the relative change no longer exceeds.
- the adapted parameters of the automation algorithm are stored S4 in the automation algorithm and the method steps are carried out again for a new intersection, a new virtual scenario.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Mechanical Engineering (AREA)
- Transportation (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102021206737.9A DE102021206737A1 (de) | 2021-06-29 | 2021-06-29 | Auslegen eines Automatisierungsalgorithmus eines Fahrzeugs |
| PCT/EP2022/059163 WO2023274594A1 (de) | 2021-06-29 | 2022-04-06 | Auslegen eines automatisierungsalgorithmus eines fahrzeugs |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP4364027A1 true EP4364027A1 (de) | 2024-05-08 |
Family
ID=81386737
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP22718724.2A Withdrawn EP4364027A1 (de) | 2021-06-29 | 2022-04-06 | Auslegen eines automatisierungsalgorithmus eines fahrzeugs |
Country Status (3)
| Country | Link |
|---|---|
| EP (1) | EP4364027A1 (de) |
| DE (1) | DE102021206737A1 (de) |
| WO (1) | WO2023274594A1 (de) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102023116617A1 (de) | 2023-06-23 | 2024-12-24 | Dr. Ing. H.C. F. Porsche Aktiengesellschaft | Verfahren zur Simulation von Testszenarien |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102017007136A1 (de) | 2017-07-27 | 2019-01-31 | Opel Automobile Gmbh | Verfahren und Vorrichtung zum Trainieren selbstlernender Algorithmen für ein automatisiert fahrbares Fahrzeug |
| US10636295B1 (en) | 2019-01-30 | 2020-04-28 | StradVision, Inc. | Method and device for creating traffic scenario with domain adaptation on virtual driving environment for testing, validating, and training autonomous vehicle |
| DE102019216836A1 (de) * | 2019-10-31 | 2021-05-06 | Psa Automobiles Sa | Verfahren zum Trainieren wenigstens eines Algorithmus für ein Steuergerät eines Kraftfahrzeugs, Computerprogrammprodukt sowie Kraftfahrzeug |
-
2021
- 2021-06-29 DE DE102021206737.9A patent/DE102021206737A1/de active Pending
-
2022
- 2022-04-06 WO PCT/EP2022/059163 patent/WO2023274594A1/de not_active Ceased
- 2022-04-06 EP EP22718724.2A patent/EP4364027A1/de not_active Withdrawn
Also Published As
| Publication number | Publication date |
|---|---|
| WO2023274594A1 (de) | 2023-01-05 |
| DE102021206737A1 (de) | 2022-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3970077B1 (de) | Verfahren zum trainieren wenigstens eines algorithmus für ein steuergerät eines kraftfahrzeugs, computerprogrammprodukt, kraftfahrzeug sowie system | |
| DE102022112059B3 (de) | Verfahren, System und Computerprogrammprodukt zur Kalibrierung und Validierung eines Fahrerassistenzsystems (ADAS) und/oder eines automatisierten Fahrsystems (ADS) | |
| EP3748453B1 (de) | Verfahren und vorrichtung zum automatischen ausführen einer steuerfunktion eines fahrzeugs | |
| DE102017213634A1 (de) | Verfahren und Vorrichtung für die Durchführung von virtuellen Tests in einer virtuellen Realitätsumgebung für ein autonom fahrendes Fahrzeug | |
| EP3942379B1 (de) | Verfahren zum trainieren wenigstens eines algorithmus für ein steuergerät eines kraftfahrzeugs, computerprogrammprodukt, kraftfahrzeug sowie system | |
| DE102019124018B4 (de) | Verfahren zum Optimieren von Tests von Regelsystemen für automatisierte Fahrdynamiksysteme | |
| DE102016109651A1 (de) | Verfahren für die Validierung von Fahrzeugfunktionen eines virtuellen autonomen Fahrzeugs | |
| DE102022132917A1 (de) | Verfahren und System zur Bestimmung der Kritikalität und Kontrollierbarkeit von Szenarien für automatisierte Fahrfunktionen | |
| DE102019134053A1 (de) | Verfahren zur kontinuierlichen Absicherung im Fahrversuch applizierter automatisierter Fahrfunktionen | |
| DE102023200314A1 (de) | Erzeugung maschinenlesbarer Szenariobeschreibungen aus menschlichen Beschreibungen | |
| WO2020249758A1 (de) | Verfahren und generator zum erzeugen von gestörten eingangsdaten für ein neuronales netz | |
| EP3748454A1 (de) | Verfahren und vorrichtung zum automatischen ausführen einer steuerfunktion eines fahrzeugs | |
| DE102022127857A1 (de) | Verfahren, System und Computerprogrammprodukt zur Sensitivitätsanalyse von Applikationsparametern für automatisierte Fahrfunktionen | |
| WO2022184363A1 (de) | Computerimplementiertes verfahren zum trainieren wenigstens eines algorithmus für eine steuereinheit eines kraftfahrzeugs, computerprogrammprodukt, steuereinheit sowie kraftfahrzeug | |
| EP4364027A1 (de) | Auslegen eines automatisierungsalgorithmus eines fahrzeugs | |
| DE102022107338B3 (de) | Verfahren zum Testen von automatisierten Fahrzeugfunktionen | |
| DE102022117841B4 (de) | Verfahren, System und Computerprogrammprodukt zur Kalibrierung und Validierung eines Fahrerassistenzsystems (ADAS) und/oder eines automatisierten Fahrsystems (ADS) unter Berücksichtigung einer subjektiven Bewertung | |
| DE102022116564A1 (de) | Verfahren, System und Computerprogrammprodukt zur Bewertung von Testfällen zum Testen und Trainieren eines Fahrerassistenzsystems (ADAS) und/oder eines automatisierten Fahrsystems (ADS) | |
| DE102020210789A1 (de) | Analyse eines neuronalen Netzes für ein Fahrzeug | |
| WO2021089499A1 (de) | Verfahren und system zum prüfen einer automatisierten fahrfunktion durch reinforcement-learning | |
| EP4544528A1 (de) | Auslegen eines automatischen fahrsteuersystems in massiv parallelen simulationen | |
| DE102023124870A1 (de) | Verfahren und System zur erweiterten automatisierten Kalibrierung von Funktionen eines Fahrerassistenzsystems | |
| EP4125007A1 (de) | Verfahren zum bereitstellen von trainingsdaten zum anlernen eines künstlichen neuronalen netzes, verfahren zum anlernen eines künstlichen neuronalen netzes, computerprogrammprodukt und datenstruktur | |
| DE102019131722B3 (de) | Verfahren zur Steuerung oder Regelung einer Klimatisierung eines Fahrzeugs | |
| WO2020208182A1 (de) | Verfahren zur bestimmung einer belastungsvorhersage für ein bauteil eines kraftfahrzeugs |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20231115 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE APPLICATION HAS BEEN WITHDRAWN |
|
| 18W | Application withdrawn |
Effective date: 20240806 |