Method and device for evaluating and / or setting a regulating parameter of a regulator

Abstract

Method for evaluating and / or setting at least one performance indicator of a steering mechanism of a steering system of a vehicle, the method having the steps of: providing (S1) measurement and / or simulation data of at least one performance indicator of the steering system, wherein the measurement and / or simulation data are generated in a drive using the vehicle and / or in a simulation of the steering mechanism of the steering system; applying (S2), based on the provided measurement and / or simulation data, a machine learning model trained by reinforcement learning to evaluate and / or set at least one performance indicator of the steering mechanism; and, evaluating and / or setting (S3) at least one performance indicator of the steering mechanism by the machine learning model, in the case of a quality function of the machine learning model, in particular until a predetermined quality criterion is met.

Description

Technical Field

[0001] The present invention relates to a method and an apparatus for evaluating and / or setting adjustment parameters of an adjuster for a vehicle's steering system. Background Technology

[0002] Steering feel is one of the key distinguishing features of modern steering systems. It describes the interaction between the vehicle's steering angle, torque, lateral acceleration, and yaw rate, and decisively influences the subjective driving experience. In current steering systems, steering feel is achieved through careful tuning of the software parameters of the motor that controls the steering system.

[0003] Traditional practices for adjusting these parameters are mostly manual. This involves repeatedly testing the vehicle, subjectively evaluating the steering feel, and iteratively matching the parameters—a process based on the experience of the vehicle testing engineer and, where necessary, on specific criteria. However, this approach is very time-consuming and heavily dependent on the engineer's individual skill and estimations.

[0004] A newer approach aims to quantitatively describe steering feel by using key performance indicators (KPIs) for predefined steering maneuvers. This method enables a simulation-based tuning process, in which steering maneuvers are simulated, KPIs are calculated, and parameters are matched according to an optimization strategy. The advantages of this approach are higher efficiency and repeatability. However, simulation-based tuning is limited by model inaccuracies, which necessitates final manual adjustments in the vehicle.

[0005] The challenge, therefore, lies in reducing the burden of manual adjustments for vehicle test engineers—whether after simulation-based optimization or in general—and further automating the process without compromising the accuracy and quality of steering feel.

[0006] Methods for adjusting such steering systems are known from existing technologies, such as DE 10 2021 206 588 A1 and WO 2022 / 139790 A1. However, further optimization potential exists.

[0007] The objective of this invention is therefore to describe an improved method and / or an improved apparatus.

[0008] This task is solved by the method according to the features of claim 1. This task is solved by the apparatus according to the features of claim 10. Summary of the Invention

[0009] According to a first aspect, a method is proposed for evaluating and / or setting at least one performance index of the steering mechanism of a vehicle's steering system, the method comprising the steps of: providing measurement and / or simulation data of at least one performance index of the steering system, wherein the measurement and / or simulation data is generated during driving of the vehicle and / or in simulation of the steering mechanism of the steering system; based on the provided measurement and / or simulation data, applying a machine learning model trained by reinforcement learning to evaluate and / or set at least one performance index of the steering mechanism; and, by means of the machine learning model, evaluating and / or setting at least one performance index of the steering mechanism, in particular until a predetermined quality criterion is met, using a quality function of the machine learning model.

[0010] An alternative expression for the feature “based on the provided measurement and / or simulation data, applying a machine learning model trained by reinforcement learning to evaluate and / or set at least one performance metric of the regulator” could be: based on the provided measurement and / or simulation data, applying a machine learning model to train the machine learning model using reinforcement learning, specifically to evaluate and / or set at least one performance metric of the steering mechanism.

[0011] During the training process of a machine learning model, the model is trained, for example, using reinforcement learning based on training data, to evaluate and / or define at least one performance metric of the steering mechanism. The training data may be measurement and / or simulation data corresponding to at least one performance metric of the steering system within a data structure, and may be supplemented and / or enriched and / or replaced as necessary by synthetic data.

[0012] It should be understood that the steps and other optional steps according to the invention do not necessarily have to be performed in the order shown, but may be performed in other orders. Additional intermediate steps may also be provided. Furthermore, each step may include one or more sub-steps without departing from the scope of the method according to the invention.

[0013] According to a second aspect, an apparatus is proposed for evaluating and / or setting at least one performance index of the steering mechanism of a vehicle's steering system, wherein the apparatus has an evaluation and calculation device configured to perform the following steps: providing measurement and / or simulation data of at least one performance index of the steering system, wherein the measurement and / or simulation data is generated during driving of the vehicle and / or in simulation of the steering mechanism of the steering system; based on the provided measurement and / or simulation data, applying a machine learning model trained by reinforcement learning to evaluate and / or set at least one performance index of the steering mechanism; and, using the machine learning model, evaluating and / or setting at least one performance index of the steering mechanism, particularly until a predetermined quality criterion is met, with regard to the quality function of the machine learning model.

[0014] The method describes a scheme for evaluating and / or setting at least one performance index (KPI) of the steering mechanism of a vehicle's steering system. Here, measurement and / or simulation data for at least one performance index are first provided. This data may originate either from actual driving of the vehicle or from simulations of the steering mechanism of the steering system. Preferably, the provided data represents at least one performance index of the steering system and serves as the basis for future analysis and optimization.

[0015] Here, a machine learning model, specifically trained through reinforcement learning, is used to evaluate and / or set at least one performance metric for the steering mechanism. The machine learning model analyzes the provided data and further processes it. Reinforcement learning enables the model to iteratively develop strategies for optimizing at least one performance metric through feedback from the data. Here, the machine learning model evaluates possible operating options and learns from the results of these evaluations to improve the performance of the steering system.

[0016] In the following steps, it is preferable to evaluate and / or set at least one performance indicator. Preferably, a machine learning model is used to evaluate at least one performance indicator, i.e., to analyze its quality and suitability, and, if necessary, to set parameters such that the desired KPI value is achieved. In this case, a quality function for evaluating the quality of at least one performance indicator is preferably used. The method continues until a predefined quality criterion is met, which defines a specified target for at least one performance indicator.

[0017] The method employs a machine learning model with reinforcement learning to provide adaptive and iterative KPI optimization. Target-oriented optimization is ensured through the use of a quality function and quality criteria. Overall, the method is an efficient and automated solution for evaluating and optimizing at least one performance metric for a steering system.

[0018] Preferably, the quality function serves as the basis for evaluating the quality of the at least one performance metric and can be defined in different ways. One possibility is an error-based definition, where the difference between the setpoint and the actual value is minimized, for example, through mean squared error or absolute error. Another possibility is preferably a KPI-based evaluation, where specific key performance indicators, such as steering accuracy, reset torque, stability, or damping characteristics, are considered. A multi-criteria approach is also preferably envisioned, where different parameters are weighted according to their correlation. Also preferably, dynamic aspects, such as time delay or overshoot, can be incorporated into the quality function by considering integral or differential criteria. Alternatively, a heuristic approach based on empirical values ​​or expert knowledge, such as through fuzzy rules, can be preferably utilized.

[0019] Preferably, a quality criterion is determined: when the quality function is considered sufficiently good and when the optimization process terminates. Preferably, the quality criterion can be defined by a threshold, below which the value of the quality function must be, such as a maximum permissible error. Alternatively, a convergence criterion can preferably be considered, whereby the process terminates when subsequent iterations no longer provide significant improvement. Process-specific criteria, such as compliance with regulatory requirements or technical standards, can also be used.

[0020] Preferably, quality criteria can be matched both manually and automatically. Manual matching is preferably performed by engineers who modify thresholds or tolerances based on experience or specific requirements. Automated matching can preferably be performed by a machine learning model that learns from feedback through reinforcement learning. Furthermore, scenario-specific matching criteria are preferred, for example, using stricter regulations for high-speed maneuvers or higher tolerances for low-speed situations. Preferably, external conditions, such as weather conditions or lane status, can also be included in the matching.

[0021] Machine learning models can be trained in various ways. In supervised learning, labeled data is used to train the model to predict the correct setpoint from input values, such as gear position. In reinforcement learning, the model is trained through interaction with an environment, such as a steering system simulation, and the model optimizes based on rewards earned for achieving defined quality criteria. Alternatively, the model can be trained unsupervised by analyzing patterns and anomalies in unstructured data. Hybrid approaches can also be used, combining supervised pre-training with reinforcement-based fine-tuning. Additionally, synthetic data from simulations can be used to train the model to cover scenarios that are difficult to reproduce in real-world testing. Another approach is transfer learning, where pre-trained models are further matched with similar contexts. These methods enable efficient and application-oriented machine learning model training.

[0022] The steering system can be a steer-by-wire system. This is a steering technology in which the mechanical connection between the steering wheel and the wheels is eliminated. Instead, the driver's steering input is detected by sensors, processed digitally, and transmitted to the wheels via electric actuators. The rack that controls the wheels is preferably moved electrically.

[0023] Preferably, the steering system adjustment is provided via a rack position adjuster. The rack position adjuster is a software component in either a steer-by-wire system or a conventional steering system. The rack position adjuster controls the precise positioning of the rack, which transmits steering action to the wheels. Here, the rack position adjuster processes control commands and position feedback to precisely position the rack to the desired location. Preferably, this is achieved via an electric actuator. In the rack position adjuster, adjustment techniques, such as PID control, or a model-based approach are preferably used, ensuring high accuracy and / or dynamics.

[0024] This invention reduces the burden on vehicle test engineers for manual adjustments following simulation-based tuning and / or for manual tuning in general. In particular, it utilizes reinforced machine learning. Machine learning is implemented on control devices and / or computers to evaluate parameters and / or generate matching suggestions and / or to (particularly proactively) match parameters.

[0025] Because reinforcement learning is not performed by a human individual, it is independent of the experience and / or motivation of the vehicle test engineer. Furthermore, using reinforcement learning can reduce tuning overhead because evaluation and / or matching can be performed at a higher frequency than could be achieved by a vehicle test engineer. This is because vehicle test engineers typically perform these steps as a sequence of driving the vehicle, evaluating its performance, and matching parameters based on that evaluation.

[0026] The method is used in the development of steering systems, particularly steer-by-wire systems and / or software components of such steering systems. Preferably, the method is used in the design phase of the steering system, that is, before the steering system has been manufactured or installed in a vehicle. In principle, the method can be used not only in prototype vehicles but also for final parameter adjustments for production samples and / or mass-production vehicles.

[0027] Preferably, the invention relies on the availability of specific performance requirements for controlling the rack position, which are preferably based on real measurements and thus enable accurate evaluation. Preferably, such requirements have been made available through advanced developments in virtual reality within the scope of online control development projects and / or single-wheel steering mechanism development projects. Here, the virtual simulation environment offers the advantage of creating near-realistic conditions under which control characteristics can be evaluated and / or optimized without having to perform costly physical testing methods. This allows for more efficient and earlier adjustments to the control system while adhering to defined performance requirements.

[0028] The implementation of this method is accordingly applicable to the device. It should be understood that linguistic variations of the method's characteristics can be restated for the device according to linguistic conventions, without needing to explicitly list such variations here.

[0029] In another aspect, it is proposed that the measurement and / or simulation data have one or more of the following parameters: the set value and / or actual value of the rack position of the steering system, the set steering wheel angle and / or the actual steering wheel angle, the set torque and / or actual torque of the rack, the set steering tie rod force and / or the actual steering tie rod force, the set yaw rate and / or the actual yaw rate of the vehicle, the set lateral acceleration and / or the actual lateral acceleration, the set travel speed and / or the actual travel speed, and the set torque and / or the actual torque of the steering assist motor.

[0030] Measurement and / or simulation data can include various parameters of the steering system. These include setpoints and / or actual values ​​of the steering rack position, setpoints and / or actual values ​​of the steering wheel angle, setpoints and / or actual torque of the rack, setpoints and / or actual tie rod forces, setpoints and / or actual yaw rates of the vehicle, setpoints and / or actual lateral accelerations, setpoints and / or actual travel speeds, and setpoints and / or actual torques of the steering assist motor. These parameters represent important characteristics of the steering system and enable detailed analysis and optimization of steering behavior considering both actual conditions and target parameters.

[0031] In another aspect, it is proposed that machine learning models have transformer models and / or neural networks and / or recurrent neural networks.

[0032] To evaluate and / or match the steering system's adjustment parameters, different machine learning models can be used based on the data and task assignment.

[0033] Preferably, neural networks, especially deep neural networks (DNNs), are suitable for modeling nonlinear relationships. Recurrent neural networks, such as LSTM or GRU, are preferred to account for temporal correlations in dynamic processes. Transformer models can be used for multivariate time series analysis because they can prioritize important features. Reinforcement learning is suitable for optimizing KPIs through interaction with simulated or real environments. Hybrid approaches that combine physical models with machine learning are also preferred to leverage domain knowledge and reduce data requirements.

[0034] In another aspect, it is proposed that the predetermined quality criteria include a combination of one or more of the following components: integral regulation deviation on the regulator's regulation object (Regelstrecke), the maximum permissible deviation between the setpoint and the actual value, convergence conditions based on changes across successive iterations of the quality function, compliance with specific performance indicators (Key performance indicators), and dynamic evaluation of time-related parameters such as reaction time, overshoot, or stability criteria.

[0035] Preferably, the quality criteria comprise a combination of multiple components used to evaluate and optimize the regulator. Among these is an integral regulation deviation on the regulated object, which enables minimization of the deviation between the setpoint and the actual value over a defined time period. Preferably, a maximum permissible deviation between the setpoint and the actual value is included to clearly define the error range. A convergence condition ensures that the optimization process terminates once no further significant improvement is achieved. Additionally or alternatively, compliance with specific key performance indicators is considered to map application-specific requirements. Preferably, time-dependent parameters, such as reaction time, overshoot, or stability criteria, are also included to ensure optimal regulator performance in dynamic scenarios.

[0036] The main technical advantages of this approach lie in the flexibility and accuracy of the quality criteria. Preferably, the integral control deviation enables a comprehensive evaluation of the regulator's performance across the entire controlled object, while the maximum deviation limit ensures strict control over the error magnitude. For example, the convergence condition is responsible for efficiency because it avoids unnecessary iterations. By considering specific key performance indicators and / or dynamic parameters, the quality criteria can preferably be used in multiple ways and are primarily adaptable to different application areas.

[0037] In another aspect, it is proposed that measurement and / or simulation data be continuously provided to a machine learning model during the driving of the vehicle, and that the machine learning model continuously performs evaluation and / or setting of at least one performance index of the steering mechanism.

[0038] If measurement and / or simulation data are continuously fed to the machine learning model during driving, this enables dynamic and continuous evaluation and matching of at least one performance metric of the steering mechanism. In this scenario, the model analyzes the incoming data in real time and compares it to setpoints. Based on this analysis, the model evaluates the regulator's performance and identifies deviations or suboptimal performance metrics of the steering mechanism.

[0039] Next, the machine learning model matches performance metrics, such as those of the steering mechanism, to optimize steering performance or steering feel, preferably taking into account certain criteria. This continuous process ensures that the steering system itself can adapt to changing conditions, such as different road conditions, speeds, or vehicle loads.

[0040] In a first variant, the invention preferably includes the following steps: A vehicle test engineer initializes the reinforcement learning system. The vehicle test engineer performs predefined maneuvers for KPI generation and detects the required measurement signals. The vehicle test engineer signals to the reinforcement learning system that a new measurement for one or more KPIs is available, i.e., a new value is available. The reinforcement learning system processes the new measurement. The reinforcement learning system displays the quality to the vehicle test engineer. If the vehicle test engineer is satisfied with the quality, the method stops and the optimal value of the parameter is found; if not, the next step is implemented. The reinforcement learning system proposes new values ​​for the parameters. The vehicle test engineer initializes and verifies the transmission of the new values ​​to the computing unit. The method returns to the step (the vehicle test engineer performs predefined maneuvers for KPI generation and detects the required measurement signals).

[0041] In another aspect, it is proposed to detect and / or simulate measurement and / or simulation data during the use of a vehicle on a predetermined test road segment, and to provide the measurement and / or simulation data to a machine learning model after the test road segment is completed.

[0042] If data is recorded during a test drive and only provided to the machine learning model after the drive has ended, the evaluation and matching of at least one performance metric of the steering mechanism is not performed in real time. Instead, the model retrospectively analyzes the recorded data by evaluating the detected settings and actual values, adjustment deviations, and other relevant parameters. Based on this analysis, the model evaluates whether at least one performance metric of the steering mechanism is correctly set or requires matching. If the analysis concludes that the current parameters do not meet pre-defined quality criteria, the model can propose another test drive. Here, previously identified deficiencies are addressed, and optimized parameters are tested to further improve adjustment quality.

[0043] Compared to real-time analysis, post-mortem analysis enables more detailed and comprehensive evaluations because the model can assess the entire test drive "in rehearsed." This approach is particularly suitable when the test environment is controlled and rapid matching is not necessary, or when simulation of the test drive is used as a supplement to accelerate the matching process.

[0044] In another aspect, it is proposed that if the regulator's adjustment parameters do not meet the predetermined quality criteria, a request for re-driving the test section will be output.

[0045] Preferably, the test drive is performed on a predetermined road segment, the length of which is selected such that all relevant driving conditions required for evaluating and optimizing at least one performance indicator of the steering mechanism of the steering system are preferably covered. Preferably, the duration of the test drive depends on the road segment length, driving speed, and complexity of the test scenario. Preferably, the test duration ranges from a few minutes to half an hour, depending on the requirements for data volume and data quality.

[0046] Preferably, the test track is a predefined track that remains identical across all test runs. This ensures comparable conditions between different drives and enables accurate evaluation of the adjustments and matches made. Preferably, the test track includes known test conditions such as straight sections, curves, different road surfaces, and possibly defined obstacles or steering scenarios. This invariance also facilitates the creation of realistic simulations, as the test conditions are fully known and can be modeled.

[0047] Preferably, the defined test road segment allows for data enrichment using simulation data based on the same road segment conditions. This enables the use of additional information from the virtual scenario to supplement the data detected during test driving, providing a more comprehensive database for the evaluation and optimization of at least one performance metric of the steering mechanism. Preferably, varying environmental conditions are also integrated into the simulation to examine the robustness of at least one performance metric of the steering mechanism under different conditions.

[0048] In a second variation, the invention preferably includes the following steps: A vehicle test engineer initializes the system for reinforcement learning. The vehicle test engineer initializes the criteria of the reinforcement learning system based on knowledge gained during previously performed simulation-based adjustments. The vehicle test engineer performs predefined maneuvers for KPI generation and detects the required measurement signals. The vehicle test engineer signals to the reinforcement learning system that a new measurement for one or more KPIs, i.e., a new value, is available. The reinforcement learning system processes the new measurement. The reinforcement learning system displays the quality to the vehicle test engineer. If the vehicle test engineer is satisfied with the quality, the method stops and the optimal value of the parameter has been found; if not, the next step is implemented. The reinforcement learning system proposes new values ​​for the parameter. The vehicle test engineer initializes and verifies the transmission of the new values ​​to the ECU. The method returns to the step where the vehicle test engineer performs predefined maneuvers for KPI generation and detects the required measurement signals.

[0049] In another aspect, protection is also claimed for a control device included in a vehicle with autonomous driving capabilities, and / or a robotic system and / or an industrial machine, and capable of implementing the method in one of its aspects.

[0050] In another aspect, it is proposed that at least one performance metric is performed by matching the software and / or transmission mechanism (Mechanik) and / or at least one hardware characteristic of the steering mechanism.

[0051] Matching performance indicators through software is preferably achieved by altering control algorithms, such as adjusting parameters, modifying characteristic curves, or optimizing sensor processing. For example, mechanical matching includes reducing friction, optimizing the gear ratio, minimizing backlash in the steering tie rod, or using lighter materials. Preferably, hardware features can be used to integrate more powerful motors, more precise sensors, or more efficient control devices to improve the steering mechanism. These measures work either individually or in combination to optimize specific performance indicators, such as accuracy, energy efficiency, or responsiveness.

[0052] In another aspect, a computer program having program code is claimed for carrying out at least a portion of the method in one of its aspects when executed on a computer. In other words, a computer program (product) is claimed that includes instructions, when executed by a computer, causing the computer to carry out the method / steps of the method in one of its aspects.

[0053] In another aspect, a computer-readable data carrier having program code of a computer program is proposed for carrying out at least a portion of the method in one of its aspects when the computer program is executed on a computer. In other words, the present invention relates to a computer-readable (storage) medium comprising instructions that, when executed by a computer, cause the computer to carry out the method / steps of the method in one of its aspects.

[0054] The described design and improvement schemes can be combined with each other arbitrarily.

[0055] Other possible designs, improvements, and implementations of the present invention also include combinations of features of the invention not explicitly mentioned in the embodiments described above or below. Attached Figure Description

[0056] The accompanying drawings should facilitate a further understanding of the embodiments of the invention. These drawings illustrate the embodiments and, in conjunction with the description, are used to explain the principles and solutions of the invention.

[0057] With reference to the accompanying drawings, many advantages among other embodiments and mentioned benefits are apparent. The elements shown in the drawings are not necessarily depicted to scale.

[0058] Figure 1 : A schematic flowchart of an embodiment of this method.

[0059] In the figures described herein, unless otherwise indicated, the same reference numerals denote the same or functionally identical elements, components or assemblies. Detailed Implementation

[0060] Figure 1 A schematic flowchart is shown for a method of evaluating and / or setting the adjustment parameters of an adjuster for a vehicle's steering system.

[0061] In any embodiment, the method may be implemented at least in part by device 100, which may include a plurality of components not shown in more detail, such as one or more providing means and / or at least one evaluation and calculation means. It should be understood that the providing means may be constructed together with the evaluation and calculation means, or may be different from the evaluation and calculation means. Furthermore, device 100, which may be part of a system, may include storage means and / or output means and / or display means and / or input means.

[0062] The computer-implemented method includes at least the following steps: In step S1, at least one performance index of the steering system is measured and / or simulation data is provided, wherein the measurement and / or simulation data is generated during driving of the vehicle and / or during simulation of the steering mechanism of the steering system.

[0063] In step S2, based on the provided measurement and / or simulation data, a machine learning model trained by reinforcement learning is applied to evaluate and / or set at least one performance metric of the steering mechanism.

[0064] In step S3, at least one performance index of the steering mechanism is evaluated and / or set using a machine learning model, specifically until a predetermined quality criterion is met, by employing the quality function of the machine learning model.

Claims

1. A method for evaluating and / or setting at least one performance index of the steering mechanism of a vehicle's steering system, the method comprising the steps of: - Provide (S1) measurement and / or simulation data of at least one performance index of the steering system, wherein the measurement and / or simulation data is generated during driving of the vehicle and / or during simulation of the steering mechanism of the steering system; - Based on the provided measurement and / or simulation data, the application (S2) evaluates and / or sets at least one performance metric for the steering mechanism using a machine learning model trained through reinforcement learning; and - Using the machine learning model, in particular, until a predetermined quality criterion is met, evaluate and / or set (S3) at least one performance index of the steering mechanism.

2. The method according to claim 1, wherein the measurement and / or simulation data has one or more of the following parameters: the set value and / or actual value of the rack position of the steering system, the set steering wheel angle and / or the actual steering wheel angle, the set torque and / or actual torque of the rack, the set steering tie rod force and / or the actual steering tie rod force, the set yaw rate and / or the actual yaw rate of the vehicle, the set lateral acceleration and / or the actual lateral acceleration, the set driving speed and / or the actual driving speed, and the set torque and / or the actual torque of the steering assist motor.

3. The method according to claim 1 or 2, wherein the machine learning model has a transformer model and / or a neural network and / or a recurrent neural network.

4. The method according to any one of the preceding claims, wherein the predetermined quality criterion comprises a combination of one or more of the following features: integral regulation deviation on the regulator's regulation object, maximum permissible deviation between the set value and the actual value, convergence condition based on changes across successive iterations of the quality function, compliance with specific performance indicators and / or dynamic evaluation of time-related parameters such as reaction time, overshoot and / or stability criteria.

5. The method according to any one of the preceding claims, wherein the measurement and / or simulation data are continuously provided to the machine learning model during the use of the vehicle, and the machine learning model continuously performs evaluation and / or setting of the adjustment parameters.

6. The method according to any one of claims 1 to 4, wherein the measurement and / or simulation data are detected and / or simulated during the driving of the vehicle on a predetermined test road segment, and the measurement and / or simulation data are provided to the machine learning model after the driving on the test road segment is completed.

7. The method of claim 6, wherein if the adjustment parameters of the regulator do not meet the predetermined quality criterion, a request to re-drive the test section is output.

8. The method according to any one of the preceding claims, wherein the at least one performance indicator is performed by matching the software and / or transmission mechanism and / or at least one hardware characteristic of the steering mechanism.

9. A computer program having program code for implementing at least a portion of the method according to any one of claims 1 to 8 when the computer program is executed on a computer.

10. A computer-readable data carrier having program code of a computer program for implementing at least a portion of the method according to any one of claims 1 to 8 when the computer program is executed on a computer.

11. An apparatus (100) for evaluating and / or setting at least one performance index of the steering mechanism of a vehicle's steering system, wherein the apparatus (100) has an evaluation and calculation device configured to perform the following steps: - Provide (S1) measurement and / or simulation data of at least one performance index of the steering system, wherein the measurement and / or simulation data is generated during driving of the vehicle and / or during simulation of the steering mechanism of the steering system; - Based on the provided measurement and / or simulation data, the application (S2) evaluates and / or sets at least one performance metric for the steering mechanism using a machine learning model trained through reinforcement learning; and - Using the machine learning model, in particular, until a predetermined quality criterion is met, evaluate and / or set (S3) at least one performance index of the steering mechanism.

Images