METHOD FOR PREDICTING THE DRIVING BEHAVIOR OF A DRIVER OF A MOTOR VEHICLE BY A SUPPORT SYSTEM, A COMPUTER PROGRAM PRODUCT, A NON-VOID COMPUTER-READABLE STORAGE MEDIUM AND A SUPPORT SYSTEM

Abstract

The present invention relates to a method for predicting the driving behavior (14) of a driver of a motor vehicle (10) by means of a support system (12), comprising the steps of: acquiring current driving behavior (14) of the driver along a route (16) as first time series data by a first acquisition device of the support system (12); acquiring current route information (18) of the route (16) as second time series data by a second acquisition device of the support system (12); determining trim information for trimming the time series data depending on the current route information (18) by means of an electronic computing device of the support system (12); trimming the first time series data and the second time series depending on the determined trim information by means of the electronic computing device;The present invention relates to the processing of pairs of trimmed first time series data and trimmed second time series data by the electronic computing device; categorizing the trimmed pairs into at least two groups depending on at least one characterizing parameter of the group by the electronic computing device; providing reinforcement learning for each group by the electronic computing device, wherein a standardized driving behavior profile is adapted depending on the reinforcement learning; and providing the adapted driving behavior profile for predicting the driving behavior (14) of the driver by the electronic computing device. The present invention further relates to a computer program product, a non-volatile computer-readable storage medium, and a support system (12).

Description

AREA OF INVENTION

[0001] The present invention relates to the field of motor vehicles. In particular, the present invention relates to a method for predicting the driving behavior of a driver of a motor vehicle by means of a support system according to the pending claim 1. Furthermore, the present invention relates to a corresponding computer program product, a corresponding non-volatile computer-readable storage medium, and a corresponding support system. BACKGROUND INFORMATION

[0002] It is known from the prior art that, to provide information about, for example, the energy consumption of a motor vehicle, a driver profile can be recorded and thus a more detailed energy consumption prediction can be made. In particular, the prior art only estimates driver behavior in a short-term prediction horizon, e.g., for the next 10 seconds, but is not able to predict driving behavior in the long term, e.g., for the next ten minutes. SUMMARY OF THE INVENTION

[0003] It is an object of the present invention to provide a method, a corresponding computer program product, a corresponding non-volatile computer-readable storage medium and a corresponding support system with which the driving behavior of a driver can be predicted in an improved manner.

[0004] This problem is solved by a method, a corresponding computer program product, a corresponding non-volatile, computer-readable storage medium, and a support system according to the independent claims. Advantageous embodiments of the method are described in the dependent claims.

[0005] One aspect of the present invention relates to a method for predicting the driving behavior of a motor vehicle driver using a driver assistance system. The driver's current driving behavior along a route is recorded as first time series data by a first acquisition unit of the assistance system. Current route information, as second time series data, is recorded by a second acquisition unit of the assistance system. Trimming information for adjusting the time series data based on the current route information is determined by an electronic processing unit of the assistance system. The first and second time series data are trimmed by the electronic processing unit based on the determined trimming information. The trimmed first and second time series data are then combined by the electronic processing unit.The trimmed pairs are divided into at least two groups by the electronic computing unit based on at least one characteristic parameter of the group. Reinforcement learning for each group is provided by the electronic computing unit, whereby a standardized driving behavior profile is adapted based on the reinforcement learning. The adapted driving behavior profile is then provided by the electronic computing unit to predict the driver's driving behavior.

[0006] The provided method delivers more accurate estimated driving behavior for downstream applications, taking into account implementations and limitations. The strategies provided can be compatible with the current vehicle control framework. They can be cost-effective in terms of hardware upgrades and can leverage cloud services.

[0007] Therefore, a method for predicting driving behavior, such as speed profiles, is provided, particularly over long prediction horizons, using offline training strategies with reinforcement learning. The provided method allows a well-trained agent to create a matrix for predicting driving behavior. Both the agent and the matrix can be easily implemented in the vehicle's electronic control unit (ECU), taking into account the current limitations of the ECU without compromising performance. The provided methods and systems can include a system for acquiring and recording driving and road data, such as ADAS (Advanced Driver Assistance Systems), a method for categorizing road data, an offline training strategy with reinforcement learning, and a method for implementing the feature in the current on-board control software.The methods and systems provided enable accurate estimation of driving behavior over a long forecast horizon and improve the performance of downstream applications, such as real-time prediction of vehicle energy consumption.

[0008] According to one embodiment, the predicted driving behavior is used to predict the travel time of the route and / or the energy consumption for the route.

[0009] In another embodiment, the current route information includes at least one speed limit / optimal speed / preferred speed along the route and / or at least one curve along the route.

[0010] In another embodiment, the position of at least one speed limit and / or a curve is taken into account.

[0011] In another embodiment, the adapted driving behavior profile is provided in the form of a driver behavior prediction matrix.

[0012] In another embodiment, the time series are grouped depending on the speed of the motor vehicle and / or a curve of the motor vehicle.

[0013] In particular, the present invention relates to a computer-implemented method. A further aspect of the invention therefore relates to a computer program product comprising program code means for carrying out a method according to the preceding aspect.

[0014] Furthermore, the present invention relates to a non-volatile, computer-readable storage medium that contains at least the computer program product according to the preceding aspect.

[0015] Furthermore, the present invention relates to a support system for predicting the driving behavior of a driver of a motor vehicle, comprising at least a first detection device and a second detection device as well as an electronic computing device, wherein the support system is configured to carry out a method according to the preceding aspect. In particular, the method is carried out by the support system.

[0016] Advantageous embodiments of the method are to be considered as advantageous embodiments of the computer program product, the non-transferable, computer-readable storage medium, and the support system. The support system therefore comprises means for carrying out a method according to the preceding aspect.

[0017] An arithmetic unit / electronic computer system can be understood, in particular, as a data processing system that includes a processing circuit. The arithmetic unit can therefore, in particular, process data in order to perform arithmetic operations. These can also include operations for performing indexed accesses to a data structure, e.g., a lookup table (LUT).

[0018] In particular, the computing unit may comprise one or more computers, one or more microcontrollers, and / or one or more integrated circuits, such as one or more application-specific integrated circuits (ASICs), one or more field-programmable gate arrays (FPGAs), and / or one or more systems-on-a-chip (SoCs). The computing unit may also include one or more processors, such as one or more microprocessors, one or more central processing units (CPUs), one or more graphics processing units (GPUs), and / or one or more signal processors, in particular one or more digital signal processors (DSPs). The computing unit may also comprise a physical or virtual cluster of computers or other units of the aforementioned type.

[0019] In various embodiments, the computing unit comprises one or more hardware and / or software interfaces and / or one or more storage units.

[0020] A storage unit can be a volatile data storage device, e.g., a dynamic random access memory (DRAM) or static random access memory (SRAM), or a non-volatile data storage device, e.g., a read-only memory (ROM), programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory or flash EEPROM, a ferroelectric random access memory (FRAM), a magnetoresistive random access memory (MRAM), or a phase-change random access memory (PCRAM).

[0021] Further advantages, features, and details of the present invention will become apparent from the following description of preferred embodiments and from the drawings. The features and combinations of features mentioned above in the description, as well as those mentioned in the following description of the figures and / or illustrated in the figures alone, can be used not only in the combinations specified, but also in any other combination or on their own, without departing from the scope of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The novel features and properties of the present disclosure are set forth in the accompanying claims. The accompanying drawings, which form part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles. The same numbers are used in the figures to indicate identical features and components. Some embodiments of systems and / or methods according to embodiments of the present subject matter are described below only by way of example and with reference to the accompanying figures.

[0023] The drawings show in: Fig. 1 a schematic flowchart according to one embodiment of the method; Fig. 2. A further schematic flowchart according to an embodiment of the method; Fig. 3. Another schematic flowchart according to an embodiment of a method; Fig. 4: Another schematic flowchart according to one embodiment of a method; Fig. 5 a schematic top view of a street situation for carrying out a method according to an embodiment of the invention; Fig. 6. A further schematic flowchart according to an embodiment of the method; Fig. 7 a schematic diagram for carrying out the method according to an embodiment of the invention; Fig. 8 a further schematic flowchart according to an embodiment of the method; Fig. 9 a further schematic flowchart according to an embodiment of the method; Fig. 10 a further schematic diagram for carrying out the method according to an embodiment of the invention; and Fig. 11. Another schematic flowchart according to one embodiment of the method.

[0024] In the figures, identical elements or elements with the same function are indicated by the same reference symbols. DETAILED DESCRIPTION

[0025] In this document, the word "exemplary" is used to mean "serving as an example, instance, or illustration." Each embodiment or implementation of the subject matter described herein as "exemplary" is not necessarily to be understood as preferred or advantageous over other embodiments.

[0026] While the present disclosure is open to various modifications and alternative forms, specific embodiments are illustrated by way of example in the drawing and are described in detail below. It should be understood, however, that the disclosure is not intended to be limited to the specific forms disclosed, but rather, on the contrary, to cover all modifications, equivalents, and alternatives that fall within the scope of the disclosure.

[0027] The terms “includes,” “contains,” or other variations thereof are intended to cover non-exclusive inclusion, so that a system, device, or process that includes a list of components or steps may contain not only those components or steps but may also contain other components or steps not expressly listed or belonging to such system, device, or process. In other words, one or more elements in a system or device preceded by the expression “includes” or “include” do not, without further limitations, preclude the presence of other elements or additional elements in the system or process.

[0028] The following detailed description of the embodiment of the present disclosure refers to the accompanying drawing, which forms part of this disclosure and illustrates a particular embodiment in which the disclosure can be implemented. This embodiment is described in sufficient detail to enable the person skilled in the art to put the disclosure into practice, and it is understood that other embodiments may be used and that modifications may be made without departing from the scope of the present disclosure. The following description is therefore not to be understood in a limiting sense.

[0029] Fig. Figure 1 shows a schematic flowchart according to one embodiment of a method. In particular, it shows Fig. 1 a flowchart for a procedure for predicting the driving behavior of a driver of a motor vehicle 10 ( Fig. 5) through a support system 12 ( Fig. 5) Current driving behavior 14 ( Fig. 5) of the driver of motor vehicle 10 along a route 16 ( Fig. 5) is recorded as the first time series data by a first acquisition unit of the support system 12. Current route information 18 ( Fig. 5) The second time series data from route 16 is acquired by a second acquisition device of the support system 12. Trimming information for trimming the time series data depending on the current route information 18 is determined by an electronic computing unit of the support system 12. The first and second time series data are trimmed by the electronic computing unit depending on the determined trimming information. The trimmed first and second time series data are paired by the electronic computing unit. The trimmed pairs are divided by the electronic computing unit into at least two groups depending on at least one characteristic parameter of the group.Reinforcement learning for each group is provided by the electronic computing unit, whereby a standardized driving behavior profile is adapted depending on the reinforcement learning. The adapted driving behavior profile is provided by the electronic computing unit to predict the driver's driving behavior 14.

[0030] Fig. Figure 1 shows in particular that in step S1.1, time series of driving behavior data and time series of road data are collected or determined. In step S1.2, the time series data are trimmed, behavior and road data pairs are generated according to their temporal orientation, and the pairs are categorized into several groups. In step 1.3, offline reinforcement learning training is performed for each group. In step S1.4, the well-trained agent / behavior prediction is implemented, for example, as a matrix for each group. In step S1.5, the prediction matrix is ​​made available for downstream applications.

[0031] In particular, it shows Fig. 1. that the system can comprise five steps, as in Fig. The first step, S1.1, is described as follows: the collection or determination of time series of driving behavior data and road data; the trimming of the time series data; the generation of behavior / road time series data pairs according to the temporal orientation; the categorization of the pairs into several groups; offline reinforcement training for each group; and the implementation of the well-trained agent / behavior prediction matrix for each group. Fig. 1 requires sufficient data collection or determination from a specific driver / driving strategy and a specific motor vehicle. 10. A more detailed description of S1.1 is in Fig. 2. The design of step S1.2 can be modified, e.g., the length of the time series pairs for the defined group numbers. Longer time series and more data pairs and groups can lead to better performance of the well-trained agent; however, the computational effort can increase significantly. Further details on step S1.2 are in Fig. Figure 3 illustrates this. The number of layers in the neural network of the reinforcement learning strategy in step S1.3 can vary depending on the complexity of the training. The offline training may rely on sufficient data collection or determination from S1.1 and a robust data categorization procedure in S1.2. The expected convergence of the road function may not be achieved due to a lack of datasets or inadequate data categorization. A more detailed description of S1.3 is provided in [reference missing]. Fig. Figure 4 illustrates this. The implementation of the well-trained agent in the vehicle's onboard control software 10 may depend on the development toolchain and the limitations of the onboard control unit. If the well-trained agent is unavailable and a hardware upgrade is not unrealistic, the well-trained agent can be completely replaced by the behavior prediction matrix generated by the agent with the same effects, but in the form of a lookup table. Further details of these two methods are included in this disclosure. While the driving behavior / assistance system 12 is implemented, the accurately estimated driving behavior 14 can be available for downstream applications over the long prediction horizon.

[0032] Fig. Figure 2 shows a flowchart according to an embodiment of the method, in particular S1.1. In particular, it shows Fig. The process begins with step S2.1. In step S2.2, time series data from vehicle sensors or from a specific driving strategy determination, for example, from ADAS or defined optimal driving profiles, are provided. In step S2.3, the driving behavior data time series are collected or determined. In step S2.4, time series of road data are collected or determined. In step S2.5, time series pairs, in particular, for example, a driving behavior data and road data pair, are generated according to the temporal comparison. In step S2.6, the generated time series pairs are stored, and in step S2.7, the step ends.

[0033] In particular, it shows Fig. 2. The detailed procedure for obtaining driving behavior data and corresponding road data from the vehicle's internal sensor system or a specific driving strategy, such as ADAS or defined optimal driving profiles, is described. During data acquisition, which serves to obtain sufficient data for subsequent offline training, data time series can be recorded while a specific driver is driving a specific vehicle or determined using a specific driving strategy with a specific vehicle. Various driving characteristics can be defined as required driving behavior data, such as speed profile, steering wheel movement, shifting behavior, or other data. The definition of the required driving behavior 14 may depend on the purpose of the training. Similarly, various road conditions can be defined as required road data, e.g.,Speed ​​limits, inclines / declines, left / right turns, traffic light sequences, or other factors. The definition of road conditions also depends on the training objective. The time-series data of driving behavior and the road data can be linked and paired according to the temporal orientation for subsequent data comparison; all data pairs can then be saved for future use.

[0034] This speed limit can be the actual speed limit sign on the road, but also some preset values ​​such as the estimated optimal maximum speed or the preferred maximum speed, which correspond to the various requirements of the downstream app.

[0035] Fig. Figure 3 shows another schematic flowchart according to an embodiment of the method. In particular, a detailed flowchart for step S1.2 is shown.

[0036] In step S3.1, the process is initiated. Specifically, in step S3.2, behavioral data and road data, particularly as a time-matched pair, are provided. In step S3.3, a rule for categorizing road data is provided. Fig. Figure 3 further shows a step S3.4 in which data record pairs with temporal alignment can be divided into different groups based on the markers / labels. Four different groups are shown in particular. After step S3.4, a step S3.5 is performed for each group, in which trimmed pairs with distance stop points are generated. A step S3.6 is performed for each group, in which these new pairs are stored, for example, in group 1. The step ends in a step S3.7.

[0037] Therefore, it shows Fig. 3 the framework for how the temporally oriented data pairs are extracted from the data acquisition or determination state Fig. 2. The data can be trimmed to smaller datasets and categorized into several groups. Each data pair from the state of data acquisition or determination can be decomposed and separated into several shorter time-series data pairs according to the rule for categorizing road data. A more detailed rule for categorizing road data is provided in Fig. Figure 4 illustrates this. After running the road data categorization rule, multiple dataset pairs can be generated by dividing the data into shorter timeframes. The road data categorization rule can tag / label each dataset pair to match the group categories, ensuring each pair is correctly assigned to the appropriate group. Then, a specific number of distance checkpoints can be assigned to each group (example in Figure 4). Fig. 7) in the dataset for future decision-making and breakpoint-by-breakpoint training. The new dataset pair can then be stored in the same group / category with other similar pairs for future training purposes. The number of groups / categories is not fixed, and the way the groups are defined depends on the training purposes (see Fig. 4).

[0038] Therefore, it shows Fig. 4. Another flowchart according to an embodiment of the method. In step S4.1, the method begins. In step S4.2, a pair of long-term behavioral data and road data is provided. In step S4.3, the long-term data is divided into short-term timeframes according to road conditions or driving behavior, e.g., driving on a straight stretch, driving on a right-hand bend, etc. In step S4.4, each short-term timeframe is provided. A step S4.5 is then provided for categorization and marking / labeling. If the speed is categorized, the data can be marked / labeled in a sub-step of S4.5; if the speed is not categorized, an error occurs, and the method terminates. After step S4.5, a step S4.6 is provided for marking / labeling the vehicle's direction of travel.If the movement is categorized, the data can be marked / labeled in a substep of S4.6. If the movement is not categorized, an error occurs and the procedure terminates. Step S4.7 follows S4.6, generating a record pair with markings / labels. The procedure concludes in step S4.8.

[0039] Fig. Figure 4 shows in particular the detailed procedure for categorizing road data using an example. In the example, the training purpose could be, for instance, to predict the driver's speed profile over a distance of 200 meters, taking into account only the speed limit on the stretch ( Fig. 5) with the levels high, medium and low, as well as the route with left and right turns and straights in S4.5 and S4.6. In this scenario, speed limits below 30 mph are defined as low speed limits, which is in Fig. 4 is represented by an L. The speed limits between 30 mph and 60 mph are defined as average speed limits, M in Fig. 4. Speed ​​limits above 60 mph are defined as high speeds, H in Fig. 4. Furthermore, this scenario assumes that only three steering wheel movements take place, namely a 90-degree left turn (+90 degrees), a 90-degree right turn (-90 degrees) and driving straight ahead, as shown in S4.6.

[0040] Fig. Figure 5 shows a corresponding example. Fig. Figure 5 shows a schematic diagram illustrating an example of splitting a pair, particularly long timeframes of behavioral data and road data, into several shorter timeframes. Along route 16, the road data was divided into seven groups related to the required road information data, e.g., speed limit 20 and curves and distance 200 meters. In this way, pairs of shorter timeframes, especially behavioral data and time-aligned road data, can be generated. Each of these short timeframe groups can be further subdivided using the methods described in Figure 5. Fig. The data is processed in the four described procedure steps, and can be marked / labeled based on the street information. Thus, the categorization logic L to L means, as shown in Fig. Figure 4 shows that the speed limit of the initial part of a group is low, and the speed limit of the final part of the same group is also low. If the group meets the requirements, it can be marked / labeled L to L and then go through the next categorization logic, which can mark / label the data based on vehicle movement. For example, if the vehicle in this group turns 90 degrees to the right, then the part will be marked as minus 90 degrees. It is obvious that actual driving and road conditions can be far more complex than those shown in Figure 4. Fig. The four examples shown illustrate this. The definition of the categories for the grouped short-term data and the rules for splitting the long-term data can depend on the training purposes and have a significant impact on agent performance.

[0041] Fig. Figure 6 shows another flowchart according to the embodiment of the method. In particular, a detailed description of step S1.3 is shown. Fig. Figure 6 shows, in particular, a step S6.1, in which the procedure begins. In a step S6.2, a set of short timeframe pairs with breakpoints is provided. In a step S6.3, the training environment is generated based on this single pair. In a step S6.4, it is checked whether this single pair is the starting pair. If so, a step S6.5 is performed, in which a decision is made regarding the first action. In a step S6.6, the training policies are updated, and an action can be chosen from the action space based on the states and the rewards from a reward function. Starting from step S6.4, step S6.6 is performed if this particular pair is not the starting pair. A step S6.7 is executed, in which states and rewards are updated. In a step S6.8, it is checked whether the rewards converge. If not, step S6.3 is executed again.When the rewards converge, step S6.9 is performed, in which the well-trained agent is obtained for the group. The process is completed in step S6.10.

[0042] Fig. Figure 6 shows, in particular, the offline training process. According to Fig. 6. Offline training using reinforcement learning is based on the groups of short-term data generated in the previous step. A well-trained agent is only possible for a specific group, meaning that depending on the number of defined groups, there can be more than one well-trained agent. Action space design and the reward function are important strategies for offline training, and optimized offline training, including reward convergence, can improve computational and training costs.

[0043] In the example scenario, which is particularly relevant in Fig. Figure 7 shows the strategy for defining the action space in the reward function for implementing the method according to one embodiment of the invention. In particular, a position P, e.g., in meters, is indicated on the x-axis. A speed V, particularly in miles per hour, is indicated on the y-axis. Furthermore, in Fig. Figure 7 shows an actual speed 22, an actual speed 24, a real braking point 24, an estimated stopping point 26, an estimated speed 28 and a difference range 30.

[0044] Fig. Figure 7 shows, in particular, that the position of the stopping point is fixed. The goal of the training is to obtain a well-trained agent for estimating the speed profile for the selected group. In this case, the training objective can be simplified by estimating the vehicle speed for each stopping point between the start and end points. Then, it can be assumed that the speed curve between each pair of stopping points rises / falls with a constant slope or maintains the same speed, allowing the estimated speed profile to be determined.In this way, if there are three actions at each intermediate stop, the agent can choose to increase the speed at the stop by a fixed value, decrease the speed by a fixed value, and maintain the speed; and if there are a total of three intermediate stops, the action space in this case can be defined as follows: A={ΔSpeed=a,a∈R1×3|(+V+V+V0,0,0−V−V−V)}, where "a" is a 1x3 action list representing the selected actions at all intermediate breakpoints, such that the entire action space is a 3x3 matrix containing all possible actions at each intermediate breakpoint, and the constant velocity "V" in the action space represents the fixed increase / decrease of the action. In this case, it may be necessary for the estimated velocity profile to have the smallest deviations from the ground truth / predetermined velocity, so that the reward function can be defined as follows: r=−∑Areai, where the area i the in Fig. The difference range shown in Figure 7 is 30, and the goal of the reward function is to find the maximum negative area sum value, i.e., the minimum positive area sum value or the smallest difference between the estimated velocity profile and the ground-truth velocity profile. The update states in training can be defined as follows: s=(Vstart,Vend,Pturn),

[0045] where V start and V end The ground truth speed of the start and end points for a pair is used in offline training in Fig. 7 flows in, and P turnThe relative position of the curve in the same pair to the starting point position is given. In Table 1 below, D is the experience buffer for training, in which all training experiences epairNum are stored. The number of training episodes is not fixed, as it depends on when the reward function can converge. The number of pairs also depends on the number of pairs in the group generated in S1.2.

[0046] Fig. Figure 8 shows a schematic flowchart according to one embodiment of the method. In particular, it shows Fig. 8 in step S8.1 that the procedure begins. In step S8.2, real-time road environment data for the required horizon (a route with a long timeframe) is provided. In step S8.3, the same road data categorization rule as in Fig. Step 4 is applied. In step S8.4, short-term data pairs with group categorization markers / labels are checked. In step S8.5, a short-term data pair is fed sequentially along the long-term data route 16. In step S8.6, the corresponding well-trained agent or behavior prediction matrix is ​​provided, which can be represented by a series of steps. In step S8.7, the corresponding action list is created by the agent, or the corresponding prediction action list with the categorization markers / labels is placed in the matrix. In step S8.8, an accurate driving behavior assessment is created. In step S8.9, it is checked whether it is the last short-term data pair from the long-term route. If not, step S8.4 is repeated. If so, step S8.10 is provided, with which this process step ends.

[0047] Therefore, it shows Fig. 8. How the well-trained agent, or the entire behavior prediction matrix if the agent is not compatible with the toolchain and cannot be applied, is used to estimate / generate driver behavior based on actual real-time road conditions within the required prediction horizon. The same categorization rule for road data as in Fig. Step 4 is used to split long-term road data into short-term road data and distance information. The short-term data can be fed into the system sequentially along the route (required forecast horizon). The system can individually identify the short-term data segments and select the appropriate, well-trained agents. If the agents can be implemented, they can then generate a corresponding action list for the fed-in segments. If the offline-trained agents are not compatible with the current toolchain and cannot be implemented in the current software, a different approach is used. Fig. Section 8 also provides a backup plan. The well-trained offline agents can generate the appropriate action list for each of the groups categorized based on the street data categorization rule. The generated action list for a group can contain a numerical vector. In this way, all well-trained agents can be simplified to a behavior prediction matrix containing all the numerical vectors for each individual group. How the action list for each group and the behavior prediction matrix are generated is described in Fig. Figure 9 illustrates this. Either the well-trained agents or the behavior prediction matrix can fulfill the purpose of estimating / regenerating driver behavior.

[0048] Fig. Figure 9 shows another schematic flowchart according to one embodiment of the procedure. In step S9.1, the procedure begins. In step S9.2, all groups based on the street categorization are provided. In step S9.3, a subset of the data pairs from a group is identified. In step S9.4, the stop-to-stop action list is created based on the appropriately well-trained agent. In step S9.5, the action list is stored in the behavior prediction matrix. In step S9.6, it is checked whether this is the last group. If not, step S9.3 is repeated. If so, step S9.7 is performed, in which the procedure ends.

[0049] Fig. Figure 10 shows another diagram for the procedure. Fig. Figure 10 specifically presents the ground truth data 32, the estimated data 34, and the previous estimated data 36. In particular, it shows Fig. Figure 10 is an example of how the procedure can estimate the driver's speed profile within the selected horizon, such as the previous horizon 40 or the shaded area for the current horizon 38. While the real-time signals from the vehicle 10 are updated, the estimated speed profile, once the route 16 is established, can be continuously updated along the selected route 16 according to the route environment data, such as the route information 18 and the corresponding action list. In this way, an accurate prediction of driver behavior can be achieved.

[0050] Fig. Figure 11 shows another schematic flowchart according to an embodiment of the method. In particular, it shows Fig. 11, that in step S11.1 the procedure begins. In step S11.2, data from known racing drivers on tracks are collected to obtain a sufficient data basis. In step S11.3, a rule for road categorization is created. In step S11.4, reinforcing offline training is carried out. In step S11.5, the well-trained agent matrix / behavior prediction matrix is ​​provided. In step S11.6, this part of the procedure is terminated.

[0051] Furthermore, step S11.7 is shown, in which road environment data from famous racetracks around the world is provided. Additionally, step S11.8 provides real-time vehicle signals and driver behavior data. Step S11.9 provides a rule for road categorization. Step S11.10 provides a corresponding, well-trained agent / prediction action list. Step S11.11 calculates the difference between the current driver behavior and the predicted behavior of the famous race car driver. Step S11.12 sets safety and aggressiveness limits. Step S11.13 provides breakpoint-to-breakpoint guidance on the user interface. After this step, step S11.6 is executed.

[0052] Therefore, in Fig. 11. Another possible application for monitoring race behavior and race control is presented. The method presented in the present invention allows the driving behavior of famous racing drivers on the racetrack or of selected cars to be recorded. If the road categorization rule is well defined, the defined groups can accurately represent most road environment data on all popular racetracks worldwide. Thus, it would be possible to estimate the driving behavior of famous racing drivers on all tracks that can be accurately represented, instead of requiring the driver to physically drive on all tracks.

[0053] As in the Fig. 11, the first flowchart on the top with S11.1-11.6, once a sufficient data set is generated from a famous driver with a car, the method in this present invention can be applied to offline training of the agent for estimating future driver behavior. When the well-trained agent matrix or the behavior prediction matrix is ​​implemented, the estimated driver behavior of the race car driver can be used as a driving guide for the customer driving the same type of car on different tracks, as in Fig. 11 shown in the second diagram below with S11.1, S11.6 and S11.7-11.13.

[0054] The various racetrack data can be processed using the same road categorization rule and classified by multiple sets of data pairs. The short road data pairs within the required timeframe can be provided to the corresponding implemented agents or action lists according to their markers, as previously discussed. Real-time vehicle signals and driver behavior can then be detected, allowing the difference between the real-time signals and the estimated driver behavior to be calculated. Taking into account other regulations and limitations, such as safety considerations, a driving guide is available, enabling customers to emulate their favorite drivers and potentially helping them to drive more professionally.Of course, a clear and easy-to-follow user interface can be provided; the present disclosure includes the projection of the shadow image onto the windshield or other user interface, with clear breakpoint-to-breakpoint instructions and / or other data points.

[0055] It is further mentioned that predicting driving time and predicting energy consumption are two possible downstream applications, but the invention can be applied to more possible applications, such as training an artificial intelligence to learn and shadow driving profiles of famous racing drivers on racetracks in order to improve / guide sports car customers on racetracks, training the artificial intelligence to learn and shadow optimal driving behavior in order to train professional racing drivers, and training the artificial intelligence to learn and shadow the safest driving behavior in order to warn customers to drive more safely, etc. Reference sign 10 motor vehicle 12 Support system 14 Driving behavior Route 16 18 Route information 20 speed limit 22 Actual speed 24 real braking point 26 estimated braking point 28 estimated speed 30 Difference range 32 Ground Truth data 34 estimated data 36 previous estimated data P Position V speed L low speed M average speed H high speed S1.1 to S11.13 Steps of the procedure

Claims

[1] Method for predicting the driving behavior (14) of a driver of a motor vehicle (10) by means of a support system (12) comprising the following steps: - Recording the driver's current driving behavior (14) along a route (16) as initial time series data by a first recording device of the support system (12); - Acquisition of current route information (18) of the route (16) as second time series data by a second acquisition device of the support system (12); - Determining trim information for trimming the time series data depending on the current route information (18) by an electronic computing device of the support system (12); - Trimming of the first time series data and the second time series depending on the determined trimming information by the electronic computing device; - Pairing of the trimmed first time series data and the trimmed second time series data by the electronic computing device; - Classification of the trimmed pairs into at least two groups depending on at least one characteristic parameter of the group by the electronic computing device; - Providing reinforcement learning for each group through the electronic computing device, whereby a standardized driving behavior profile is adapted depending on the reinforcement learning; and - Provision of the adapted driving behavior profile for predicting the driving behavior (14) of the driver by the electronic computing device. [2] Method according to claim 1, characterized by , that the predicted driving behavior (14) is used to predict a travel time of the route (16) and / or an energy consumption for the route (16). [3] Method according to claim 1 or 2, characterized by, that as current route information (18) at least one speed limit (20) along the route (16) and / or at least one curve along the route (16) is recorded. [4] Method according to claim 3, characterized by , that the location of at least one speed limit (20) and / or a curve is taken into account. [5] Method according to any one of claims 1 to 4, characterized by , that the adapted driving behavior profile is provided in the form of a driver behavior prediction matrix. [6] Method according to any one of claims 1 to 5, characterized by , that the grouping of the time series is based on the speed of the motor vehicle (10) and / or the cornering of the motor vehicle (10). [7] Computer program product comprising program code means for carrying out a method according to any one of claims 1 to 6. [8] Non-volatile computer-readable storage medium containing at least the computer program product according to claim 7. [9] Support system (12) for predicting the driving behavior (14) of a driver of a motor vehicle (10), comprising at least a first detection device, a second detection device and an electronic computing device, wherein the support system (12) is configured to carry out a method according to any one of claims 1 to 6.

Images