Method and system for improving the automated calibration of functions of a driver assistance system

Abstract

Method for the automated calibration of functions of a driver assistance system (Advanced Driver Assistance System ADAS / Automated Driving System, ADS) for a variety of scenarios and vehicle configurations, comprising: - Training (S10) a learning reinforcement agent (410) of an optimization module (400) with a training data set (240), wherein the optimization module (400) comprises at least the learning reinforcement agent (410), an action module (420) with an action space, an environment module (430), a state module (440) and a reward module (450), and wherein the training data set (240) contains calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) of an ADAS / ADS system (10); - generating (S20) a strategy of the learning reinforcement agent (410) for varying calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) in the form of actions (A i) for optimising calibration data sets (15) of ADAS / ADS systems (10); - Entering (S30) a calibration data set (15) of the ADAS / ADS system (10) with calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) via a user interface (220) of an input module (200) or from a database (300); - Selecting (S40) a large number of test cases (T i ) from the database (300) and generating simulation environments with simulation scenarios from the test cases (T i ) in the environment module (430); - performing (S50) simulations of the behavior of the ADAS / ADS system (10) in the simulation scenarios and generating simulation results, wherein the simulation results reflect the system behavior of the ADAS / ADS system (10) in the simulation scenarios; - Evaluate (S60) the simulation results in the form of key performance indicators (KPI i ) and performance values; - Determining (S70) states (S i ) from the key performance indicators (KPI i ) and rewards (R i ) from the performance values; - Calculate (S80) modeled calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) to achieve an improved system behavior of the ADAS / ADS system (10) in the simulation scenarios by the trained learning reinforcement agent (410) in several iterations until an optimized calibration data set (17) has been generated, wherein to determine the decisions of the learning reinforcement agent (410) regarding the actions to be carried out (A i ) the action space of the action module (420) into discrete parameter step sizes with which a certain parameter (P i) during which the optimization is adapted or updated, is divided such that, in addition to a first parameter step size, which is predetermined by the discretization of the action space, a second parameter step size is added, the amount of which is smaller than the amount of the first parameter step size, and wherein the learning reinforcement agent (410) determines the calibration parameter value (PVcal i ) of a calibration parameter to be varied (Pcal i ) by one or more parameter increments, such that the learning reinforcement agent (410) decides whether to use a value of a calibration parameter (Pcal i ) by a certain number of the given first step size of the action area and by a certain number of the smaller second step size by adding, subtracting or leaving the first and second step sizes unchanged; - generating and outputting (S90) output results (550) by an output module (500), wherein the output results (550) contain the optimized calibration data set (17).

Description

[0001] The invention relates to a method, a system and a computer program product for improving the automated calibration of functions of a driver assistance system (Advanced Driver Assistance System ADAS / Automated Driving System, ADS).

[0002] Modern advanced driver assistance systems (ADAS) and automated driving systems (ADS) encompass a range of technologies designed to improve vehicle safety and driving comfort. ADAS / ADS systems utilize various sensors, cameras, and algorithms to analyze the vehicle's surroundings and provide real-time assistance to the driver. To this end, the ADAS / ADS system incorporates a wide range of functions that are continually being developed to further improve vehicle safety and driving comfort.

[0003] A typical function is adaptive cruise control (ACC), which uses radar or laser sensors to maintain a safe following distance from the vehicle in front. It automatically adjusts the vehicle's speed to the flow of traffic, eliminating the need for the driver to constantly adjust their speed. Another function is lane departure warning (LDW), which uses cameras or sensors to detect unintentional lane departures. The driver is warned by visual, audible, or haptic signals and prompted to correct their lane. Another function is automatic emergency braking (AEB), which automatically applies the brakes to avoid a collision with a vehicle or pedestrian.

[0004] The Society of Automotive Engineers (SAE) International has classified the capabilities of automated driving into six levels, from Level 0 (no automation) to Level 5 (full automation). Level 2 represents a higher degree of automation than the lower levels but is still referred to as partial automation. Level 3 includes automated passenger transport under certain conditions. Level 4 encompasses fully autonomous driving under certain conditions.

[0005] A particular challenge in the development of driver assistance systems is the integration and testing process, which includes the verification, validation, and calibration of various hardware and software versions prior to release. This involves the question of how to demonstrate the correct functionality of ADAS / ADS systems in all scenarios within the operating conditions – the so-called Operational Design Domain (ODD). As the performance of ADAS / ADS systems increases, the number of scenarios that must be recognized and correctly processed grows to a level that can hardly be described in a specification such as for conventional systems. This characteristic of ADAS / ADS systems is referred to as functional complexity and is the reason why conventional testing methods are only of limited use, especially for systems above SAE Level 3.

[0006] The functional complexity of ADAS / ADS systems poses a particular challenge for the calibration process. Calibration is the modification of parameters in a data set to adapt the ADAS / ADS system to a desired behavior without changing the software code. Efficient calibration can establish a natural driving behavior of the vehicle, which increases customer acceptance of ADAS / ADS systems. Furthermore, the safety and reliability of an ADAS / ADS system, and thus road safety, can be improved.

[0007] ADAS / ADS system datasets play a critical role in developing, validating, and calibrating algorithms, training machine learning models, and evaluating system performance. They provide the fundamental information the ADAS / ADS system needs to perceive its environment, make decisions, and optimally support the driver.

[0008] When calibrating ADAS / ADS systems, some parameters of the dataset can be derived from the vehicle configuration and system specification, while others must be adjusted to achieve optimal system behavior across the entire ODD. However, in real-world driving tests on the road or at a proving ground, only small subsets of scenarios from the ODD can be tested with limited reproducibility. However, repeated and reproducible testing of the same scenarios with different datasets is necessary to optimize the calibration parameters. Furthermore, some parameter changes only occur in certain scenarios, so they are often ignored. The problem of functional complexity thus affects not only the validation process but also the calibration process.

[0009] Furthermore, the multitude of different driving scenarios and driving tasks considered in the development of ADAS / ADS systems increases the number of calibration parameters that can influence the system behavior of an ADAS / ADS system within these scenarios. Some of these calibration parameters can even influence each other, especially if the ADAS / ADS system is divided into subsystems running on different ECUs. However, the interactions of parameter changes are difficult to predict with a large number of scenarios.

[0010] While conventional systems and simple ADAS systems can be calibrated on the test track or in road traffic using a trial-and-error approach, this is hardly possible with complex ADAS / ADS systems at a reasonable effort and cost. Therefore, new approaches to calibrating ADAS / ADS systems are required.

[0011] CN 1 14 715 119 A discloses a method for real-time trajectory planning in which a neural network trained by reinforcement learning is used for decision-making. Discrete parameters are used for training.

[0012] WO 2022 / 090800 A1 discloses a method for autonomous control of a vehicle using trained systems, wherein reinforcement learning is used for training.

[0013] US 2021 / 0 263 526 A1 discloses a method for supporting maneuver planning for an automated driving vehicle, wherein a state space of an environment of the vehicle is described by a Markov decision process and optimal action values ​​for individual actions for discrete states in the state space are determined by means of dynamic programming before the application of a learning reinforcement agent.

[0014] DE 10 2022 104 313 A1 discloses a method for calibrating an electric drive system.

[0015] DE 10 2021 110 083 A1 discloses further prior art.

[0016] The invention is based on the object of creating possibilities for an improved and more precise calibration of an ADAS / ADS system in order to increase the reliability and accuracy of the calibration and thus the driving comfort and road safety of the ADAS system.

[0017] This object is achieved according to the invention with respect to a method by the features of patent claim 1, with respect to a system by the features of patent claim 8, with respect to an ADAS system by the features of patent claim 9, and with respect to a computer program product by the features of patent claim 10. The further claims relate to preferred embodiments of the invention.

[0018] For the calibration of an ADAS / ADS system, the invention uses a virtual simulation method based on reinforcement learning methods. This allows for reliable calibration results and shortens the duration of the calibration process.

[0019] With the present invention, the calibration of a calibration data set of an ADAS / ADS system can be performed efficiently and reliably using a virtual approach with a learning algorithm. The calibration can be performed using a large number of test cases that define a variety of simulation scenarios, and the trained learning algorithm leads to stable and valid calibration results. Since the simulation scenarios generated by the test cases are reproducible, the virtual calibration is not dependent on the random occurrence of relevant scenarios in a real environment. By generating virtual scenarios that simulate real driving conditions, the performance of the driver assistance system can be tested and analyzed. Performance indicators such as detection accuracy, reaction time, decision-making, and system behavior can be measured and evaluated.According to the invention, they are used to optimize calibration parameters and calibration parameter values.

[0020] The learning algorithm offers a cost-effective way to iterate and refine the system behavior of the ADAS / ADS system to generate an optimized configuration of a calibration dataset. Using a discrete, multi-stage approach, in which the action space of the action module is divided into discrete parameter steps and the learning algorithm is provided with a predefined framework for its decisions, an optimized calibration dataset can be determined more quickly, as every conceivable variation no longer needs to be run through, thus allowing a more targeted solution to be found.

[0021] This allows calibration to be performed quickly and efficiently in a wide variety of different and complex driving scenarios that are difficult to reproduce in reality. This allows the calibration parameters of an ADAS / ADS system to be tested under various conditions, such as heavy traffic or rare events. By simulating a wide range of scenarios, potential weak points and edge cases can be uncovered that may not occur in real-world testing. Furthermore, virtual calibration offers significant cost and time savings. Instead of conducting extensive real-world tests, which can be expensive and time-consuming, virtual simulations enable rapid and repeated testing in a controlled environment. A wide variety of scenarios can be simulated, allowing calibration to be improved more efficiently.

[0022] According to a first aspect, the invention provides a method for the automated calibration of functions of a driver assistance system (Advanced Driver Assistance System, ADAS / Automated Driving System, ADS) for a variety of scenarios and vehicle configurations. The method comprises the following steps: - Training a learning reinforcement agent of an optimization module with a training data set, wherein the optimization module comprises at least the learning reinforcement agent, an action module with an action space, an environment module, a state module, and a reward module, wherein the training data set contains calibration parameters and calibration parameter values ​​of an ADAS / ADS system; - Generating a strategy of the learning reinforcement agent for varying calibration parameters and calibration parameter values ​​in the form of actions for optimizing calibration data sets of ADAS / ADS systems; - Entering a calibration data set of the ADAS / ADS system with calibration parameters and calibration parameter values ​​via a user interface of an input module or from a database; - Selecting a large number of test cases from the database and creating simulation environments with simulation scenarios from the test cases; - Performing simulations of the behavior of the ADAS / ADS system in the simulation scenarios and generating simulation results, whereby the simulation results reflect the system behavior of the ADAS / ADS system in the simulation scenarios; - Evaluating the simulation results in the form of performance indicators and performance values; - Determining states from performance indicators and rewards from performance values; - Calculating modeled calibration parameters and calibration parameter values ​​to achieve improved system behavior of the ADAS / ADS system in the simulation scenarios by the trained learning reinforcement agent in several iterations until an optimized calibration data set has been generated, wherein, to determine the decisions of the learning reinforcement agent regarding the actions to be performed, the action space of the action module is divided into discrete parameter step sizes with which a specific parameter is adjusted or updated during the optimization, such that in addition to a first parameter step size predetermined by the discretization of the action space, a second parameter step size is added, the amount of which is smaller than the amount of the first parameter step size, and wherein the learning reinforcement agent reduces the calibration parameter value of a calibration parameter to be varied by one or more parameter step sizes,keeps constant or increases, such that the learning reinforcement agent decides whether to change a value of a calibration parameter by a certain number of the specified first step size of the action space and by a certain number of the smaller second step size by adding, subtracting or leaving the first and second step size unchanged, respectively; - Generating and outputting output results by an output module, wherein the output results contain the optimized calibration data set.

[0023] In a further development, it is provided that a state of the ADAS / ADS system is determined by the state module and includes calibration parameters and calibration parameter values.

[0024] In an advantageous embodiment, it is provided that the software application of the environment module uses at least one neural network, in particular recurrent neural networks or convolutional neural networks, as algorithm.

[0025] In a further embodiment, it is provided that the reinforcement learning algorithm of the optimization module is designed as a Markov decision process or as Temporal Difference Learning (TD-Learning) or as Q-Learning or as SARSA or as Monte Carlo simulation or as Actor-Critic or as Proximal Policy Optimization.

[0026] In particular, a calibration parameter is a physical quantity, a controller parameter, a chemical quantity, a torque, a rotational speed, a voltage, a current, an acceleration, a speed, a braking value, a specific sensor, a measured value of a sensor, a resolution of a sensor, an installation position of a sensor, a technical specification of a sensor, a direction, an angle, a radius, a vehicle property such as engine power, a material, a shape, a specific fuel consumption, a load and / or a specific vehicle configuration, wherein a sensor used is an image-capturing 2D / 3D camera in the visible, IR and / or UV range, a radar system with one or more radar sensors, a LIDAR system for optical distance and speed measurement, an ultrasonic sensor and / or a GPS system.

[0027] Advantageously, it is provided that a calculation function and / or an action based on a policy for the state is selected for modifying at least one calibration parameter and / or a calibration parameter value of the calibration data set of the ADAS / ADS system by the learning reinforcement agent; wherein a new state is calculated by the environment module based on the modified value for the calibration parameter and / or the calibration parameter value; wherein the new state is compared with an old state and a deviation for a comparison result is calculated in the state module; and wherein a reward is determined by the reward module for the comparison result.

[0028] In a further training, the output results are intended to include design plans, images and graphics, video sequences, acoustic sequences, holograms, key figures, diagrams, mathematical formulas and / or PowerPoint presentations, etc.

[0029] In the second aspect, the invention provides a system for the automated calibration of functions of a driver assistance system (Advanced Driver Assistance System, ADAS / Automated Driving System, ADS) for a variety of scenarios and vehicle configurations. The system comprises an input module, a database, an optimization module, and an output module, wherein the optimization module comprises at least one learning reinforcement agent, an action module with an action space, an environment module, a state module, and a reward module. The system is configured to carry out the method according to the first aspect.

[0030] According to a third aspect, the invention provides an ADAS system for a vehicle, wherein the method according to the first aspect is used for calibrating a calibration data set of the ADAS / ADS system.

[0031] According to a fourth aspect, the invention provides a computer program product comprising executable program code configured to carry out the method according to the first aspect when executed.

[0032] The invention is explained in more detail below with reference to embodiments shown in the drawing.

[0033] It shows: Fig. 1 is a block diagram illustrating an embodiment of a system according to the invention; Fig. 2 a flow chart explaining the individual method steps of a method according to the invention; Fig. 3 is a block diagram of a computer program product according to an embodiment of the third aspect of the invention.

[0034] Further features, aspects and advantages of the invention and its embodiments are explained in the following description in conjunction with the patent claims.

[0035] The system 100 according to the invention is based on reinforcement learning methods and is Fig. 1. The system 100 comprises an ADAS / ADS system 10, an input module 200, a database 300, an optimization module 400, and an output module 500. The optimization module 400 comprises a learning reinforcement agent (LV agent) 410, an action module 420 with an action space, an environment module 430, a state module 440, and a reward module 450. The environment module 430 comprises a software application 435 for performing a simulation of the system behavior of the ADAS / ADS system 10 in a specific scenario SZc.i .

[0036] The input module 200, the database 300, the optimization module 400, and the output module 500 can be configured as independent computing units or integrated into a computing unit. In particular, they can each be provided with a processor and / or a memory unit.

[0037] In the context of the invention, a “processor” can be, for example, a machine or an electronic circuit. A processor can in particular be a main processor (Central Processing Unit, CPU), a microprocessor or a microcontroller, e.g. an application-specific integrated circuit or a digital signal processor, optionally in combination with a memory unit for storing program instructions. A processor can also be a virtualized processor, a virtual machine or a soft CPU. It can also be, for example, a programmable processor which is equipped with configuration steps for carrying out the aforementioned method according to the invention or is configured with configuration steps such that the programmable processor implements the inventive features of the method, the entity, the modules or other aspects and / or sub-aspects of the invention.In particular, the processor can contain highly parallel computing units and powerful graphics modules.

[0038] In the context of the invention, a "storage unit" or "storage module" and the like can be understood as, for example, a volatile memory in the form of random access memory (RAM), a permanent memory such as a hard drive or a data storage device, or, for example, a removable storage module. The storage module can also be a cloud storage solution.

[0039] In the context of the invention, a "module" can be understood, for example, as a processor and / or a memory unit for storing program instructions. For example, the processor is specifically configured to execute the program instructions in such a way that the processor and / or the control unit performs functions to execute or implement the method according to the invention or a step of the method according to the invention.

[0040] In the context of the invention, “data” refers to both raw data and already processed data from measurement results from sensors and other data sources.

[0041] The term “database” refers to both a storage algorithm and the hardware in the form of a storage unit.

[0042] In particular, the database 300 and the optimization module 400 can be integrated into a cloud computing infrastructure. This allows for a fast calculation result, as cloud-based solutions offer the advantage of high and thus fast computing power. The term cloud computing infrastructure refers to the entirety of the hardware, software, network, and storage resources used to support cloud computing services. The cloud computing infrastructure provides a platform for providing a wide range of services, including data storage and management, application development and deployment, and virtual computing resources. It includes services for providing computing resources such as computing power and storage to support cloud-based applications.Security technologies such as firewalls, encryption, and identity and access management tools are used to protect data and applications from unauthorized access. A cloud computing infrastructure can significantly increase computing power and data processing speed.

[0043] Furthermore, communication connections for the exchange and transmission of data between the individual modules can be provided, which are designed in particular as a wireless communication connection, e.g. as a mobile radio connection, and / or as a near-field communication connection, e.g. Bluetooth®, Ethernet, NFC (near field communication) or Wi-Fi®.

[0044] The calibration of a specific ADAS / ADS system 10 according to the invention is carried out on the basis of parameterized scenarios SZp i , which are also called logical scenarios. As a parameterized scenario SZp iIn the context of the invention, a scenario is referred to that is written in machine-readable code. A parameterized scenario SZp i is determined by various scenario parameters P1, P2, ..., P n from a set of possible scenario parameters P i and associated scenario parameter values ​​PV1, PV2, ..., PV n from a set of possible scenario parameter values ​​PV i defined. Scenario parameter P i characterize properties and characteristics of a parameterized scenario SZp i . A scenario parameter value PV i sets the value range or a specific value for a scenario parameter P i firmly.

[0045] For example, a scenario parameter P i a parameterized scenario SZp i the speed of an ego vehicle. For this scenario parameter “Speed”, the value range of the corresponding scenario parameter value PV icover the range from 100 km / h to 180 km / h or a specific value, e.g. 120 km / h, can be specified. For another parameterized scenario SZp k the value range of the parameter value PV i For the scenario parameter “speed”, however, the range is from 40 km / h to 70 km / h.

[0046] Additional scenario parameters P iare, for example, a physical quantity, a chemical quantity, a torque, a speed, a voltage, a current, an acceleration, a speed, a braking value, a direction, an angle, a radius, a location, a number, a moving object such as a motor vehicle, a person or a cyclist, an immovable object such as a building or a tree, a road configuration such as a motorway, a traffic light, a tunnel, a roundabout, a turning lane, a traffic volume, a traffic sign, a topographical structure such as a gradient, a time of day, a temperature, a precipitation value and / or a season.

[0047] Within the scope of the present invention, a distinction is made between a parameterized scenario SZp and a concrete scenario SZc. A parameterized scenario SZp is defined within the scope of this invention as a scenario in which both the scenario parameters P ias well as the associated scenario parameter values ​​PV i not all are specified. A concrete scenario SZc is a scenario SZ in which the concrete scenario parameters Pc i and the corresponding concrete scenario parameter values ​​PVc i or value ranges of the concrete scenario parameter values ​​PVc i are defined. Both scenarios, the parameterized scenario SZp and the concrete scenario SZc, are scenarios written in machine-readable code.

[0048] For the creation of parameterized scenarios SZp iVarious sources can be used, such as requirement specifications, expert knowledge, and / or measurements taken on public roads or on test sites using sensors. The sensors used can include, in particular, image-capturing 2D / 3D cameras in the visible, IR, and / or UV ranges, radar systems with one or more radar sensors, LIDAR systems for optical distance and speed measurement, ultrasonic sensors, and / or GPS systems.

[0049] These parameterized scenarios SZp generated from various data sources i and scenario parameter P i , Scenario parameter values ​​PV i , concrete scenarios SZc i , concrete scenario parameters Pc i , concrete scenario parameter values ​​PVc i and, if necessary, further information is stored in the database 300.

[0050] For a test case T iIt is a matter of selecting a suitable parameterized scenario SZp i , concrete scenario parameter Pc i and specific scenario parameter values ​​PVc i and, if necessary, further information for simulating the behavior of the ADAS / ADS system 10 in a specific driving situation or for handling a specific driving task. An example simulation task is the simulation of an overtaking maneuver on a highway (cut-in scenario). According to the invention, the behavior of the ADAS / ADS system 10 is simulated in the environment module 430.

[0051] In particular, a test agent is provided that executes a plurality of test cases T i using a test strategy and stores it in the database 300. This way, concrete scenario parameters Pc i selected depending on an application case and scenario parameter values ​​PVc i determined, while other concrete scenario parameters Pci be kept constant. The data for a test case T i can contain a test identification number (Test ID), a user name, a creation time and the name of the selected parameterized scenario Szp i In addition, the specific scenario parameters Pc i stored that the parameterized scenario Szp i into a concrete scenario Szc i and thus into a test case T i transform. The test agent can be integrated in the input module 200 or the database 300 and includes various algorithms used to create the test cases T i suitable, such as classification and extraction algorithms.

[0052] At least a selection of the test cases stored in the database 300 T i is passed to the optimization module 400 for the creation of simulation environments in the environment module 430.

[0053] The ADAS / ADS system 10 comprises a calibration data set 15 with calibration parameters Pcal i . The respective selection of the calibration parameters Pcal i and the associated parameter values ​​PVcal i The calibration parameters determine the specific calibration data set 15 of the ADAS / ADS system 10. In addition, an ADAS / ADS system 10 has additional algorithms, particularly extraction and classification algorithms, to extract and classify key features from the sensor data acquired by the vehicle sensors. Artificial intelligence algorithms such as neural networks are particularly used in the extraction and classification algorithms. Furthermore, a multitude of other algorithms are required to implement the various functions of the ADAS / ADS system 10.

[0054] Calibration parameter Pcal iare, for example, a physical quantity, a controller parameter, a chemical quantity, a torque, a speed, a voltage, a current, an acceleration, a speed, a braking value, a specific sensor, a measured value of a sensor, a resolution of a sensor, an installation position of a sensor, a technical specification of a sensor, a direction, an angle, a radius, a vehicle property such as engine power, a material, a shape, a specific fuel consumption, a load and / or a specific vehicle configuration.

[0055] In particular, the parameters of a vehicle's sensor system are important for the functionality of an ADAS / ADS system 10, as an ADAS / ADS system 10 uses various sensors to perceive and record the environment, thus creating the basis for the execution of the various functions of the ADAS / ADS system 10. The sensors used by an ADAS / ADS system 10 include cameras for capturing visual information, including images and videos, to enable a visual understanding of the environment. They are used for tasks such as object detection, lane detection, traffic sign recognition, and pedestrian detection. Radar sensors use radio waves to detect objects and measure their distance, relative speed, and angle.They are suitable for detecting and tracking objects, estimating their speed, and providing information for adaptive cruise control, forward collision warning, and blind spot detection systems. Lidar sensors emit laser beams and measure the time it takes the beams to return after hitting objects. This data can be used to create detailed 3D maps of the surroundings, enabling precise object detection, localization, and tracking. Lidar is particularly valuable for tasks such as pedestrian detection, object avoidance, and environmental mapping. Ultrasonic sensors use sound waves to detect proximity and measure distance. They are often used for parking assistance, close-range object detection, and low-speed maneuvering.Each sensor type has its strengths and limitations, and combining information from multiple sensor types helps improve the overall perception and reliability of the ADAS / ADS system 10.

[0056] Calibration of vehicle sensors is therefore important for the reliable operation of an ADAS / ADS system 10 for several reasons. Calibration ensures that the sensors provide accurate and precise readings. It takes into account potential misalignments, distortions, or systematic errors in the sensor readings. Furthermore, an ADAS / ADS system 10 uses sensor fusion, which combines data from multiple sensors to improve perception and understanding of the environment. Calibration ensures that the data from the various sensors is correctly aligned and synchronized to create a coherent picture of the environment. Accurate sensor measurements help detect objects, estimate their speed and trajectory, and make informed decisions or generate appropriate responses.Calibrated sensors contribute to accurate perception of the environment and enable the ADAS / ADS system 10 to make informed decisions and effectively support the driver. They thus form the basis for the high reliability and performance of the ADAS / ADS system 10.

[0057] According to the invention, the calibration parameters Pcal i and calibration parameter values ​​PVcal i by the optimization module 400 for a specific ADAS / ADS system 10 (System under Test, SuT). The optimization module 400 is based on reinforcement learning. Here, the reinforcement learning agent 410 selects for a specific state S i ∈ S from a set of available states at least one action A i ∈ A from a set of available actions. For the actions A i According to the invention, it is the selection of one or more calibration parameters Pcal iand / or one or more calibration parameter values ​​PVcal i . For the selected action A i the LV agent 410 receives a reward that can be represented positively, neutrally, or negatively as an integer or real number, preferably in a range from -1 to + 1. The states S i receives the LV agent 410 from the state module 440, which stores the states S i derived from the simulation results of the environment module 430. A state S i ∈ S is determined by the selection of certain calibration parameters Pcal i , calibration parameter values ​​PVcal i and the mean values ​​of performance indicators KPI i of the calibration data set 15 of the ADAS / ADS system 10. In addition, further data from the various sensors and / or from the database 300 can be processed. Thus, a specific configuration of the calibration data set 15 is assigned a state S i which the LV agent 410 can access.

[0058] The choice of the selected action A i is based on a strategy or policy. For the selected action A i for the state S i Agent 410 receives a reward R i ∈ R from the reward module 450. The strategy, in turn, changes because it is based on the received rewards R i The strategy determines which action A i ∈ A from the set of available actions for a given state S i ∈ S is to be selected from the set of available states. This creates a new state S i+1 generated, for which the LV agent 410 receives a reward R i A strategy thus defines the assignment between a state S i and an action A i and thus specifies which action to be performed A i for a state S iThe goal of the LV agent 410 is to increase the achieved rewards R i to maximize.

[0059] In the action module 420, the actions A selected by the LV agent 410 are i executed. By an action A i an adjustment of at least one calibration parameter Pcal i and / or at least one calibration parameter value PVcal i made.

[0060] According to the invention, a discrete multi-stage approach is used to determine the action decisions of the LV agent 410. Here, an action space of the action module 420 is divided into discrete parameter increments.

[0061] The term “action space” refers to the set of all possible actions A i that the LV Agent 410 can perform in a specific environment. It defines the scope of actions A i, which are available to the LV agent 410 for each of its decisions. Depending on the nature of the problem, the action space can be discrete or continuous. Within the scope of the present invention, a discrete action space is provided, comparable to a game such as chess, in which the action space consists, for example, of the various possible moves, such as moving a specific piece to a specific location. The choice of the action space influences the complexity and feasibility of the optimization task. The action space, together with the state space and the reward structure, forms the basis for modeling the reinforcement learning algorithm.

[0062] The term “parameter step size” refers to the size or step size with which a particular parameter P iadjusted or updated during the optimization or tuning process. As already mentioned, the ADAS / ADS system 10 relies on various parameters P i to control behavior and performance. These parameters P i can relate to sensor calibration, control algorithms, filtering techniques, or decision-making processes. The parameter step size determines how strongly a parameter P i is changed during each iteration of the optimization process by the LV agent 410. The parameter step size thus influences the convergence speed and the accuracy of the actions A performed by the RL agent 410. iA larger step size leads to faster convergence, but may also result in exceeding the optimal solution. On the other hand, a smaller step size may result in slower convergence but higher accuracy. Choosing an appropriate parameter step size is a compromise between convergence speed and accuracy. Different step sizes may be required for the different ADAS / ADS systems 10 to be calibrated, and these must be determined through careful evaluation.

[0063] A discrete action space means that it comprises a smaller number of parameter combinations. For each calibration parameter value PVcal to be varied i a calibration parameter Pcal i The LV agent 410 has several discrete parameter step sizes available with which it can calculate the calibration parameter value PVcal iIn addition, the LV Agent 410 has the option of setting a calibration parameter value PVcal i unchanged. The LV agent 410 decides on the respective number of discrete step sizes based on an estimate of the parameter value change. For this estimation, the LV agent 410 can use various optimization algorithms to find the optimal number of step sizes within the discretized action space. These algorithms aim to determine the number of step sizes for the various calibration parameters Pcal i of a calibration data set 15 that maximize the expected reward and thus generate the optimized data set 17 in a shorter time.

[0064] The discrete multi-stage approach allows the LV agent 21 to reduce the complexity of decision making. This leads to a more efficient and faster learning process, since the selection of possible actions A i This can significantly reduce the time required to generate an optimized data set 17.

[0065] According to the invention, in addition to the first parameter step size, which is predetermined by the discretization of the action space, a second parameter step size is added, the magnitude of which is smaller than that of the first parameter step size. The LV agent 410 can then decide whether to use a value of a calibration parameter PVcal iby a certain number of the specified first step size of the action space and by a certain number of the smaller second step size by adding, subtracting, or leaving the first and second step sizes unchanged. This allows the resolution of the performed actions A i increased so that the LV Agent 410 can move even faster, especially in complex and confusing areas of the action space, thus accelerating the convergence to an optimal solution.

[0066] The environment module 430 calculates based on the selected action A i and taking into account the performance indicators KPI determined by the simulation results i and, if necessary, further specified requirements and constraints, the states S iA constraint can, for example, relate to the cost structure, energy costs, environmental impact, availability, or the delivery situation. These constraints can be transmitted separately to the optimization module 400 via the input module 200 or from the database 300.

[0067] In the reward module 450, the degree of deviation Δ between the calculated value for the old state S i and the value of a new state S i+1 a reward R i Since the degree of deviation Δ from the selection of the respective action A i depends, preferably in a matrix or a database the reward R i for the selected action A i saved. A reward R i preferably has the values ​​+1 and -1, whereby a small or positive deviation Δ between the old state S i and the new state S i+1is rewarded with +1 and thus reinforced, while a significant negative deviation Δ is rewarded with -1 and thus negatively evaluated. However, it is also conceivable that values ​​> 1 and values ​​< 1 and / or real numbers are used.

[0068] Preferably, a Markov decision process is used as the learning algorithm for the LV agent 410. However, it may also be provided to use a Temporal Difference Learning (TD-Learning) algorithm. The LV agent 410 with a TD-Learning algorithm performs the adaptation of the actions A i not only when he receives the reward, but after each action α i based on an estimated expected reward. Furthermore, algorithms such as Q-learning and SARSA, Actor-Critic, or Monte Carlo simulations are also conceivable, allowing for dynamic programming and strategy adaptation through iterative procedures.

[0069] In addition, the LV agent 410 and / or the action module 420 and / or the environment module 430 and / or the state module 440 and / or the reward module 450 contain calculation methods and algorithms for i for mathematical regression methods or physical model calculations. For the mathematical functions f i These can be statistical methods such as mean values, minimum and maximum values, lookup tables, models of expected values, linear regression methods or Gaussian processes, fast Fourier transforms, integral and differential calculus, Markov methods, probability methods such as Monte Carlo methods, temporal difference learning, but also extended Kalman filters, radial basis functions, data fields, convergent neural networks, deep neural networks, and / or feedback neural networks. Based on the actions A i and the rewards R ithe LV agent 410 and / or the action module 420 and / or the environment module 430 and / or the state module 440 selects / selects for a state S i one or more of these calculation functions f i out of.

[0070] Now a second cycle begins to optimize the calibration parameters Pcal i and / or calibration parameter values ​​PVcal of the calibration data set 15. Here, the LV agent 410 can perform another action A i+1 and / or another calculation function f i+1 and / or another calibration parameter Pcal i+1 and / or another calibration parameter value PVcal i+1 according to the defined strategy or policy. The result is then fed to the state module 440, and the result of the comparison is evaluated in the reward module 450. The LV agent 410 repeats the optimization process for all planned actions A i , calculation functions f i , calibration parameter pi and calibration parameter values ​​PVcal i until a result state of the optimization of the calibration data set 15 of the ADAS / ADS system 10 is reached. This is preferably the case when a deviation Δ from specified targets or performance indicators KPI i in the range of + / - 5%. The LV agent 410 thus optimizes its behavior and thus the strategy with which an action A i is selected until the calculated states S i converge. The optimized result of configuring an optimized calibration data set 17 by modified calibration parameters Pcal i and / or calibration parameter values ​​PVcal i is output in the form of output results 550 by the output module 500.

[0071] In the optimized result state, the optimized values ​​for the calibration parameters Pcal i and / or calibration parameter values ​​PVcal iof the optimized calibration data set 17, with which a specific behavior of the ADAS / ADS system 10 can be achieved, such as during an overtaking maneuver.

[0072] According to the invention, the LV agent 410 is trained in a training phase. The calibration of the calibration data set 15 is then carried out using the trained LV agent 410, which in particular has a defined strategy.

[0073] To create the strategy of the LV agent 410, a training data set 240 is entered by a user via a user interface 220 of the input module 200 or the training data set 240 is created using a software application 250 and data from the database 300.

[0074] The input module 200 can be integrated into a computer, a tablet, a smartphone, etc., and has a user interface 220 for entering input data such as the training data set 240 and other calibration data sets 15. The user interface 220 is designed, in particular, as a touchscreen, via which numerical values, a drawing or a diagram, and other inputs can be entered. The input module 200 can additionally have a keyboard, a microphone, and / or a camera. Furthermore, the input module 200 is connected to the database 300 and the optimization module 400.

[0075] In particular, historical data in the form of images, graphics, time series, parameters, etc. are stored in the database 300. For example, target and target values ​​that define a safety standard can be stored in the database 300. This data can be transferred to the optimization module 400.

[0076] The training data set 240 comprises a prepared or random selection of calibration parameters Pcal i and calibration parameter values ​​PVcal i a calibration data set 15 of the ADAS / ADS system 10. With this training data set 240, the behavior of the ADAS / ADS system 10 is now simulated in a simulation environment in the environment module 430. The simulation environment is defined by a test case T i that describes a specific scenario. For this purpose, the environment module 430 has a software application 435 that is designed to create a simulation environment using a test case T i and, if necessary, further data, to carry out the simulation of the behavior of the ADAS / ADS system 10 in the simulation environment to generate simulation results and to evaluate these simulation results.

[0077] To simulate the behavior of an ADAS / ADS system 10, the software application 435 contains a detailed simulation model with various modules. The level of detail of the simulation model of the software application 435 can vary and depends on the specific requirements for a calibration and thus on the desired accuracy of the simulation.

[0078] The simulation model includes a module for simulating the behavior and properties of the various sensors used in an ADAS / ADS system 10, such as cameras, lidar, radar, and ultrasonic sensors. This involves modeling sensor properties such as field of view, range, resolution, noise, and distortion. In addition, perception algorithms, in particular neural networks, are used to process the sensor data and detect and track objects in the environment. This can include techniques such as feature extraction, object detection, and sensor fusion to obtain a comprehensive representation of the surrounding objects. Furthermore, a motion planning module is provided, which generates a detailed trajectory for the vehicle, taking into account factors such as speed limits, traffic regulations, obstacle avoidance, and comfort restrictions. In addition, a decision-making module is provided, e.g.B. for lane changes, overtaking, evasive maneuvers and passing through intersections. Rule-based systems and / or machine learning methods can be used as algorithms. In addition, the dynamic behavior of the vehicle is modeled, including acceleration, braking, steering and suspension. The model also includes control algorithms that convert desired trajectories or actions into input signals for the vehicle's control units, e.g. accelerator pedal, brakes, steering angle and gear selection. In addition, simulation environments are created that include a virtual environment with road networks, intersections, traffic signals and traffic flows. In addition, weather conditions such as rain, fog or snow and their effects on sensor perception and vehicle behavior as well as different lighting conditions such as day, night or poor lighting are taken into account in the simulation.

[0079] The software application 435 uses neural networks as algorithms, which have been trained with suitable training data sets. A neural network consists of neurons arranged in several layers and connected to each other in various ways. A neuron is able to receive information at its input from outside or from another neuron, evaluate this information in a specific way, and pass it on in a modified form at the neuron's output to another neuron or output it as the final result. Hidden neurons are located between the input neurons and the output neurons. Depending on the type of network, there can be several layers of hidden neurons. They ensure the forwarding and processing of the information. Output neurons ultimately provide a result and pass it on to the outside world. The arrangement and interconnection of the neurons creates different types of neural networks, such as:Feed-forward networks, recurrent networks, or convolutional neural networks. These networks can be trained using unsupervised or supervised learning.

[0080] The convolutional neural network (CNN) is a special form of artificial neural network. It has multiple convolutional layers and is ideally suited for machine learning and artificial intelligence (AI) applications in the field of pattern recognition. The individual layers of the CNN are the convolutional layer, the pooling layer, and the fully connected layer. The convolutional layer is the actual convolutional layer and is capable of detecting and extracting individual features from the input data. In pattern and image recognition, these can be features such as lines, edges, or specific shapes. The input data is processed in the form of tensors such as a matrix or vectors. The pooling layer, also called the subsampling layer, condenses and reduces the resolution of the detected features using suitable filter functions. The reduced data volume increases the computing speed.Because the CNN is divided into several local, partially connected layers, it requires significantly less memory than fully connected neural networks. The training time of the convolutional neural network is also significantly shorter. By using modern graphics processors, CNNs can be trained very efficiently.

[0081] The environment module 400 uses the software application 435 to calculate simulation results that include a prediction of the system behavior of the ADAS / ADS system 10 in a simulation scenario, such as the reaction of a vehicle when merging onto a highway. The simulation results can generate a variety of data, such as synthetic sensor data that mimics the results of real sensors. This includes the generation of simulated images, point clouds, or radar measurements that represent the virtual environment. The synthetic sensor data serves as the basis for evaluating the performance of the perception algorithms of the ADAS / ADS system 10. Furthermore, the simulations enable the analysis of sensor fusion results, in which data from different sensors is combined to obtain a holistic perception of the environment.The results can include fused sensor data representations such as integrated point clouds, fused object detections, or combined sensor traces. These results demonstrate how the ADAS / ADS system 10 integrates and fuses information from different sensors. Furthermore, simulation results for object detection and tracking are generated. The algorithms of the ADAS / ADS system 10 can detect and track virtual objects in the simulated environment and provide information about the position, speed, size, and other properties of these objects. These results enable an evaluation of the accuracy, robustness, and real-time performance of the ADAS / ADS system 10 in object detection tasks. Another aspect of the simulations is the evaluation of sensor performance under different conditions.By manipulating parameters such as noise, occlusion, or sensor failure, the simulation can provide insight into the impact of these factors on the quality and reliability of the sensor data. This analysis contributes to understanding the resilience of the ADAS / ADS system 10 to sensor limitations and aids in the development of sensor redundancy or fault detection mechanisms. This enables the analysis of the sensor calibration performed by the LV agent 410. By varying the calibration parameters Pcal. i and the calibration parameter values ​​PVcal i the accuracy and reliability of the perception and decision-making processes of the ADAS / ADS system 10 may be affected.

[0082] In order to evaluate the performance and functionality of one or more functions or the overall performance of the ADAS / ADS system 10 in the simulation environment, performance indicators KPI iKey Performance Indicators (KPIs) are used as metrics to evaluate simulation results. i are used to describe the performance of an ADAS / ADS system 10, whereby different performance indicators KPI are used for different evaluation categories such as performance, comfort, safety, natural driving, recognition accuracy, reaction time, decision making, trajectory accuracy and system behavior. i be defined. An example of a performance indicator KPI i is the evaluation of a minimum distance to another vehicle or an average acceleration in a deceleration scenario. For a performance indicator KPI i A performance value is calculated, which can be represented by numerical values ​​or Boolean values. For example, a performance value of a key performance indicator (KPI) i a scale of 1 to 100 should be provided.

[0083] The software application 435 extracts suitable performance indicators KPI from the simulation results i and the respective performance values. The behavior of the ADAS / ADS system 10 is tested in a large number of selected test cases T i simulated and for the performance values ​​of the key performance indicators KPI extracted from the simulation results i Averages are created. In addition, the averages of performance values ​​for a number of performance indicators (KPI) i an overall performance can be calculated.

[0084] The T for a variety of test cases i extracted key performance indicators KPI i are passed to the state module 440 and their performance values ​​or the mean values ​​of the performance values ​​to the reward module 450. The LV agent 410 is now trained using this evaluation data, the calibration parameters Pcal i and the calibration parameter values ​​PVcal iof the training data set 240 until acceptable performance values ​​for the performance indicators KPI i which describe the behavior of the ADAS / ADS system 10 in the numerous test cases T i described simulated concrete scenarios. Through the large number of iterations, the strategy of the LV agent 410 is optimized during the training phase.

[0085] The pre-trained LV agent 410 with an optimized strategy is now used according to the invention for the calibration of similar calibration data sets 15 for an ADAS / ADS system 10 for a plurality of relevant concrete scenarios SZc iand different vehicle configurations with different technical specifications are used. To obtain meaningful and reliable calibration results, it is advisable that the calibration data sets 15 to be optimized do not deviate too significantly from the training data set 240, since then the learned strategy of the learning reinforcement agent 410 is all the more suitable for creating an optimized calibration data set 17.

[0086] For a given calibration data set 15, the RV agent 410 makes suggestions regarding the variation of the calibration parameters Pcal based on its trained strategy i and the calibration parameter values ​​PVcal i and these are then directly recalculated with the software application 435 of the environment module 430 to generate simulation results from which new performance indicators KPI j and performance values ​​for new states S j and new rewards R jThe iteration is repeated until an optimized calibration data set 17 is created.

[0087] Since a fast calculation result is often desired, the computing speed is crucial for the numerous computing operations in the optimization module 400. To enable the execution of the computing operations in real time, the use of cloud computing infrastructure is advantageous, as this ensures fast calculation.

[0088] In particular, it can be provided that the simulation and calculation results are stored in the cloud computing infrastructure in the form of states, actions, rewards, and strategies and are each available via the Internet. The LV agent 410, the action module 420, the environment module 430, the state module 440, and the reward module 450 have the necessary technical interfaces and protocols for accessing the cloud computing infrastructure. This can increase computing efficiency through simplified access options and speeds to previously calculated states, actions, rewards, and strategies.

[0089] This optimized calibration data set 17 is transferred to the output module 500, which outputs the calibration data set 17 directly or generates output results 550 therefrom. The output module 500 can also be integrated into a hardware device such as a computer, a tablet, a smartphone, etc., wherein, in particular, the hardware device of the input module 200 and the hardware device of the output module 500 can be identical. The output data 550 can be, for example, design plans, images and graphics, video sequences, audio sequences, holograms, key figures, diagrams, mathematical formulas, PowerPoint presentations, etc., to represent the actual adjustment of the calibration parameters Pcal i and calibration parameter values ​​PVcal i of a real ADAS / ADS system 10.

[0090] In Fig. Figure 2 shows the process steps for the automated calibration of functions of a driver assistance system (Advanced Driver Assistance System ADAS / Automated Driving System, ADS) for a variety of scenarios and vehicle configurations.

[0091] In a step S10, a learning reinforcement agent 410 of an optimization module 400 is trained with a training data set 240, wherein the optimization module 400 comprises at least the learning reinforcement agent 410, an action module 420 with an action space, an environment module 430, a state module 440 and a reward module 450, and wherein the training data set 240 contains calibration parameters Pcal i and calibration parameter values ​​PVcal i an ADAS / ADS system 10.

[0092] In a step S20, a strategy of the learning reinforcement agent 410 for varying calibration parameters Pcal i and calibration parameter values ​​PVcal i in the form of actions A ito optimize calibration data sets 15 of ADAS / ADS systems 10.

[0093] In a step S30, a calibration data set 15 of the ADAS / ADS system 10 with calibration parameters Pcal i and calibration parameter values ​​PVcal i entered via a user interface 220 of an input module 200 or from a database 300.

[0094] In a step S40, a plurality of test cases T i selected from the database 300 and from the test cases T i Simulation environments with simulation scenarios are created.

[0095] In a step S50, simulations of the behavior of the ADAS / ADS system 10 in the simulation scenarios are carried out and simulation results are generated, wherein the simulation results reflect the system behavior of the ADAS / ADS system 10 in the simulation scenarios.

[0096] In a step S60, the simulation results are presented in the form of performance indicators KPI i and performance values.

[0097] In a step S70, states S i from the performance indicators KPI i and rewards R i determined from the performance values.

[0098] In a step S80, modeled calibration parameters Pcal i and calibration parameter values ​​PVcal i to achieve an improved system behavior of the ADAS / ADS system 10 in the simulation scenarios by the trained learning reinforcement agent 410 in several iterations until an optimized calibration data set 17 has been generated, wherein to determine the decisions of the learning reinforcement agent 410 regarding the actions to be performed A ithe action space of the action module 420 is divided into discrete parameter step sizes and the learning reinforcement agent 410 determines the calibration parameter value PVcal i a calibration parameter Pcal to be varied i reduced, kept constant or increased by one or more parameter increments.

[0099] In a step S90, output results 550 are generated and output by an output module 500, wherein the output results 550 contain the optimized calibration data set 17.

[0100] Fig. 3 schematically illustrates a computer program product 900 comprising an executable program code 950 configured to perform the method according to the first aspect of the present invention.

[0101] With the present invention, the calibration of a calibration data set of an ADAS / ADS system can be performed efficiently and reliably using a virtual approach with a learning algorithm. The calibration can be performed using a large number of test cases that define a variety of simulation scenarios, and the trained learning algorithm leads to stable and valid calibration results. Since the simulation scenarios generated by the test cases are reproducible, the virtual calibration is not dependent on the random occurrence of relevant scenarios in a real environment. By generating virtual scenarios that simulate real driving conditions, the performance of the driver assistance system can be tested and analyzed.Performance indicators such as detection accuracy, reaction time, decision making and system behavior can be measured and evaluated and are used according to the invention to optimize the calibration parameters and calibration parameter values.

[0102] The learning algorithm offers a cost-effective way to iterate and refine the system behavior of the ADAS / ADS system to generate an optimized configuration of a calibration dataset. The discrete multi-stage approach, in which the action space of the action module is divided into discrete parameter steps and the learning algorithm is provided with a predefined framework for its decisions, allows for an optimized calibration dataset to be determined more quickly, as every conceivable variation no longer needs to be run through, thus allowing a more targeted solution to be found.

[0103] This allows calibration to be performed quickly and efficiently in a wide variety of different and complex driving scenarios that are difficult to reproduce in reality. This allows the calibration parameters of an ADAS / ADS system to be tested under various conditions, such as heavy traffic or rare events. By simulating a wide range of scenarios, potential weak points and edge cases can be uncovered that may not occur in real-world testing. Furthermore, virtual calibration offers significant cost and time savings. Instead of conducting extensive real-world tests, which can be expensive and time-consuming, virtual simulations enable rapid and repeated testing in a controlled environment. A wide variety of scenarios can be simulated, allowing calibration to be improved more efficiently. Reference symbol 10 ADAS / ADS system 15 Calibration data set 17 optimized calibration data set 100 systems 200 input module 220 User Interface 240 training data sets 250 software applications 300 database 400 Optimization Module 410 Learning reinforcement agent 420 Action Module 430 Environment module 435 Software application 440 State module 450 Reward Module 500 output module 550 output results 900 Computer program product 950 program code

Claims

[1] Method for the automated calibration of functions of a driver assistance system (Advanced Driver Assistance System ADAS / Automated Driving System, ADS) for a variety of scenarios and vehicle configurations, comprising: - Training (S10) a learning reinforcement agent (410) of an optimization module (400) with a training data set (240), wherein the optimization module (400) comprises at least the learning reinforcement agent (410), an action module (420) with an action space, an environment module (430), a state module (440) and a reward module (450), and wherein the training data set (240) contains calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) of an ADAS / ADS system (10); - generating (S20) a strategy of the learning reinforcement agent (410) for varying calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) in the form of actions (A i) for optimising calibration data sets (15) of ADAS / ADS systems (10); - Entering (S30) a calibration data set (15) of the ADAS / ADS system (10) with calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) via a user interface (220) of an input module (200) or from a database (300); - Selecting (S40) a large number of test cases (T i ) from the database (300) and generating simulation environments with simulation scenarios from the test cases (T i ) in the environment module (430); - performing (S50) simulations of the behavior of the ADAS / ADS system (10) in the simulation scenarios and generating simulation results, wherein the simulation results reflect the system behavior of the ADAS / ADS system (10) in the simulation scenarios; - Evaluate (S60) the simulation results in the form of key performance indicators (KPI i ) and performance values; - Determining (S70) states (S i ) from the key performance indicators (KPI i ) and rewards (R i ) from the performance values; - Calculate (S80) modeled calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) to achieve an improved system behavior of the ADAS / ADS system (10) in the simulation scenarios by the trained learning reinforcement agent (410) in several iterations until an optimized calibration data set (17) has been generated, wherein to determine the decisions of the learning reinforcement agent (410) regarding the actions to be carried out (A i ) the action space of the action module (420) into discrete parameter step sizes with which a certain parameter (P i) during which the optimization is adapted or updated, is divided such that, in addition to a first parameter step size, which is predetermined by the discretization of the action space, a second parameter step size is added, the amount of which is smaller than the amount of the first parameter step size, and wherein the learning reinforcement agent (410) determines the calibration parameter value (PVcal i ) of a calibration parameter to be varied (Pcal i ) by one or more parameter increments, such that the learning reinforcement agent (410) decides whether to use a value of a calibration parameter (Pcal i ) by a certain number of the given first step size of the action area and by a certain number of the smaller second step size by adding, subtracting or leaving the first and second step sizes unchanged; - generating and outputting (S90) output results (550) by an output module (500), wherein the output results (550) contain the optimized calibration data set (17). [2] Method according to claim 1, wherein a state (S i ) of the ADAS / ADS system (10) is determined by the state module (440) and calibration parameters (Pcal i ) and calibration parameter values ​​(PVcal i ) includes. [3] Method according to claim 1 or 2, wherein the software application (435) of the environment module (400) uses at least one neural network, in particular recurrent neural networks or convolutional neural networks, as algorithm. [4] Method according to one of claims 1 to 3, wherein the reinforcement learning algorithm of the optimization module (400) is designed as a Markov decision process or as Temporal Difference Learning (TD-Learning) or as Q-Learning or as SARSA or as Monte Carlo simulation or as Actor-Critic or as Proximal Policy Optimization. [5] Method according to one of claims 1 to 4, wherein at least one calibration parameter (Pcal i) is a physical quantity, a controller parameter, a chemical quantity, a torque, a rotational speed, a voltage, a current, an acceleration, a speed, a braking value, a specific sensor, a measured value of a sensor, a resolution of a sensor, an installation position of a sensor, a technical specification of a sensor, a direction, an angle, a radius, a vehicle property such as engine power, a material, a shape, a specific fuel consumption, a load and / or a specific vehicle configuration, and wherein a sensor used is an image-capturing 2D / 3D camera in the visible, IR and / or UV range, a radar system with one or more radar sensors, a LIDAR system for optical distance and speed measurement, an ultrasonic sensor and / or a GPS system. [6] Method according to one of claims 2 to 5, wherein a calculation function (f i) and / or an action (A i ) based on a policy for the condition (S i ) is selected for the modification of at least one calibration parameter (Pcal i ) and / or a calibration parameter value (PVcal i ) of the calibration data set (15) of the ADAS / ADS system (10) by the learning reinforcement agent (410); wherein a new state (S i+1 ) from the environment module (430) based on the modified value for the calibration parameter (Pcal i ) and / or the calibration parameter value (PVcal i ) is calculated; where the new state (S i+1 ) with the old state (S i ) and a deviation (Δ) for a comparison result is calculated in the state module (440); and wherein a reward (R i ) is determined by the reward module (450) for the comparison result. [7] Method according to one of claims 1 to 6, wherein the output results (550) contain design plans, images and graphics, video sequences, acoustic sequences, holograms, key figures, diagrams, mathematical formulas and / or Power Point presentations, etc. [8] System (100) for the automated calibration of functions of a driver assistance system (Advanced Driver Assistance System ADAS / Automated Driving System, ADS) for a plurality of scenarios and vehicle configurations, comprising an input module (200), a database (300), an optimization module (400) and an output module (500), wherein the optimization module (400) comprises a learning reinforcement agent (410), an action module (420) with an action space, an environment module (430), a state module (440) and a reward module (450), and wherein the system (100) is designed to carry out the method according to one of claims 1 to 7. [9] ADAS / ADS system (10) for a vehicle, wherein the method according to one of claims 1 to 7 is used for calibrating a calibration data set (15) of the ADAS / ADS system (10). [10] A computer program product (900) comprising an executable program code (950) configured to carry out the method according to any one of claims 1 to 7 when executed.

Images