DE102018217192A1 - Method and system for providing a coefficient of friction information to a vehicle - Google Patents

Abstract

The invention relates to an improved method for providing a coefficient of friction information for a traffic area section (200) to a vehicle (300). With the method, the most precise possible information about the coefficient of friction can be achieved through reinforcing machine learning. The method provides for determining a coefficient of friction prediction by an AI module (120), to which at least one input variable (400, 500, 600) that can be assigned to the traffic area section (200) is supplied. In addition, the method comprises determining a weighted deviation between the coefficient of friction prediction and a coefficient of friction measurement (700) for a common point in time or time period, and supplying the determined weighted deviation to the AI module (120) and adapting the determination of the friction value prediction on the basis of the weighted Deviation. The invention further relates to a system (100) for providing a coefficient of friction information for a traffic area section (200) to a vehicle (300). Method and system for providing a coefficient of friction information to a vehicle.

Description

The present invention relates to a device for providing a coefficient of friction information for a traffic area section to a vehicle. The invention further relates to a system for providing a coefficient of friction information for a traffic area section to a vehicle.

State of the art

The coefficient of friction between a vehicle, in particular its tire, and a traffic area section, such as a roadway, influences the dynamic driving limits of a vehicle with regard to longitudinal and transverse guidance. In a conventional vehicle driven by a human driver, the driver may e.g. Using empirical values, brief braking, visual perception or the like, at least to some extent, determine whether the current coefficient of friction is rather high or rather low. For example, the current coefficient of friction on a well-developed and clean road surface and when it is dry can be rather high and thus promote good static friction between the vehicle and the road surface, whereas wetness, black ice, dirt on the road surface or the like can adversely affect the coefficient of friction.

In the case of a partially or fully autonomous driving operation of a vehicle, it is more difficult to estimate the current coefficient of friction of a traffic area section, since the information of the human vehicle driver is at least partially eliminated. To counter this problem, for example, the WO 2016/120092 A1 a database-based coefficient of friction map, in which information sent by sending vehicles is received and stored in the database, the information comprising at least the data describing the coefficient of friction of a road segment, the location data describing the geometric position of this segment of the road, and the time data describing the time of determination of the coefficient of friction data, and in data stored in the database can be called up by receiving vehicles. Although the vehicle can thus be provided with useful information on the current coefficient of friction of a traffic area section, there is a desire to be able to provide a vehicle with even more precise coefficient of friction data.

Disclosure of the invention

Embodiments of the invention provide an improved method and an improved system for providing a coefficient of friction information to a vehicle according to the independent claims. Advantageous further developments of the invention result from the dependent claims, the description and the accompanying figures.

• - Es wird eine Reibwertprädiktion durch ein KI-Modul bestimmt, dem eine einzige oder eine Mehrzahl von dem Verkehrsflächenabschnitt zuordbaren Eingangsgrößen zugeführt werden. Der Reibwert lässt sich ohne eine lokale, spezialisierte Messeinrichtung üblicherweise nicht genau, sondern nur näherungsweise bestimmen. Deshalb wird der Reibwert hier prädiziert, idealerweise für einen zukünftigen Zeitpunkt oder Zeitabschnitt, in dem das Fahrzeug den Verkehrsflächenabschnitt voraussichtlich erreichen wird.
• - Es wird eine Abweichung zwischen der Reibwertprädiktion und einer Reibwertmessung für einen gemeinsamen oder ähnlichen Zeitpunkt, Zeitabschnitt usw. bestimmt und diese Abweichung bewertet. In anderen Worten wird aus der Abweichung eine Art Belohnung, die positiv oder negativ sein kann, für das KI-Modul bestimmt, um insbesondere eine möglichst kleine Abweichung zu belohnen und für eine zu große Abweichung zu bestrafen. Unter Reibwertmessung kann zwar, aber muss hier keine gesonderte Reibwertmessung durch ein mit spezieller Messtechnik ausgestattetes Messfahrzeug verstanden werden. Vielmehr kann die Reibwertmessung durch ein oder mehrere Fahrzeuge erfolgen, die den betrachteten Verkehrsflächenabschnitt befahren oder bereits befahren haben bzw. in Nachbarschaft dazu passieren bzw. bereits passiert haben. Die Fahrzeuge können ein elektronisches Steuergerät oder einen elektronischen Steuergeräteverbund aufweisen, der aus z.B. Daten einer Fahrdynamikregelung, einem Regensensor usw. einen Reibwert des Verkehrsflächenabschnitts fahrzeugseitig bestimmen bzw. schätzen, also eine indirekte Messung durchführen.
• - Zuführen der bestimmten bewerteten Abweichung an das KI-Modul und Anpassen der Bestimmung, z.B. einer Bestimmungsmethodik, der Reibwertprädiktion auf Basis der bewerteten Abweichung. In anderen Worten kann aufgrund der bewerteten Abweichung, einer Art Belohnung oder Bestrafung, das KI-Modul dazu veranlasst werden, eine Methode zum Bestimmen der Reibwertprädiktion anzupassen und ein zukünftiges Bestimmen dann unter Berücksichtigung dieser Anpassung durchzuführen.

The proposed method can be carried out, for example, with a computer system, which can have a processor, a memory for program instructions, a data interface, etc. The computer system can be, for example, a server device to which a plurality of vehicles can have access via a data network. The method is particularly suitable for providing a vehicle with a coefficient of friction information for a traffic area section, for example a roadway, a road, or other traffic area that is mapped as far as possible. The process involves the following steps:

• A coefficient of friction prediction is determined by an AI module, to which a single or a plurality of input variables that can be assigned to the traffic area section are supplied. Without a local, specialized measuring device, the coefficient of friction cannot usually be determined exactly, but only approximately. The coefficient of friction is therefore predicted here, ideally for a future point in time or time period in which the vehicle is likely to reach the traffic area section.
• - A deviation between the coefficient of friction prediction and a coefficient of friction measurement for a common or similar point in time, time period, etc. is determined and this deviation is assessed. In other words, the deviation is used to determine a kind of reward, which can be positive or negative, for the AI module, in particular to reward the smallest possible deviation and to punish for a deviation that is too large. Coefficient of friction measurement can, but does not have to be understood here as a separate measurement of the coefficient of friction by a measuring vehicle equipped with special measuring technology. Rather, the coefficient of friction measurement can be carried out by one or more vehicles which are or have already traveled on the traffic area section under consideration or pass in the vicinity or have already passed. The vehicles can have an electronic control unit or an electronic control unit network, which determine or estimate a coefficient of friction of the traffic area section on the vehicle side, for example from data of a vehicle dynamics control, a rain sensor, and thus carry out an indirect measurement.
• - supplying the determined evaluated deviation to the AI module and adapting the determination, for example a determination methodology, the coefficient of friction prediction based on the assessed deviation. In other words, based on the assessed deviation, a kind of reward or punishment, the AI module can be made to adapt a method for determining the coefficient of friction prediction and then to carry out a future determination taking this adaptation into account.

The method can generate more precise coefficient of friction information or can continuously improve during the operation of a computer system operating according to the method. The AI module receives direct feedback through the evaluation of the deviation, i.e. the reward or punishment, so that the accuracy of the friction coefficient prediction can be gradually improved. This enables even more precise information on the coefficient of friction to be made available.

In a further development, the deviation can be assessed as positive if it is smaller than a threshold value, and as negative if it is greater than the threshold value. The deviation can be recorded on any scale, for example with values between 0 and 1, where 0 means no deviation and 1 no agreement, that is to say the greatest possible deviation of the comparison values.

According to a further development, the AI module can provide the coefficient of friction prediction for the vehicle before it reaches the traffic area section. As a result, the vehicle, which can also be an at least partially automated vehicle, can use this information for e.g. consider route planning, trajectory planning, speed planning, generally determining a driving strategy, etc. In particular, a traffic area section that is iced up based on the coefficient of friction prediction can be avoided, in particular bypassed.

In a further development, the input variable can be supplied to at least one artificial neural network (KNN) of the AI module. This can be formed in one or more layers, folding, etc. It is possible that the KNN is trained with training data records, it being possible for the training data records to have been created using the method described here, possibly also a slightly modified method.

According to a further development, the coefficient of friction prediction can include a change in the coefficient of friction over time. In other words, the AI module can be set up to assign a respective friction coefficient to a time series. In this way, for example, it can be determined even more precisely for the vehicle when the traffic area section can be used with which driving strategy.

In a further development, the input variable can include vehicle-side friction coefficient data of the traffic area section. As indicated above, one or more vehicles that pass or have passed the traffic area section can determine or estimate a coefficient of friction on the vehicle side and can estimate this via e.g. provide a radio connection as an input variable for the AI module. This can provide a good basis for predicting the coefficient of friction as accurately as possible.

According to another development, the input variable can include road condition data from a road sensor. This can e.g. detect a number of vehicles that travel on the traffic area section, provide a road temperature, etc. In this way, the coefficient of friction can be predicted even more precisely.

In a further development, the input variable can include weather-related environmental data of the traffic area section. These can either be through local or nearby weather measuring points, a weather service or the like e.g. be made available via the data interface. For example, a precipitation amount, a fog density, a rain amount, a snowfall amount, sun intensity or the like can be recorded. In this way, the coefficient of friction can be predicted even more precisely.

In another development, the input variable can also include information about the development and / or planting in the area of the traffic area section or adjacent thereto. this can be used, for example, to determine whether or to what extent the traffic area section is exposed to wind, sun, etc., in order to be able to determine from this how long it will take for the moist or wet traffic area section to dry, etc. The coefficient of friction can be predicted even more precisely.

According to a further development, an intermediate layer of the traffic area section can be determined on the basis of at least the weather-related environmental data as a further input variable for the AI module. The intermediate layer can be determined by a first part of the KI module and made available to a second part of the KI module provided for predicting the coefficient of friction. Another system or AI module for determining the interlayer can also be provided. In this context, a medium, such as water, snow, ice, leaves, crushed stone or water, can be found underneath an intermediate layer on a surface of the traffic area section similar, to be understood. The KI module can also be set up to determine a temporal change in the intermediate layer, for example a time series, after the determination of the intermediate layer and to use this as an input variable for the coefficient of friction prediction. In this way, the coefficient of friction can be predicted even more precisely.

The invention also relates to a system for providing a coefficient of friction information for a traffic area section to a vehicle. The system has a server device, which has an artificial intelligence module, KI module, and a data interface, the server device being set up to determine a coefficient of friction prediction by the AI module, to which a single or a plurality of data is transmitted via the data interface input variables assignable to the traffic area section are supplied. Furthermore, the KI module is set up to determine an evaluated deviation between the coefficient of friction prediction and a coefficient of friction measurement for a common point in time and to adapt the determination of the coefficient of friction prediction on the basis of the evaluated deviation.

The system can be operated, for example, with the methods explained above in one or more of the described embodiment variants and can therefore offer the advantages explained above.

Further measures improving the invention are described in more detail below together with the description of the preferred exemplary embodiments of the invention with reference to figures.

Figure list

• 1 eine schematische Übersicht über ein System zum Bereitstellen einer Reibwertinformation an ein Fahrzeug und
• 2 ein Blockdiagramm eines Systems zum Bereitstellen einer Reibwertinformation an ein Fahrzeug.

In the following, advantageous exemplary embodiments of the invention are described in detail with reference to the accompanying figures. Show it:

• 1 a schematic overview of a system for providing a coefficient of friction information to a vehicle and
• 2nd a block diagram of a system for providing a coefficient of friction information to a vehicle.

The figures are only schematic and are not to scale. In the figures, the same, equivalent or similar elements are provided with the same reference numerals throughout.

Embodiments of the invention

1 shows a system 100 which is suitable for the electronic provision of a coefficient of friction information, for example a coefficient of friction, a time series of a coefficient of friction, etc. This coefficient of friction information relates in particular to a coefficient of friction of a surface of a traffic area section 200 , for example, a mapped road or a road section, and a plurality of vehicles 300 provided centrally, only one of which is shown here as an example.

The system 100 has a server 110 , which has a memory for program instructions, a processor for executing the program instructions, etc., and via a communication interface to a data network, which in particular enables a radio connection. In the server 110 is an artificial intelligence module 120 , Kl module (see 2nd ), for example implemented by program instructions, which comprises a machine learning model set up for reinforcing learning in the form of an artificial neural network (KNN). The KNN has a single-layer, multi-layer, folding, etc. structure and ideally has already been trained with training data sets, although this can also be dispensed with in the case of reinforcing learning. The AI module 120 can also generate training data sets itself in order to make them available to other applications.

The system 100 are different input variables via the communication interface 400 , 500 , 600 made available. Here are the input variables 400 , 500 , 600 the AI module 120 , for example as input variables of an input layer of the KNN.

According to 1 become the system 100 the input variables exemplarily in the form of weather data 400 for the traffic area section 200 are determined and / or provided by weather stations, local weather sensors, weather services or the like from vehicle-side sensor data 500 that are detected by vehicle sensors that cover the traffic area section 200 drive or pass nearby and from road sensor data 600 provided that in the traffic area section 200 are arranged close to the surface or integrated in them or arranged adjacent to them.

2nd shows the system 100 in a block diagram. The server setup 110 includes the AI module 120 and a reward determination module 130 includes, here separately for better illustration for the AI module 120 is shown, but also part of the reinforcing machine learning model or the KNN of the AI module 120 can be.

The reward determination module 130 includes two inputs for input variables and one output for an output variable. According to 2nd there is a first feedback, for example a first data path, from one Output of the AI module 120 to the first entry of the reward determination module 130 so that an output variable of the AI module 120 the reward determination module 130 can be supplied. A second feedback, for example a second data path, exists between the output of the reward determination module 130 to another input of the KL module 120 . Thus, the AI module 120 and the reward determination module 130 by exchanging data and considering a determination result of the other module mutually influence what reinforcing learning, especially of the AI module 120 , can be used.

The AI module 120 is set up the input variables 400 , 500 , 600 to get that depending on the accuracy of the system 100 may include more or less different data from one or more systems. Exemplary here are the AI module 120 the weather data 400 supplied, which contain, for example, a rainfall, a rainfall, a snowfall, a sun intensity, a wind force, a fog density, etc. The AI module is also exemplary 120 the on-board sensor data 500 supplied, which contain, for example, an estimated or determined coefficient of friction on the vehicle side, raw sensor data from a vehicle dynamics control system, etc. The AI module is further exemplary 120 the street sensor data 600 supplied, for example, a street temperature, a number of the traffic area section 200 vehicles driving 300 etc. included. In addition, the AI module 120 exemplary also supplied (not specified) topographical properties, which are a topographical property in the vicinity of the traffic area section 200 , such as buildings and / or vegetation, which, for example, lead to the formation of shadows on the traffic area section 200 lead and thus, for example, impair the evaporation of water.

The AI module 120 is also set up on the basis of the input variables 400 , 500 , 600 a coefficient of friction prediction for the traffic area section 200 to determine and output this as an output variable. On the one hand, the coefficient of friction prediction becomes the reward determination module 130 supplied as an input variable and on the other hand to the vehicle 300 made available.

According to 2nd is the reward determination module 130 set up to obtain two input variables, of which a first input variable is the output variable of the KI module that includes the coefficient of friction prediction 120 is and a second input variable data of a coefficient of friction measurement 700 at the traffic area section 200 includes. The coefficient of friction measurement 700 takes place here indirectly via the majority of vehicles, which also include the input variables 500 provide. That is, the vehicles that make up the traffic area section 200 drive or suggest pass or have already done so, e.g. provide data from a vehicle dynamics control, a rain sensor, etc., to determine or estimate the coefficient of friction therefrom or already do this on the vehicle side in an electronic control unit or control unit network. Alternatively or additionally, a local measuring device can also be used for measuring the coefficient of friction. This becomes the second input variable 700 for the reward determination module 130 formed and fed to this.

The reward determination module 130 is also set up to, for example, amount, quantitative, qualitative, etc. deviation between the two input variables, namely the coefficient of friction prediction of the AI module 120 and the coefficient of friction measurement 700 , determine and evaluate. This deviation or its assessment can be on any scale. As an example, an evaluation of the deviation from 0 can indicate that there is no deviation, that is to say both input variables match. Likewise, an assessment of the deviation from 1 can indicate that the deviation is so large that, in particular, that of the AI module 120 input variable is practically unusable.

Through the feedback between the reward determination module 130 and the AI module 120 this receives the rating of the reward determination module 130 as direct feedback, depending on the value, as a reward or punishment for the determination result of the coefficient of friction prediction by the AI module 120 can be interpreted so that the AI module 130 can adapt the determination, for example a determination methodology, a calculation model, etc. So the AI module can 120 gradually provide a more precise prediction of the coefficient of friction through reinforcing machine learning.

The AI module 120 is also set up, for example, to be trained using training data records from at least a subset of the input variables 400 , 500 , 600 the presence of an intermediate layer on a surface of the traffic area section 200 to determine. For example, weather data 400 used to infer an existing water film from a quantity of rain or an ice formation from a quantity of precipitation in connection with a surface temperature. In addition, due to the vehicle's sensor data 500 conclude that the surface of the road is icy or has reduced friction if there are several different vehicles detect a vehicle dynamics control intervention. The AI module has this for you 120 via one or more models that allow a determination of the intermediate layer. In addition, the AI module 120 set up to determine an intermediate layer property assigned to the intermediate layer, such as, for example, geometric data in the form of a water film thickness, a temperature of the road, a snow layer thickness, an ice layer thickness, a leaf layer thickness, etc. Through the feedback of the reward determination module 130 the determination of the intermediate layer can also be gradually improved.

Using that from the AI module 120 The vehicle can use the friction coefficient information provided 300 , which can also be, for example, an at least partially automated vehicle, plan its route, driving strategy, driving trajectory, etc. or, if the coefficient of friction is unfavorable, the traffic area section 200 avoid or bypass.

QUOTES INCLUDE IN THE DESCRIPTION

This list of documents listed by the applicant has been generated automatically and is only included 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

• WO 2016/120092 A1 [0003]

Claims (10)

Method for providing a coefficient of friction information for a traffic area section (200) to a vehicle (300), comprising the steps: - Determining a coefficient of friction prediction by an AI module (120), to which at least one input variable (400, 500, 600) that can be assigned to the traffic area section (200) is fed, - Determining a weighted deviation between the coefficient of friction prediction and a coefficient of friction measurement (700) for a common point in time or time period, and - Feeding the determined evaluated deviation to the KI module (120) and adapting the determination of the coefficient of friction prediction based on the evaluated deviation. Procedure according to Claim 1 , characterized in that the deviation is rated as positive if it is less than a threshold value, and is rated as negative if it is greater than the threshold value. Procedure according to Claim 1 or 2nd , characterized in that the AI module (120) provides the coefficient of friction prediction for the vehicle (300) before reaching the traffic area section (200). Method according to one of the preceding claims, characterized in that the input variable (400, 500, 600) are supplied to an artificial neural network of the AI module (120). Method according to one of the preceding claims, characterized in that the coefficient of friction prediction comprises a change in the coefficient of friction over time. Method according to one of the preceding claims, characterized in that the input variable (400, 500, 600) comprises on the vehicle side determined coefficient of friction data of the traffic area section (200). Method according to one of the preceding claims, characterized in that the input variable (400, 500, 600) comprises road condition data of a road sensor. Method according to one of the preceding claims, characterized in that the input variable (400, 500, 600) comprises weather-related environmental data of the traffic area section. Procedure according to Claim 8 that on the basis of at least the weather-related environmental data, an intermediate layer of the traffic area section is determined as a further input variable for the AI module. System (100) for providing a coefficient of friction information for a traffic area section to a vehicle (300), with a server device (110) having an artificial intelligence module (120), a Kl module, and a data interface, the server device (110 ) is set up to - to determine a coefficient of friction prediction by the AI module (120), to which at least one input variable (400, 500, 600) that can be assigned to the traffic area section (200) is fed via the data interface, - determine a weighted deviation between the coefficient of friction prediction and a coefficient of friction measurement for a common point in time or time period and - adjust the determination of the coefficient of friction based on the assessed deviation.

Images