A method for predicting a driving behavior of a driver of a motor vehicle by a support system, a computer program product

Abstract

A method and system for predicting a driving behavior 14 of a driver of a motor vehicle 10 by a support system 12. Comprising capturing a current driving behavior of the driver along a route 16 as fir

Description

[0001] The present invention relates to the field of automobiles. More specifically, the present invention relates to a method for predicting a driving behavior of a driver of a motor vehicle by a support system according to the pending claim 1. Furthermore, the present invention relates to a corresponding computer program product, a corresponding non-transitory computer-readable storage medium, as well as to a corresponding support system. BACKGROUND INFORMATION

[0002] From the state of the art it is known that in order to provide, for example, an information about an energy consumption of a motor vehicle a driver profile may be captured, and therefore a more detailed energy consumption may be predicted. In particular the state of the art just estimates a driver behavior in a short term prediction horizon, for example the next 10 seconds, but the state of the art is not capable of predicting driving behavior for a long term, for example 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-transitory computer-readable storage medium as well as a corresponding support system, by which a driving behavior of a driver may be predicted in an improved manner.

[0004] This object is solved by a method, a corresponding computer program product, a corresponding non-transitory computer-readable storage medium, as well as a support system according to the independent claims. Advantageous embodiments of the method are presented in the dependent claims.

[0005] One aspect of the present invention relates to a method for predicting a driving behavior of a driver of a motor vehicle by a support system. A current driving behavior of the driver is captured along a route as a first time series data by a first capturing device of the support system. A current route information of the route as a second time series data is captured by a second capturing device of the support system. A trimming information for trimming the time series data depending on the current route information is determined by an electronic computing device of the support system. The first time series data and the second time series data are trimmed depending on the determined trimming information by the electronic computing device. The trimmed first time series data and the trimmed second time series data are paired by the electronic computing device. The trimmed pairs are categorized into at least two groups depending on at least one characterizing parameter of the group by the electronic computing device. A reinforcement learning for each group is provided by the electronic computing device, wherein a standardized driving behavior profile is adapted depending on the reinforcement learning. The adapted driving behavior profile is provided for predicting the driving behavior of the driver by the electronic computing device.

[0006] With the provided method a more accurate estimated driving behavior for downstream apps is provided, wherein the implementations and constrains / limits are also considered. The provided strategies may compile to the current vehicle’s on-board control frame. The provided strategies may be cost effective regarding hardware upgrades and may utilize cloud services.

[0007] Therefore, a method is provided to predict driving behaviors, such as driving speed profiles, in particular in long prediction horizon with applying reinforcement learning offline training strategies. With the provided method, a driver behavior prediction matrix may be generated by the well-trained agent. Both, the agent and the matrix may be easily implemented to the motor vehicle control frame while considering the current on-board control unit limits without comprising performance. The provided methods and systems may include a driving data and road data sensing and recording system, such as ADAS, a road data categorization method, a reinforcement learning offline training strategy, and a method to implement the feature to the current on-board control software. With the provided methods and systems, the driving behavior in long prediction horizon may be accurately estimated and may improve the performance of the downstream applications, such as vehicle-real-time power consumption prediction.

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

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

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

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

[0012] In another embodiment, grouping of the time series is performed depending on a speed of the motor vehicle and / or a turn of the motor vehicle.

[0013] In particular, the presented present invention is a computer-implemented method. Therefore, another aspect of the invention relates to a computer program product comprising program code means for performing a method according to the preceding aspect.

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

[0015] Furthermore, the present invention relates to a support system for predicting a driving behavior of a driver of a motor vehicle, comprising at least one first capturing device, and one second capturing device, and one electronic computing device, wherein the support system is configured for performing a method according to the preceding aspect. In particular, the method is performed by the support system.

[0016] Advantageous embodiments of the method are to be regarded as advantageous embodiments of the computer program product, the non-transitory computer-readable storage medium, as well as the support system. The support system therefore comprises means for performing a method according to the preceding aspect.

[0017] A computing unit / electronic computing device may in particular be understood as a data processing device, which comprises processing circuitry. The computing unit can therefore in particular process data to perform computing operations. This may also include operations to perform indexed accesses to a data structure, for example a look-up table, LUT.

[0018] In particular, the computing unit may include one or more computers, one or more microcontrollers, and / or one or more integrated circuits, for example, one or more application-specific integrated circuits, ASIC, one or more field-programmable gate arrays, FPGA, and / or one or more systems on a chip, SoC. The computing unit may also include one or more processors, for example one or more microprocessors, one or more central processing units, CPU, one or more graphics processing units, GPU, and / or one or more signal processors, in particular one or more digital signal processors, DSP. The computing unit may also include a physical or a virtual cluster of computers or other of said units.

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

[0020] A memory unit may be implemented as a volatile data memory, for example a dynamic random access memory, DRAM, or a static random access memory, SRAM, or as a non-volatile data memory, for example a read-only memory, ROM, a 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 derive from the following description of preferred embodiments as well as from the drawings. The features and feature combinations previously mentioned in the description as well as the features and feature combinations mentioned in the following description of the figures and / or shown in the figures alone can be employed not only in the respectively indicated combination but also in any other combination or taken alone without leaving the scope of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The novel features and characteristic of the present disclosure are set forth in the appended claims. The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and together with the description, serve to explain the disclosed principles. The same numbers are used throughout the figures to reference like features and components. Some embodiments of system and / or methods in accordance with embodiments of the present subject matter are now described below, by way of example only, and with reference to the accompanying figures.

[0023] The drawings show in:

[0024] Fig. 1 a schematic flow chart according to an embodiment of the method;

[0025] Fig. 2 another schematic flow chart according to an embodiment of the method;

[0026] Fig. 3 another schematic flow chart according to an embodiment of a method;

[0027] Fig. 4 another schematic flow chart according to an embodiment of a method;

[0028] Fig. 5 a schematic top view of a road situation for performing a method according to an embodiment of the invention;

[0029] Fig. 6 another schematic flow chart according to an embodiment of the method;

[0030] Fig. 7 a schematic diagram for implementing the method according to an embodiment of the invention;

[0031] Fig. 8 another schematic flow chart according to an embodiment of the method;

[0032] Fig. 9 another schematic flow chart according to an embodiment of the method;

[0033] Fig. 10 another schematic diagram for implementing the method according to an embodiment of the invention; and

[0034] Fig. 11 another schematic flow chart according to an embodiment of the method.

[0035] In the figures the same elements or elements having the same function are indicated by the same reference signs. DETAILED DESCRIPTION

[0036] In the present document, the word "exemplary" is used herein to mean "serving as an example, instance, or illustration". Any embodiment or implementation of the present subject matter described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

[0037] While the present disclosure is susceptible to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the drawing and will be described in detail below. It should be understood, however, that it is not intended to limit the disclosure to the particular forms disclosed, but on the contrary, the disclosure is to cover all modifications, equivalents, and alternatives falling within the scope of the disclosure.

[0038] The terms “comprises”, “comprising”, or any other variations thereof, are intended to cover a non-exclusive inclusion so that a setup, device or method that comprises a list of components or steps does not include only those components or steps but may include other components or steps not expressly listed or inherent to such setup or device or method. In other words, one or more elements in a system or apparatus preceded by “comprises” or “comprise” does not or do not, without more constraints, preclude the existence of other elements or additional elements in the system or method.

[0039] In the following detailed description of the embodiment of the present disclosure, reference is made to the accompanying drawing that forms part hereof, and in which is shown by way of illustration a specific embodiment in which the disclosure may be practiced. This embodiment is described in sufficient detail to enable those skilled in the art to practice the disclosure, and it is to be understood that other embodiments may be utilized and that changes may be made without departing from the scope of the present disclosure. The following description is, therefore, not to be taken in a limiting sense.

[0040] Fig. 1 shows a schematic flow chart according to an embodiment of a method. In particular, Fig. 1 shows a flow chart for a method for predicting a driving behavior of a driver of a motor vehicle 10 (Fig. 5) by a support system 12 (Fig. 5). A current driving behavior 14 (Fig. 5) of the driver of the motor vehicle 10 along a route 16 (Fig. 5) as first time series data is captured by a first capturing device of the support system 12. A current route information 18 (Fig. 5) of the route 16 as second time series data is captured by a second capturing device of the support system 12. A trimming information for trimming the time series data depending on the current route information 18 is determined by an electronic computing device of the support system 12. The first time series data and the second time series data are trimmed depending on the determined trimming information by the electronic computing device. The trimmed first time series data and the trimmed second time series data are paired by the electronic computing device. The trimmed pairs are categorized into at least two groups depending on at least one characterizing parameter of the group by the electronic computing device. A reinforcement learning for each group is provided by the electronic computing device, wherein a standardized driving behavior profile is adapted depending on the reinforcement learning. The adapted driving behavior profile is provided for predicting the driving behavior 14 of the driver by the electronic computing device.

[0041] In particular, Fig. 1 shows that in a step S1.1 driving behavior data time series and road data time series are collected or determined. In step S1.2, the time series data are trimmed, behavior-road time series data pairs are generated according to the time alignment, and the pairs are categorized into multiple groups. In a step 1.3 reinforcement learning offline training for each group is performed. In a step S1.4 the well-trained agent / behavior prediction is implemented as, for example, a matrix of each group. In a step S1.5, the prediction matrix is provided for downstream applications.

[0042] In particular, Fig. 1 shows that the system may include five steps as shown in Fig. 1, which are driving behavior data time series and road data time series collection or determination, trim the time series data, generate behavior-road time series data pairs according to the time alignment, categorize the pairs into multiple groups, reinforcement learning offline training for each group, and implement the well-trained agent / behavior prediction matrix for each group. The first step S1.1 in Fig. 1 requires sufficient data collection or determination from one specific driver / driving strategy and one specific motor vehicle 10. More detailed description of S1.1 may be described in Fig. 2. The design of the step S1.2 may be modified, such as the length of the time series pairs at the defined group numbers. Longer time series and more data pairs and groups may lead to better performance of the well-trained agent, however the calculation cost may be increased significantly. More details about the step S1.2 is shown in Fig. 3. The number of, for example, neural network layers of the reinforcement learning strategy in the step S1.3 may vary based on the complexity of the training. The offline training may rely on sufficient data collection or determination from the S1.1 and a solid data categorization method in the S1.2. The expected convergence of the road function may not be achieved due to lack of data sets or designing the data categorization unproperly. More detailed description of the S1.3 is shown in the Fig. 4. Implementing the well-trained agent in the motor vehicle 10 on-board control software may depend on the development tool chain and the onboard control unit limits. If the well-trained agent is not available and is not unrealistic for hardware upgrading, the well-trained agent may be fully substituted by the behavior prediction matrix, which is generated by the agent with the same effects but in a look-up table form. More details of these two methods are included in the present disclosure. While the driving behavior prediction system / support system 12 is implemented, the accurate estimated driving behavior 14 on the long-prediction horizon may be available for the downstream applications.

[0043] Fig. 2 shows a flow chart according to an embodiment of the method, in particular, of the S1.1. In particular, Fig. 2 shows a step S2.1, wherein the method starts. In a step S2.2, data time series from the vehicle sensors or from specific driving strategy determination, for example from ADAS or defined optimal driving profiles are provided. In a step S2.3, the driving behavior data time series are collected or determined. In a step S2.4, road data time series are collected or determined. In step S2.5, time series pairs, in particular, for example, a behavior data and road data pair is generated according to the time alignment. In a step S2.6, the generated time series pairs are saved, and in a step S2.7 the step ends.

[0044] In particular, Fig. 2 shows the detailed procedure of that how the driving behavior data and the corresponding road data that may be obtained from the vehicle on-board sensing system or specific driving strategy determination, such as ADAS or defined optimal driving profiles. The data collection or determination state, which is to obtain sufficient data for the following offline training, the data time series may be recorded while one specific driver is driving one specific car or be determined by one specific driving strategy with one specific car. Various driving characteristics may be defined as the demanded driving behavior data, such as speed profile, steering wheel movement, gear shift behavior, or other data. Defining the demanded driving behavior 14 may depend on the purpose of the training. Similarly, various road conditions may be defined as the demanded road data, such as speed limit, up / downhill, left / right turn, traffic light numbers, or furthermore. Defining the road conditions depends on the training purpose as well. The time series data of the driving behavior and road data may be linked and paired to each other according to the time alignment for the following data trimming purposes, then all the pairs of data can be saved for future use.

[0045] This speed limit can be the actual real speed limit sign on the road, it also can be some preset values such as the estimated optimal speed limit or preferred speed limit corresponding to different downstream app requirements.

[0046] Fig. 3 shows another schematic flow chart according to an embodiment of the method. In particular, a detailed flow chart for the step S1.2 is shown.

[0047] In a step S3.1, the method starts. In particular, in a step S3.2, a behavior data, and road data, in particular as a pair with time alignment, are provided. In a step S3.3, a road data categorization rule is provided. Fig. 3 further shows a step S3.4, wherein data set pairs with time alignment may be grouped to different group based on the marks / labels. In particular, four different groups are shown. After the step S3.4, in particular for each group, a step S3.5 is performed, wherein trimmed pairs with distance breakpoints are generated. A step S3.6 is performed for each group, wherein these new pairs are saved to, for example, group 1. In a step S3.7 the step ends.

[0048] Therefore, Fig. 3 shows the frame of that how the data pairs with time alignment from the data collection or determination state from Fig. 2 are trimmed to smaller data sets and categorized into several groups. For each data pair from the data collection or determination state, it may be trimmed and separated into multiple shorter pieces of time series data pairs following the road data categorization rule. A more detailed road data categorization rule is shown in Fig. 4. After going through the road data categorization rule, multiple data set pairs may be generated by dividing the data into shorter time frames. The road data categorization rule may mark / label each of the data set pairs to match with the group categories to ensure that each pair can be grouped into the correct group. Then, for each group a certain number of distance breakpoints (example shown in Fig. 7) may be defined in the data set for the future breakpoint-to-breakpoint decision making and training. Then, the new data set pair may be saved in the same group / category with other similar pairs for the future training purposes. The group / category number is not fixed and how to define the groups depends on the training purposes, an example is provided in Fig. 4.

[0049] Therefore, Fig. 4 shows another flow chart according to an embodiment of the method. In a step S4.1 the method starts. In step S4.2, one pair of the long time frame behavior data and the road data is provided. In a step S4.3, the long time frame data is divided into short time frame pieces according to the road conditions or driving behaviors, e.g. driving on a straight road piece, driving on a right turn piece, etc. In a step S4.4, each short time frame piece is provided. After that, a step S4.5 is provided for categorization and marking / labeling. If the speed is categorized, then the data may be marked / labeled in a sub-step of S4.5, and if the speed is not categorized, there is an error and the method ends. After the step S4.5 a step S4.6 for marking / labeling the vehicle movement direction is provided. If the movement is categorized, then the data may be marked / labeled in a sub-step of S4.6, and if the movement is not categorized, there is an error and the method ends. After the step S4.6, a step S4.7 is provided, wherein one set pair with marks / labels is generated. In a step S4.8 the method ends.

[0050] In particular, Fig. 4 shows the detailed road data categorization method with an example. In the example, the training purpose may be to predict, for example, the driver’s speed profile next to 200 meter driving while only considering the route speed limit (Fig. 5) with high, medium, low levels and route with left, right turns and straight in S4.5 and S4.6. In this scenario, the speed limit lower than 30 miles per hours are defined as low speed limits, which is shown with an L in Fig. 4. The speed limits between 30 miles and 60 miles are defined as medium level speed, M in Fig. 4. The speed limits higher than 60 miles per hour are defined as high level speeds, H in Fig. 4. Furthermore, in this scenario, it is assumed that only three steering wheel movements happened which are the vehicle left 90 degrees turn (+90 deg), the vehicle right 90 degrees turn (-90 deg), and the vehicle keeps moving straight as shown with S4.6.

[0051] Fig. 5 shows a corresponding example. In Fig. 5, the schematic diagram shows an example of dividing one pair, in particular long time frame behavior data and road data, into multiple shorter time frames pieces. Along the driving route 16, the road data has been separated into seven groups with respect to the demanded road information data, for example speed limit 20 and turns and distance 200 meter. So, the shorter time frame, in particular behavior data and road data with time alinement, pairs can be generated. Each of these short time frame groups may be processed using the disclose method steps in Fig. 4, and the data may be marked / labeled based upon the road information. Such, as the categorization logic L to L as shown in the Fig. 4 means the speed limit of the beginning part of one group is in low level and the speed limit of the end of the same group is also in low level. If the group can fulfil the requirements, it may be marked / labeled with L to L and then go the next categorization logic, which may mark / label the data based upon the vehicle movement. Such as, if the vehicle is turned 90 degrees right in this group, then the piece will be marked as minus 90 degrees. It is obvious that the actual driving and road conditions could be way more complex than the examples shown in Fig. 4. Defining the categories for the grouped short time frame data and the rules of dividing the long time frame data may be dependent on the training purposes and may have significantly effects on the agent performances.

[0052] Fig. 6 shows another flow chart according to the embodiment of the method. In particular, a detailed description of the step S1.3 is shown. In particular, Fig. 6 shows a step S6.1, wherein the method starts. In a step S6.2 one group of short time frame pairs with breakpoints is provided. In a step S6.3, the training environment based on that one pair is generated. In a step S6.4 it is checked, if this particular pair is the initial pair. If yes, a step S6.5 is performed, wherein initial action decision is made. In a step S6.6 updating the training policies and may take an action from the action space based on the states and rewards from a reward function. Coming from the step S6.4, if this particular pair is not the initial pair, the step S6.6 is performed. A step S6.7 is performed, wherein states and rewards are updated. In a step S6.8 it is checked, if the rewards converge. If not, the step S6.3 is performed again. If the rewards converge, a step S6.9 is performed, wherein the well-trained agent for the group is obtained. In step S6.10 the method ends.

[0053] In particular, Fig. 6 shows the offline training process. According to the Fig. 6, the reinforcement learning offline training is based upon the groups of short time frame data, which is generated in the former step. The well-trained agent is only feasible for a specific group, which means there may be more than one well-trained agents according to the number of the defined groups. Designing the action space and the reward function are important strategy for the offline training, and the optimized offline training including the convergence of the rewards may improve the computational and training costs.

[0054] In the example scenario, in particular shown in Fig. 7, the strategy that how to define the action space in the reward function is provided for implementing the method according to an embodiment of the invention. In particular, on the x-axis, a position P, for example in meter, is provided. On the y-axis, a velocity V, in particular in miles per hour is provided. Furthermore, Fig. 7 shows an actual velocity 22, an actual speed 24, a real breakpoint 24, an estimated breakpoints 26, an estimated velocity 28 and a difference area 30.

[0055] In particular, Fig. 7 shows, that the breakpoint position is fixed. The training purpose is to obtain a well-trained agent for speed profile estimation for the selected group. In this case, the training purpose may be simplified to estimate the vehicle speed for each breakpoint between the initial and end points. Then, the speed curve changing between every two breakpoints may be assumed as increasing / decreasing with constant slope or maintaining the same speed, so the estimated speed profile may be obtained. In this way, if there are three actions on each in-between breakpoints, the agent can choose, which are increasing the on-breakpoint speed by a fixed value, decreasing the speed by a fixed value and maintaining the speed; and if there are three in-between breakpoints in total, the action space in the case may be defined as: ( +F * F W ) J “ “ a, a € 0 „ 0 , 0 |L ( ■ -F -F -F ) where “a” is a one by three action list, which represents the selected actions on all the inbetween breakpoints, so the whole action space is a three-by-three matrix which includes all the possible actions on each in-between breakpoints, and the constant velocity “V” in the action space represents the fixed increment / decrement of the action. In this case, the estimated speed profile may be required to have the smallest differences from the ground truth / pre-determined speed, so the reward function can be defined as: r “ where the Area, isibsiiIif.tthe difference area 30 shown in Fig. 7, and the aim of the reward function is to find the maximum negative area summation value which means the minimum positive area summation value or the smallest difference between the estimated speed profile and the ground truth speed profile. The updating states in the training may be defined as:

[0056] where Vstart and Vend are the ground truth speed of the start, which may be predetermined and end points for one pair which feeds the offline training in the Fig. 7, and Ptum is the relative position of the turn in the same pair to the start point position. In the Table 1 below, the D is the experience buffer for the training where all the training experience epairNum is stored. The training episode number is not fixed since it will depend on when the reward function may converge. The pairNum also depends on the number of pairs in the group which are generated in S1.2. initialization: Initialize replay memory D; Initialize actton-wfoe f u ncti on s; for episode-M do Initialize the state %; for paWum - 1, T do Either select a random action apair^wH e w^h probability e or select greedy action which maximizes the target; Execute 8^«»» observe the reward an^ the new State * Store the experience: h A» Sample a random mintbatch from 0 and train the (Hetwork G, based on gradient descent; Synchronize the target network with the CHetwork; end end

[0057] Fig. 8 shows in a schematic flow chart according to an embodiment of the method. In particular, Fig. 8 shows in a step S8.1 that the method starts. In step S8.2 realtime road environment data in the required horizon (a long time frame route) is provided. In step S8.3 the same road data categorization rule, as in Fig. 4 is applied. In a step S8.4 short time frame data set pairs with group categorization marks / labels are checked. In a step S8.5, one short time frame pair sequentially along the long time frame route 16 is feed. In the step S8.6, the corresponding well-trained agent or behavior prediction matrix is provided which may be represented with a series of steps. In a step S8.7, the corresponding action list is made by the agent or the corresponding prediction action list with the categorization marks / labels is located in the matrix. In a step S8.8, an accurate driving behavior estimation is provided. In a step S8.9 it is checked, if it is the last short time frame pair from the long time frame route. If not, the step S8.4 is provided again. If yes, a step S8.10 is provided, wherein this method step ends.

[0058] Therefore, Fig. 8 shows how to use the well-trained agent, or the whole behavior prediction matrix if the agent is not compatible with the toolchain and cannot be applied, to estimate / regenerate the drivers’ behaviors based on the actual real-time road conditions in the demanded prediction horizon. The same road data categorization rule as in Fig. 4, is applied to divide the long time frame road data into short time frame road data with distance alinement. The short time frame data pieces may be feed to the system sequentially along the route (demanded prediction horizon). The system may identify the short time frame data pieces one by one and select the corresponding well-trained agents if the agents may be implemented, then the agents may generate a corresponding action list one by one for the feed pieces. If the offline trained agents are incompatible with the current toolchain and cannot be implemented on the current software, a substitute plan is provided in Fig. 8 as well. The offline well-trained agents may generate the corresponding action list for each of the groups which are categorized based on the road data categorization rule. The generated action list for one group may include a vector of numbers, in this way, all the well-trained agents may be simplified as a behavior prediction matrix which includes all the vectors of numbers for each single group. How to generate the action list for each group and how to generate the behavior prediction matrix are shown in Fig. 9. Either the well-trained agents or the behavior prediction matrix may achieve the purposes of drivers’ behaviors estimation / regeneration.

[0059] Fig. 9 shows another schematic flow chart according to an embodiment of the method. In a step S9.1 the method starts. In a step S9.2 all groups based upon the road categorization are provided. In a step S9.3, one piece of the data set pairs from one group is identified. In the step S9.4 the breakpoint-to-breakpoint action list is generated based on the correspondingly well-trained agent. In the step S9.5, the action list is saved to the behavior prediction matrix. In a step S9.6, it is checked, if it is the last group. If not, the step S9.3 is performed again. If yes, a step S9.7 is performed, wherein the method step ends.

[0060] Fig. 10 shows another diagram for the method. In particular, Fig. 10 shows ground truth data 32, estimated data 34 and previous estimated data 36. In particular, Fig. 10 shows an example of how the method may estimate the driver’s speed profile in the selected horizon such as the previous horizon 40 or the shaded region for the current horizon 38. While the real-time signals of the motor vehicle 10 is updating, once the route 16 is fixed, the estimated speed profile may keep updating along the selected route 16 according to the route environment data such as route information 18 and the corresponding action list. In this way, the accurate driver’s behavior prediction may be obtained.

[0061] Fig. 11 shows another schematic flow chart according to an embodiment of the method. In particular, Fig. 11 that in a step S11.1 the method starts. In a step S11.2, famous racing driver’s data on tracks are collected to obtain sufficient data base. In a step S11.3, a road categorization rule is provided. In a step S11.4, reinforcement offline training is performed. In a step S11.5, the well-trained agent matrix / behavior prediction matrix is provided. In a step S11.6, this part of the method ends.

[0062] Furthermore, a step S11.7 is shown, wherein road environment data of famous racing tracks all over the world are provided. Furthermore, in a step S11.8 real-time vehicle signals and driver behaviors are provided. In a step S11.9, a road categorization rule is provided. In a step S11.10, a corresponding well-trained agent / prediction action list is provided. In step S11.11, the difference between the current driver behavior and the estimated famous racing driver’s behavior is calculated. In a step S11.12, safety and aggressive level constraints are provided. In a step S11.13, the breakpoint-to-breakpoint guidance on the user interface is provided. After this step, the step S11.6 is provided.

[0063] Therefore, in Fig. 11 another possible racing behavior shadow and racing guidance application is presented. With the presented method in this present invention, famous racing drivers on-track driving behaviors or selected cars can be recorded. When the road categorization rule is well-defined, the defining groups can well represent most road environment data pieces on all the popular racing tracks all over the world. So, it would be possible to estimate the famous racing drivers’ driving behavior on all the tracks that may be well represented instead of asking the driver to physically drive on all tracks.

[0064] As shown in the Fig. 11, the first flow chart on the top with S11.1 -11.6, once a sufficient data base of one famous driver with one car is generated, the method provided in this present invention can be applied to offline training the agent for future driver behavior estimation. When the well-trained agent matrix or behavior prediction matrix are implemented, the estimated driver behaviors of the racing driver may be used as a driving guidance for the customer which are the same kind of car on different tracks as shown in Fig. 11 in the second chart at the bottom with S11.1, S11.6, and S11.7-11.13.

[0065] The different racing track data may go through the same road categorization rule and may be classified by multiple groups of data pairs. The short road data pairs within the required horizon may be provided to the corresponding implemented agents or action lists according to the marks they have as it has been discussed before. Then, the realtime vehicle signals and driver behaviors can be detected, so the difference between the real-time signals and the estimated racing driving behaviors can be calculated, while considering other regulations and constraints, such as safety issues, a driving guidance which may provide an opportunity for the customers to drive their favorite racing drivers' shadows, and also can help them drive more professionally, is available. Of course, a concise and easy to track user interface may be provided, the present disclosure includes the projection of the shadow image on the windshield or another user interface, with clear breakpoint-to-breakpoint guide and / or other data points.

[0066] It is additionally mentioned, that to predict the travel time and to predict the energy consumption there are two possible downstream applications described, but the invention can be applied on more possible applications such as, training an artificial intelligence to learn and shadow famous racing drivers’ driving profiles on tracks to improve / guide the sports car customers on racing tracks, training the artificial intelligence to learn and shadow the optimal driving behaviors to train the professional racers and training the artificial intelligence to learn and shadow the safest driving behaviors to warn the customers to drive more safely, etc. Reference signs 10 12 14 16 18 20 22 24 26 28 30 32 34 36 P V L M H S1.1 to S11.13 motor vehicle support system driving behavior route route information speed limit actual velocity real brake point estimated brake point estimated velocity difference area ground truth data estimated data previous estimated data position velocity low speed medium speed high speed steps of the method

Claims

1. A method for predicting a driving behavior (14) of a driver of a motor vehicle (10) by a support system (12), comprising the steps of:- capturing a current driving behavior (14) of the driver along a route (16) as first time series data by a first capturing device of the support system (12);- capturing a current route information (18) of the route (16) as second time series data by a second capturing device of the support system (12);- determining a trimming 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 the first time series data and the second time series depending on the determined trimming information by the electronic computing device;- pairing the trimmed first time series data and the 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 a 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.

2. The method according to claim 1, characterized in thatthe predicted driving behavior (14) is used for predicting a travel time of the route (16) and / or an energy consumption for the route (16).

3. The method according to claim 1 or 2, characterized in thatas the 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 captured.

4. The method according to claim 3, characterized in thata position of the at least one speed limit (20) and / or one curve is taken into consideration.

5. The method according to any one of claims 1 to 4, characterized in thatthe adapted driving behavior profile is provided as a driver behavior prediction matrix.

6. The method according to any one of claims 1 to 5, characterized in thatgrouping of the time series is performed depending on a speed of the motor vehicle (10) and / or a turn of the motor vehicle (10).

7. A computer program product comprising program code means for performing a method according to any one of claims 1 to 6.

8. A non-transitory computer-readable storage medium comprising at least the computer program product according to claim 7.

9. An support system (12) for predicting a driving behavior (14) of a driver of a motor vehicle (10), comprising at least one first capturing device, one second capturing device, and one electronic computing device, wherein the support system (12) is configured for performing a method according to any one of claims 1 to 6.

Images