DE102021214081A1 - Performance-modeled cyclists as a simulation component for vehicles - Google Patents

Abstract

The invention relates to a method for testing and/or validating an automated driving control function of a vehicle (1), a simulation being carried out and the simulation including a cyclist (3) among other road users, the movements of the cyclist (3) being based on a performance model of the cyclist (3) are simulated, with the performance model modeling torque inputs through muscle movement of the cyclist (3), and in the case of an electrified bicycle of the cyclist (3) additionally through the power of the electric drive of the bicycle, depending on environmental variables on a resulting movement of the Cyclist (3) depicts, and wherein the automated driving control function is linked to the simulation, so that the automated driving control function can react to the simulated movements of the cyclist (3).

Description

The invention relates to a method for testing and/or validating an automated driving control function of a vehicle, and a simulation system for testing and/or validating an automated driving control function of a vehicle.

Simulation-based methods are almost indispensable for the development and validation of automated vehicles. The behavior of automated vehicles can be developed and tested with traffic simulations in particular. In an urban area, the traffic situation is often very complex due to the large number and variety of other road users. So that simulation results from urban traffic areas can still be used for validation, it is necessary that all modeled road users move realistically and interact with each other.

The US 9,495,874 B1 relates to a method and a device to generate one or more behavior models that are used by autonomous vehicles to predict the behavior of a detected object. In this case, the autonomous vehicle can detect and record a behavior of the object by using one or more sensors. The autonomous vehicle can then communicate the recorded behavior of the object to a server that is designed to determine behavior models. The server can determine the respective behavior model according to a predetermined object classification, according to relevant actions of the object, as well as according to the object's perceived properties of the environment.

The WO 2020/079685 A1 also relates to a method for training a model for generating simulation data for training an automatic driver model, comprising the steps: analyzing real data collected from a driving environment in order to identify a plurality of environment classes and a plurality of agent classes, and a variety of motion profile classes; creating a training environment according to an environment class; and in at least one training iteration: generating, by a simulation generation model, a simulated driver environment according to the training environment and according to the plurality of generated training agents, wherein each of the training agents is linked to one of the plurality of agent classes and linked to one of the plurality of motion profiles becomes; determining simulated driving data from the simulation environment; and modifying at least one model parameter of the simulation generation model to minimize a difference between a simulation statistical fingerprint and a real statistical fingerprint, wherein the statistical fingerprint is generated by simulated driving data and the real statistical fingerprint is calculated from real data.

The object of the invention is, by creating realistic models of cyclists in a simulation environment, to use these for the validation and testing of automated vehicles in order to make their driving behavior safer and less unnecessarily defensive.

The invention results from the features of the independent claims. Advantageous developments and refinements are the subject matter of the dependent claims.

A first aspect of the invention relates to a method for testing and/or validating an automated driving control function of a vehicle, a simulation being carried out and the simulation including a cyclist among other road users, the movements of the cyclist being simulated on the basis of an empirical performance model of the cyclist, wherein the performance model maps modeled torque inputs by muscular movement of the cyclist, and in the case of an electrified bicycle by the cyclist additionally by the power of the electric drive of the bicycle, depending on environmental variables, to a resulting movement of the cyclist, and wherein the automated driving control function is linked to the simulation, so that the automated ride control function can react to the simulated movements of the cyclist.

The driving control function is in particular a function for automatically controlling the vehicle, for example for specifying a steering movement, braking actions, acceleration, or actuating another component of the vehicle, such as controlling a signal horn, etc.; each action takes place in particular in response to an external command or as a reaction to the surroundings of the vehicle. The environment is recorded in particular by sensors of the vehicle. This driving control function should be tested and/or validated in the simulation.

In order to be able to depict the acceleration and deceleration of cyclists more realistically in such a simulation, cyclists are physically modeled according to the invention and move on under power control. Such a property of the virtual cycling models primarily, and therefore essentially, only affects acceleration and deceleration in the longitudinal direction of the cyclist, i.e. in his current direction of travel, and consequently on the speed and in a further integration stage on the position of the cyclist at a given time.

The empirical performance model relies in particular on the performance of the cyclist himself, which is achieved through muscle work. When modeling e-bikes, this power can be increased by the power provided by the drive. In the simplest case, the statistical relationship between weight and size is used for the currently required power for constant speed or acceleration. Statistics from medical and/or sports-scientific performance diagnostics can be used to determine, among other things, maximum values for performance depending on weight and size. From this, values can be derived for the performance, preferably in the unit "Watt", in the comfort zone, e.g. in normal city traffic without being in a hurry and what performance can be provided over a certain period of time when the cyclist is in a hurry. Speeds can then be calculated based on the performance and environmental variables such as the incline of the road. Another way of estimating the relationship between the power input and the prevailing conditions is to use machine learning.

The performance model is empirical, i. H. it is based on real data. The data can come from various and different sources.

Real cyclists equipped with sensor(s) who move in city traffic can be considered here. A (d)GPS tracker, a power meter such as "measuring pedals" or a "measuring crank" can be used as sensors, or a "measurement directly in the bottom bracket" can be used. Other sensors that supplement the database can be: speedometers, acceleration and gyro sensors, cameras, heart rate monitors. From this, conclusions can be drawn as to how much effort the cyclist is making for the current performance. This information can be used to derive the performance that another cyclist with a different performance level would perform in a similar place with a similar intention (hurry, normal). A group of cyclists can be fitted with sensors specifically for this purpose. Alternatively, databases from fitness tracking platforms can be used. Bike rides with a sporty-motivated background are mainly stored on these platforms, but also commute rides and everyday rides. Due to the sheer number of bike rides stored there with different sensor setups, different behavior and realistic performance values can be derived depending on different driver profiles in different traffic environments. If a database is already available in which only GPS data is available, performance values can be back-calculated or estimated using the speed and the incline. For this purpose, typical formulas of the equations of motion can be used again. If weight values are also available, the estimation can be carried out more precisely. If necessary, wind data can be included in the estimation in order to further increase the quality. If a database has been built up as explained above, this can be used to derive the realistic behavior with regard to acceleration and deceleration. Machine learning methods are preferably used for this purpose. With so-called “supervised learning”, for example, an artificial neural network is trained in such a way that it uses the input parameters provided to estimate the performance that the cyclist delivers in the current situation (can be derived from the input values). This performance also includes the braking performance, for example when driving downhill quickly. The input data for the artificial neural network are then in particular the above parameters of the cyclist (weight, age, level of performance, intention of the cyclist (e.g. in a hurry, emergency braking, normal braking to a standstill, leisurely ride) and parameters that describe the current traffic environment. However, at least the gradient of the roadway, a curve radius and, in particular, the nature of the ground are preferred. If such an artificial neural network has been sufficiently trained with the database, it can be entered into the behavior model. From the performance received as a response from the artificial neural network the speed of the cyclist can be calculated with the formulas mentioned above, or accelerations and decelerations then result from the performance rendered.Furthermore, effects can be taken from the database that can arise during very slow journeys, especially uphill : The "snaking", i.e. deliberately driving in a serpentine manner, in order to artificially reduce the incline. Here it can be determined for different driver types, speeds and gradients when this effect occurs and to what extent, if suitable sensors were used during the recording. This can then be installed as an additional option in the simulated behavior of the cyclists and taken into account in the simulation if the determined limit values for the respective driver type and parameters are exceeded. Machine learning can also be used to determine the limit values (speed, incline). be used according to the above principle. This allows the estimation to be generalized and cyclist settings to be applied to all parameters. Furthermore, the database described above can be replaced by statistical evaluations of medical or sports science databases for fitness tests. In general, services can also be derived from this database that are realistically provided for different cyclists in different situations: For example, the comfort zone in which the cyclist moves from the starting point to the destination without great physical exertion or the performance that can be provided by him for a short time e.g. to cope with an incline.

It is an advantageous effect of the invention that, based on an evaluation of real data, simulation models of cyclists can be controlled in a performance-oriented manner. This means that the power that a cyclist transfers to the pedals is used to calculate the cyclist's longitudinal acceleration - resulting in a realistic representation of the cyclist's movement. The movement sequences of cyclists are therefore advantageously modeled precisely even on steep roads, in strong winds, on unusual or uneven road surfaces and when there are changes in inclination. It is precisely in these situations that the use of a simple model of a cyclist based solely on the maximum acceleration and a maximum speed of the respective cyclist would not be sufficiently realistic.

More realistic scenarios can be created by incorporating performance-oriented cycling models into a simulation environment used to train and/or test an automated vehicle. This makes it possible, among other things, for a computing unit of an automated vehicle to be able to recognize before a steep climb that the cyclist in front will lose a great deal of speed and will possibly take a wavy line. Furthermore, the computing unit can determine a probability that the automated vehicle will be overtaken again by a cyclist on a steep downhill slope, e.g. in a zone with a greatly reduced speed limit, such as 30 km/h. With established methods for validation and training or with criticality metrics, it can be ensured that even the scenarios with realistic cyclists can be solved by the automated vehicle (in the simulation) sufficiently to meet the safety requirements. Thus, a further increase in the degree of reality is possible compared to pure training of the AI module with speed-based cyclist agents. Test drives and developers are not endangered.

1. a) Jegliche Situationen, in der ein Radfahrer beschleunigt. Insbesondere wenn nicht-regelkonformes Verhalten simuliert werden soll, z.B. ein Wechsel vom Bürgersteig auf eine Straße oder umgekehrt. In diesen Situationen bremst und beschleunigt ein Radfahrer in der Regel sehr oft, was mit einfacheren Modellen eines Radfahrers nicht realistisch modelliert werden kann.
2. b) Steigungen und Gefälle: Durch die Schwerkraft kommt es zu erheblichen Geschwindigkeitsänderungen, da deutlich mehr Leistung oder weniger Leistung benötigt wird, um den Radfahrer vorwärts zu bewegen. In bisherigen Modellen wird dieses nicht berücksichtigt.
3. c) Gegen/Rückenwind: Dieser kann ebenfalls berücksichtigt werden. Was für motorisierte Personenkraftwagen / Lastkraftwagen im Allgemeinen vernachlässigt werden kann, ist für Radfahrer ein entscheidender Faktor, der auch zu abrupten und unerwarteten Geschwindigkeitsänderungen führen kann, was wiederum zu sicherheitskritischen Situationen mit dem automatisierten Fahrzeug führen könnte.
4. d) Unterschiedliche Typen von Radfahrern können einfacher simuliert und realistischer simuliert werden, da mehr Parameter angepasst werden können, die nicht nur Einfluss auf die maximale Geschwindigkeit haben. Z.B. kann ein langsamer Radfahrer trotzdem sehr schnell bergab fahren.

The following situations, among others, can therefore be mapped more realistically in the simulation:

1. a) Any situation where a cyclist accelerates. Especially when behavior that does not conform to the rules is to be simulated, eg changing from the sidewalk to a street or vice versa. In these situations, a cyclist typically brakes and accelerates very often, which cannot be realistically modeled with simpler models of a cyclist.
2. b) Uphill and downhill gradients: Gravity causes significant changes in speed as significantly more power or less power is required to propel the cyclist forward. This is not taken into account in previous models.
3. c) Headwind/tailwind: This can also be taken into account. What can be neglected for motorized passenger cars/trucks in general is a crucial factor for cyclists, which can also lead to abrupt and unexpected changes in speed, which in turn could lead to safety-critical situations with the automated vehicle.
4. d) Different types of cyclists can be simulated more easily and more realistically, since more parameters can be adjusted, which not only affect the maximum speed. For example, a slow cyclist can still ride downhill very quickly.

According to an advantageous embodiment, the movements of the cyclist resulting from the performance model are accelerations in the longitudinal direction of the cyclist and kinematic variables obtained from the integration of the accelerations over time are defined in the longitudinal direction of the cyclist, while steering movements and a route taken by the cyclist via social Force models and/or single track models are simulated.

According to this embodiment, the longitudinal movement and the lateral movement of the cyclist are considered separately, the longitudinal movement being simulated based on the modeled performances and the lateral movement based on well-known methods such as the mentioned social forces models and/or single track models. The connection between the performance and the primary effect of this on the movement thus becomes advantageous in the instantaneous longitudinal direction of the cyclist.

According to a further advantageous embodiment, the performance model includes a total mass of the cyclist including his bicycle and luggage, and a body size of the cyclist or a size aerodynamically equivalent to the body size, in particular an effective frontal area.

In this way, the influence of inclines/slopes (hereinafter referred to generally as the "increase" of the road or lane), as well as the current air resistance, can be taken into account.

• - Art des Fahrrads,
• - körperliche Kondition des Radfahrers,
• - Alter des Radfahrers,
• - Verhaltensparameter des Radfahrers, insbesondere Eile;

According to a further advantageous embodiment, the performance model includes at least one of the following parameters:

• - type of bike,
• - physical condition of the cyclist,
• - age of the cyclist,
• - Behavioral parameters of the cyclist, especially haste;

The type of bike indicates what the bike was designed for, such as speed-optimized bikes or terrain-optimized bikes. Possible types of bicycle are therefore in particular racing bikes, mountain bikes, gravel bikes, e-bikes (in particular with power assistance up to 25 km/h), trekking bikes, city bikes, triathlon bikes, recumbent bikes, cargo bikes, etc.; the physical condition of the cyclist serves in particular to estimate the possible maximum achievable performance through muscle power of the cyclist. The age of the cyclist is also used to model the dynamic range, the possible force input on the pedals and the behavior of the cyclist. In particular, the cyclist's haste is part of the behavior, which indicates what level of aggressive driving style and speed the cyclist can expect, as well as whether shortened reaction times can be present.

• - Anstieg einer vom Radfahrer aktuell virtuell befahrenen Straße,
• - Wind, der auf den Radfahrer wirkt,
• - Reibungseinfluss einer Fahrbahn, die vom Radfahrer in der Simulation virtuell aktuell befahren wird,
• - Verkehrszeichen,
• - Simulierte Verkehrssituation in Bezug auf andere Verkehrsteilnehmer,
• - Außentemperatur,
• - Zustand der Fahrbahnoberfläche, insbesondere Nässe, Glätte, Trockenheit;

According to a further advantageous embodiment, the environment variables have at least one of the following:

• - ascent of a road currently virtually traveled by the cyclist,
• - wind acting on the cyclist,
• - Influence of friction of a roadway that is currently being traveled virtually by the cyclist in the simulation,
• - traffic signs,
• - Simulated traffic situation in relation to other road users,
• - outside temperature,
• - Condition of the road surface, especially wet, slippery, dry;

According to a further advantageous embodiment, during a steep uphill climb for the cyclist, the simulation checks whether, according to the performance model, the speed of the cyclist falls below a predetermined limit value, with the cyclist being converted into a pedestrian pushing his bicycle in the simulation if this limit value is not reached becomes.

According to a further advantageous embodiment, when there is a steep uphill climb in the simulation, the cyclist deliberately driving in serpentines is simulated if a predetermined condition is met.

Depending on the total mass, the condition of the cyclist, the type of bicycle and the incline of the road, cyclists may have a tendency to deliberately meander in order to reduce the current incline. Namely, on steep climbs, the cyclist's speed can become very low, such that a cyclist can tend to "snake" in order to climb the hill without falling over. Furthermore, the simulation can take into account that the cyclist dismounts and becomes a "pedestrian". A simulation that takes into account the above effects can advantageously be used to test how the automated vehicle would behave in relation to such behavior on the part of the cyclist.

According to a further advantageous embodiment, braking of the cyclist is provided in the simulation in the event of a steep ascent downhill, so that a maximum, predefined speed of the cyclist is not exceeded in the simulation.

On steep descents, cyclists will reach a higher speed than could be observed with a simple non-performance-oriented model, due to the downhill force, which is also advantageously taken into account in the simulation, since in the latter a constant speed would simply be assigned. To increase realism, a maximum speed can be set at which cyclists actively brake to slow down the speed.

According to a further advantageous embodiment, the performance model takes into account a Change in speed of the cyclist when changing from a pavement to the street and vice versa.

According to a further advantageous embodiment, the performance model includes forces and accelerations of the cyclist when cornering, with the forces and accelerations being determined in particular as a function of the total mass of the cyclist including luggage and bicycle.

The modeled total mass of the cyclist can also be used to better estimate the maximum speed at which a cyclist can corner. Furthermore, if the speed is too fast, the turning radius can be adjusted so that the cyclist is carried further out. Braking and acceleration before and after curves can thus also be described realistically.

According to a further advantageous embodiment, the performance model takes into account non-rule-compliant behavior of the cyclist, which includes a sudden change in speed.

According to a further advantageous embodiment, the performance model is executed by an artificial neural network, the artificial neural network being trained with an empirical database, the input variables of the artificial neural network being the environmental variables and parameters of the performance model and the output variable of the artificial neural network being those of the cyclist and/or the power applied to the bicycle when driving the bicycle.

Another aspect of the invention relates to a simulation system for testing and/or validating an automated driving control function of a vehicle, the simulation system being designed to carry out a simulation, the simulation including a cyclist among other road users, the movements of the cyclist based on a performance model of the cyclist are simulated, with the performance model mapping modeled torque inputs through muscle movement of the cyclist, and in the case of an electrified bicycle of the cyclist additionally through the power of the electric drive of the bicycle, depending on environmental variables, to a resulting movement of the cyclist, and the automated driving control function with linked to the simulation so that the automated ride control function can react to the simulated movements of the cyclist.

According to a further advantageous embodiment, the simulation system has an interface to a real vehicle in order to replace sensor signals from sensors of the vehicle with simulated sensor signals generated by the simulation.

According to a further advantageous embodiment, the driving control function of the vehicle is implemented as a module in the simulation itself.

Advantages and preferred developments of the proposed simulation system result from an analogous and analogous transfer of the statements made above in connection with the proposed method.

Further advantages, features and details result from the following description, in which at least one exemplary embodiment is described in detail-if necessary with reference to the drawing. Identical, similar and/or functionally identical parts are provided with the same reference symbols.

• 1: Ein Simulationssystem zum Ausführen eines Verfahrens zum Testen und/oder Validieren einer automatisierten Fahrsteuerfunktion eines Fahrzeugs gemäß einem Ausführungsbeispiel der Erfindung.

It shows:

• 1 : A simulation system for executing a method for testing and/or validating an automated driving control function of a vehicle according to an embodiment of the invention.

The representations in the figure are schematic and not to scale.

1 shows a simulation system 5 on which a method for testing and validating an automated driving control function of the counterpart of a real vehicle 1 is carried out virtually. A virtual cyclist 3 is provided in the simulation, with the movements of the cyclist 3 being simulated on the basis of an empirically determined performance model of the cyclist 3 . The performance model includes modeled torque entries due to muscle movement of the cyclist 3. These torque entries are mapped to a resulting movement of the cyclist 3 depending on environmental variables. The automated driving control function is implemented directly in the simulation, so that the automated driving control function can react to the simulated movements of the cyclist 3 . In the figure, a situation is shown in which the digital twin of the vehicle 1 is driving behind the virtual cyclist 3, the cyclist 3 being located immediately in front of an increasing gradient of the road. The empirical model contains data that assigns a corresponding decrease in speed to a gradient angle resulting from a change in altitude, against which the cyclist 3 exerts muscle power brought power to the bike acts. The simulation can thus take into account the decrease in speed of the cyclist 3 and the reaction of the driver assistance system of the vehicle 1 to it can be tested and validated before delivery.

Although the invention has been illustrated and explained in more detail by means of preferred exemplary embodiments, the invention is not restricted by the disclosed examples and other variations can be derived therefrom by a person skilled in the art without departing from the protective scope of the invention. It is therefore clear that a large number of possible variations exist. It is also understood that the exemplary embodiments given are really only examples and should not be construed as limiting in any way the scope, applications or configuration of the invention. Rather, the preceding description and the description of the figures enable the person skilled in the art to concretely implement the exemplary embodiments, whereby the person skilled in the art, knowing the disclosed inventive concept, can make a variety of changes, for example with regard to the function or the arrangement of individual elements mentioned in an exemplary embodiment. without departing from the scope of protection defined by the claims and their legal equivalents, such as further explanations in the description.

Reference List

1

vehicle

3

cyclist

5

simulation system

QUOTES INCLUDED IN DESCRIPTION

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

Patent Literature Cited

• US9495874B1 [0003]
• WO 2020079685 A1 [0004]

Claims (15)

Method for testing and/or validating an automated driving control function of a vehicle (1), wherein a simulation is carried out and the simulation includes a cyclist (3) among other road users, wherein the movements of the cyclist (3) are based on an empirical performance model of the cyclist ( 3) are simulated, with the performance model modeling torque inputs through muscle movement of the cyclist (3), and in the case of an electrified bicycle of the cyclist (3) additionally through the power of the electric drive of the bicycle, depending on environmental variables on a resulting movement of the cyclist (3 ) and wherein the automated ride control function is linked to the simulation such that the automated ride control function can respond to the simulated movements of the cyclist (3). procedure after claim 1 , where the movements of the cyclist (3) resulting from the performance model are accelerations in the longitudinal direction of the cyclist (3) and kinematic quantities in the longitudinal direction of the cyclist (3) obtained from the integration of the accelerations over time are defined during steering movements and a route taken of the cyclist (3) can be simulated using social force models and/or single-track models. Method according to one of the preceding claims, wherein the performance model comprises a total mass of the cyclist (3) including his bicycle and luggage, and a body size of the cyclist (3) or a size aerodynamically equivalent to body size, in particular an effective frontal area. Method according to one of the preceding claims, wherein the performance model comprises at least one of the following parameters: - type of bike, - physical condition of the cyclist (3), - Age of the cyclist (3), - Behavioral parameters of the cyclist (3), in particular haste; A method according to any one of the preceding claims, wherein the environmental variables include at least one of the following: - ascent of a road currently virtually traveled by the cyclist (3), - wind acting on the cyclist (3), - Influence of friction on a roadway that the cyclist (3) is virtually currently driving on in the simulation, - traffic signs, - Simulated traffic situation in relation to other road users, - outside temperature, - Condition of the road surface, especially wet, slippery, dry; procedure after claim 5 , whereby in the event of a steep uphill climb for the cyclist (3), the simulation checks whether, according to the performance model, the speed of the cyclist (3) falls below a specified limit value, with the cyclist (3) in the simulation falling below this limit value in transformed into a pedestrian pushing his bicycle. procedure after claim 5 , with the conscious driving of wavy lines by the cyclist (3) being simulated on a steep uphill climb in the simulation if a predetermined condition is met. procedure after claim 5 , wherein braking of the cyclist (3) is provided in the case of a steep downhill climb in the simulation, so that a maximum, predetermined speed of the cyclist (3) is not exceeded in the simulation. Method according to one of the preceding claims, in which the performance model takes into account a change in speed of the cyclist (3) when changing from a pavement to the road and vice versa. Method according to one of the preceding claims, wherein the performance model includes forces and accelerations of the cyclist (3) when cornering, the forces and accelerations being determined in particular as a function of the total mass of the cyclist (3) including luggage and bicycle. Method according to one of the preceding claims, wherein the performance model takes into account non-compliant behavior of the cyclist (3) which comprises a sudden change in speed. Method according to one of the preceding claims, wherein the performance model is executed by an artificial neural network, the artificial neural network being trained with an empirical database, the input variables of the artificial neural network being the environmental variables and parameters of the performance model and the output variable of the artificial neural network is the power applied to the bicycle by the cyclist (3) and/or the drive of the bicycle. Simulation system (5) for testing and/or validating an automated driving control function of a vehicle (1), the simulation system (5) being designed to carry out a simulation, the simulation including a cyclist (3) among other road users, the movements of the Cyclist (3) are simulated on the basis of a performance model of the cyclist (3), the performance model modeling torque inputs through muscle movement of the cyclist (3), and in the case of an electrified bicycle of the cyclist (3) additionally through the power of the electric drive of the bicycle, depending on environmental variables on a resulting movement of the cyclist (3), and wherein the automated driving control function is linked to the simulation, so that the automated driving control function can react to the simulated movements of the cyclist (3). Simulation system (5) according to Claim 13 , wherein the simulation system has an interface to a real vehicle (1) in order to replace sensor signals from sensors of the vehicle (1) with simulated sensor signals generated by the simulation. Simulation system (5) according to Claim 13 , wherein the driving control function of the vehicle (1) is implemented as a module in the simulation itself.

Images