Method for simulating the riding performance of a motorcycle on a curve, method for determining a motorcycle target speed for a curve and controller

Abstract

The invention relates to a method for simulating a riding behavior (100) of a motorcycle in a bend (102), wherein motorcycle data (110) and bend data (108) are provided to a trainable algorithm (104), wherein the motorcycle data (110) represents a motorcycle trajectory actually detected when the motorcycle drives through the bend (102), and the bend data (108) represents a geometry and constraints of the bend (102), in order to train the algorithm (104) based on a relationship between the geometry on the bend (102) and the motorcycle trajectory, wherein the motorcycle trajectory for each bend (102) contains at least one motorcycle target speed (106) for a target point (112) of the bend (102).

Description

Technical Field

[0001] The present invention relates to a method for simulating the driving performance of a motorcycle on a curve, a method for determining a target speed of a motorcycle on a curve, a corresponding controller, and a corresponding computer program product. Background Technology

[0002] Motorcycles can be equipped with assistive functions to determine a target speed for a target point ahead of a curve. For example, a motorcyclist might be warned if they are traveling too fast before a curve. Similarly, other motorcycle assistive functions can use this target speed as a target variable and automatically adjust the speed before the curve to ensure the motorcycle travels at the target speed at the target point.

[0003] The target speed can be read from a target speed database. This database can be based on the motorcycle's previous travel patterns through the curve. During these travels, at least one target speed can be recorded for at least one target point at at least one curve.

[0004] For example, DE10258167 B4 describes a method for automobiles. Summary of the Invention

[0005] Against this backdrop, based on the solution presented herein, a method for simulating the performance of a motorcycle on a curve, a method for determining a target speed for a motorcycle on a curve, a corresponding controller, and a corresponding computer program product are proposed according to the independent claim. Advantageous further designs and modifications of the solution presented herein are obtained from the specification and are described in the dependent claims.

[0006] In the proposed scheme, based on the curves and map information of these curves, the relationship between map information and target speed is learned, and for the curves, there is at least one target point where the target speed of the motorcycle actually exists.

[0007] Once this relationship is learned, it can be used to determine the target speed for curves where only map information is available.

[0008] Using the approach proposed here, warnings or appropriate information can be provided to motorcyclists ahead of curves, even without a pre-measured target speed. Alternatively or additionally, the motorcycle assistance system can adjust the motorcycle's speed based on the curve ahead, even without an actual measured target speed.

[0009] According to the first aspect, a method for simulating the driving behavior of a motorcycle on a curve is proposed, wherein motorcycle data and curve data are provided to a trainable algorithm, wherein the motorcycle data represents the actual detected motorcycle trajectory when the motorcycle passes through the curve, and the curve data represents the geometry and constraints of the curve, so as to train the algorithm based on the relationship between the curve geometry and the motorcycle trajectory, wherein the motorcycle trajectory of each curve contains at least one target motorcycle speed for a target point of the curve.

[0010] According to the second aspect, a method for determining a target speed of a motorcycle on a curve is proposed, wherein curve data is provided to an algorithm trained using the method according to the first aspect, wherein the curve data represents the geometry and constraints of the curve, and wherein the algorithm outputs at least one target speed of the motorcycle for a target point on the curve.

[0011] The ideas behind the embodiments of the present invention can be particularly considered to be based on the concepts and discoveries described below.

[0012] Motorcycles (or rather, motorcyclists) behave uniquely when cornering. Due to their narrower frame, motorcycles utilize available lane width more effectively than multi-rail vehicles such as cars. This allows them to take corners with a turning radius greater than the average corner radius of the lane. On the other hand, motorcycles have only two points of contact with the lane, resulting in potentially lower grip.

[0013] Motorcycle data can represent the distance traveled through known curves. This data can come from many different drivers who have all ridden the specific curves. It is not necessary for the same group of motorcycles or drivers to have ridden all the curves used. For some curves, the number of times they have been ridden may be more than for others.

[0014] Curve data can be exported from a map. This curve data may also include information about road signs and / or the area surrounding the curve. In particular, curve data can simulate the terrain and / or road surface conditions within the curve. Curve data can also simulate visual obstructions on curves caused by objects near the lane, such as trees, bushes, infrastructure, or rocks.

[0015] An algorithm can be a transfer function or a model. Algorithms can be embodied in artificial intelligence (AI), meaning they can be trained. During training, an algorithm can identify and represent relationships between motorcycle data and curve data. For example, for similar curves, it can look for similarities and / or differences in motorcycle data.

[0016] Car data can also be provided to trainable algorithms. Conversely, car data can be provided to trained algorithms. Car data can refer to reference data for categories of vehicles other than motorcycles or two-wheeled vehicles. For example, car data can also include truck data. Car data can represent the actual detected trajectory of a car as it navigates a curve. Car data can be used to train algorithms based on the relationships between geometry, car trajectories, and motorcycle trajectories. The car trajectory can be defined for each curve and include at least one target speed for the car at a specific point within the curve.

[0017] Car driving behavior can differ significantly from motorcycle driving behavior. However, certain specific conditions on curves can lead to similarities in driving behavior. Algorithms can learn which curves exhibit similar driving patterns. Since the number of registered cars is far greater than the number of registered motorcycles, the available data for cars is likely much more abundant than that for motorcycles. This allows car data to be used even for curves where there is little or no motorcycle data.

[0018] A geometric model can represent at least one value for each curve, derived from the following set of parameters: curve radius, curve direction change angle, road features before the curve, lane width, curve direction, visibility distance within the curve, and lane yaw angle within the curve. The geometric model can represent the corresponding curve as completely as possible. In particular, the geometric model can represent the visibility conditions while navigating the curve.

[0019] The trained algorithm can output a target speed curve for the motorcycle, which includes at least two target speeds for different target points on the curve. The target speed curve can include a reference point. The target points can be located at the reference point. The target points are placed at key points on the curve. Using multiple target points or speed curves, the assistance system can be controlled very conveniently.

[0020] The target speed of the motorcycle can be output for target points at predetermined intervals. For example, these intervals can be between 1 meter and 15 meters. Specifically, the intervals can be between 2.5 meters and 10 meters. In one embodiment, the interval can be five meters. The interval can represent a trade-off between accuracy and data volume. Smaller intervals allow for precise control but require more computing power, storage space, and / or transmission capacity. Larger intervals save on computing power, storage space, and / or transmission capacity but also result in coarser control accuracy.

[0021] This method is preferably implemented by a computer, for example, in software or hardware or in a hybrid form, such as in a driver assistance system.

[0022] The scheme described herein also proposes a controller configured to execute, control, and implement the steps of the method variants described herein in the appropriate device.

[0023] The controller can be an electronic device having at least one computing unit for processing signals or data, at least one storage unit for storing signals or data, and at least one interface and / or communication interface for reading or outputting data embedded in a communication protocol. The computing unit can be, for example, a signal processor, a so-called application-specific integrated circuit (ASIC), or a microcontroller, for processing sensor signals and outputting data signals based on the sensor signals. The storage unit can be, for example, flash memory, EPROM, or magnetic storage. The interface can be configured as a sensor interface for reading sensor signals from a sensor and / or an actuator interface for outputting data signals and / or control signals to an actuator. The communication interface can be configured to read or output data wirelessly and / or via a wired connection. The interface can also be a software module, which, for example, resides on the microcontroller along with other software modules.

[0024] Advantageously, there is also a computer program product or computer program containing program code, which can be stored on a machine-readable carrier or storage medium such as semiconductor memory, hard disk memory or optical memory, and is used to perform, implement and / or control the steps of the method described in any of the above embodiments, especially when the program product or program is executed on a computer or device.

[0025] It should be noted that this document describes some possible features and advantages of the invention with reference to different embodiments. Those skilled in the art will recognize that the features of the controller and method can be appropriately combined, adjusted, or replaced to obtain other embodiments of the invention. Attached Figure Description

[0026] Embodiments of the present invention are described below with reference to the accompanying drawings, which should not be construed as limiting the present invention.

[0027] Figure 1 A method for simulating the driving behavior of a motorcycle on a curve, and a flow chart for determining a target speed of a motorcycle for a curve, according to an embodiment, are shown.

[0028] This diagram is for illustrative purposes only and is not drawn to scale. The same reference numerals indicate the same or equivalent features. Detailed Implementation

[0029] Figure 1 A method flow according to an embodiment is shown. On one hand, Figure 1A method for modeling the driving behavior 100 of a motorcycle on a curve 102 using a trainable algorithm 104, according to one embodiment, is shown. On the other hand, Figure 1 A method for determining a target speed 106 of a motorcycle on a curve 102 using a trained algorithm 104 is shown according to one embodiment.

[0030] During training, curve data 108 and motorcycle data 110 are read in. Curve data 108 represents curve 102 and reflects the characteristics of each curve 102. Curve data 108 can also describe the corresponding visibility of curve 102. Motorcycle data 110 represents the driving behavior 100 of different motorcycles or different motorcycle drivers on curve 102. A relationship is learned from curve data 108 and motorcycle data 110 and embodied in algorithm 104. This relationship specifically represents how the characteristics of the curve affect the turning speed of the motorcycle.

[0031] In order to determine the target speed 106 of a motorcycle for a curve 102 without motorcycle data 110, the trained algorithm 104 reads the curve data 108 of the relevant curve 102 and, using the learned relationship, outputs at least one target speed 106 of a motorcycle for at least one characteristic target point 112 on the curve 102.

[0032] In one embodiment, motorcycle data 110 at least represents the speed curve of a motorcycle traveling on curve 102. Algorithm 104 then represents the relationship between the characteristics of curve 102 and the speed curve of the motorcycle on curve 102. This allows for the output of multiple target motorcycle speeds 106 for different target points 112 on an unknown curve 102. For example, the multiple target points 112 can be distributed at regular intervals along curve 102.

[0033] In one embodiment, during training, in addition to motorcycle data 110, car data 114 is also read for curve 102 as represented in curve data 108. Car data 114 represents the car's driving behavior 116 in curve 102. Algorithm 104 thus learns the similarities and differences between motorcycle driving behavior 100 and car driving behavior 116 on the same curve 102. When determining the target speed 106 of the motorcycle, the expected motorcycle driving behavior 100 can then be inferred for curve 102 using the known car driving behavior 116.

[0034] The possible designs of the present invention are summarized again below, or described in slightly different terms.

[0035] A method using machine learning, employing map data and, in particular, vehicle data, is proposed to calculate the target speed applicable to motorcycle cornering assist systems.

[0036] Motorcyclists are warned when approaching a curve at excessive speed and are given a speed assist selection for the curve. This can be achieved, for example, by deriving the appropriate curve speed and target speed for the curve speed warning system from the motorcyclist's actual speed profile on the relevant curve.

[0037] The proposed method involves an improved estimation or calculation of the target speed or target speed curve of a motorcycle, based on map data and vehicle data.

[0038] These values ​​can be used, for example, in warning systems that alert motorcyclists to exceeding a safe speed appropriate for the current situation. They can also be used in control systems, such as enhanced adaptive cruise control (ACC) systems. Currently, such systems on motorcycles only adjust based on a fixed target speed set by the driver, the speed of the vehicle ahead, or a legal speed limit (e.g., obtained from a digital map). Using the proposed method, adjustments can also be made based on a target speed that deviates from the aforementioned values, for example, during a turn, thus allowing for safe passage.

[0039] In 2023, 4% of all motorcycle accidents in Germany were caused by improper speed selection when entering a corner. In fatal accidents, this figure reached 6%, equivalent to 25 lives. While the specific causes of accidents can be further broken down into factors such as road conditions, improper braking or steering, and poor trajectory selection, most factors can be controlled by choosing an appropriate speed.

[0040] The corresponding assistance system requires one or more suitable target cornering speeds (v_target). The following section uses a simplified cornering model, which can be further refined in subsequent expansions. The basic model divides the corner into three phases.

[0041] The first stage is to approach the curve and decelerate so that the curve speed (v_target) is reached approximately at or shortly after entering the curve.

[0042] The second stage involves cornering at a near-constant speed (v_target) ("speed plateau") until reaching the apex of the corner.

[0043] The third stage is exiting the curve, which means leaving the platform and accelerating to the next suitable speed or the previously selected speed.

[0044] Determining the v_target value is crucial for the system's usability and acceptability. A value that is too low can lead to overuse, meaning the assistance system, based on this value, will unnecessarily warn the driver frequently or be programmed to take curves at unnecessarily slow speeds. This is known to cause user acceptance issues and may also affect traffic behind. If the value is chosen too high, the system will obviously offer no benefit. Assist warnings will be lacking, or even adventurous drivers might deliberately speed to trigger warnings, creating additional risks.

[0045] Besides the curve radius, the speed v_target depends on many other factors, such as the angle of change of direction, the characteristics of the road ahead, the lane width, rainfall, the direction of the curve, the distance from the line of sight, and the inclination of the lane.

[0046] For example, there are often two types of curves that have the same geometry (radius, length, gradient), but test drivers often pass them at different speeds. Analysis shows that in this case, the difference in visibility is the decisive factor. In the second curve, due to the increased terrain and tree obstruction, the exit is almost invisible, so the test drivers travel at significantly lower speeds than in the first curve.

[0047] To determine a meaningful target speed for a specific curve (ideally all curves globally), it is impractical to acquire comprehensive and statistically significant data, particularly on motorcycle driving. Traditional methods cannot determine a suitable target speed solely based on information from digital maps; therefore, acquiring empirical data remains indispensable.

[0048] The proposed solution addresses the problem of estimating the cornering speed of motorcycle drivers by deriving the cornering speed of motorcycle drivers from existing passenger car cornering speed data.

[0049] By 2024, vehicle-to-everything (V2X) systems for anonymous passenger vehicle data collection had rapidly become widespread. This was entirely due to the significant increase in the number of vehicles and the higher level of connectivity-assisted systems deployed. Preliminary investigations indicate that, in Germany alone, a nationwide and rapidly growing database already exists, with most curves possessing ample data points for aggregation.

[0050] Many factors affecting cornering speed are the same for cars (four-wheeled) and motorcycles (two-wheeled), especially the field of vision mentioned above—although seat height causes differences in field of vision, the degree of reduction in field of vision is relatively similar. Therefore, at the same cornering radius, the speeds of every two exemplary test cars, both two-wheeled and four-wheeled, are similar.

[0051] Despite these significant similarities, the differences between the different car groups are also evident. Motorcycles, for example, may have a smaller turning radius due to their narrower frame, but require more frequent speed adjustments on rough roads. Furthermore, motorcycles generally have stronger acceleration—largely due to their more sporty riding style compared to cars.

[0052] Preliminary tests have found that on the same curves, motorcyclists' average turning speed is about 10% higher than that of cars, even though the actual maximum possible speed limit of a motorcycle on a curve is usually lower than that of a car.

[0053] Speed ​​differences on many curves vary significantly depending on the individual characteristics of each curve, so simply using a scaling factor to convert a car's turning speed to a motorcycle's turning speed is not effective.

[0054] Therefore, the basic idea here is to deduce the target speed of the motorcycle on the same curve based on the average speed of a car passing through the same curve, that is, in the simplest case: v_target_2W=f(v_target_4W).

[0055] Of course, it's possible that a car travels through two curves at the same speed, but a motorcycle travels through them at different speeds. Therefore, the previously explained influencing factors come into play again and are incorporated into the transfer function: v_target_2W=f(v_target_4W,g), Here, g is a feature vector that describes all relevant properties of the curve, such as: g = (radius, road width, slope, μ, ...).

[0056] The proposed method applies the classic regression process in machine learning to a novel combination of input features (v_target_4W,g) and output target velocity v_target_2W.

[0057] This method involves several steps. The first step is creating training data. First, existing 2W real-world data (possibly combined with map data) is used to identify the corresponding curves. Then, the target speed v_target_2W is extracted from the 2W data. In a simple design, this can be calculated as v_target_2W := the 50th percentile of the existing speed at the midpoint of the curve. For the same curves, the same method is used to determine the target speed v_target_4W in the 4W database. For the identified curves, the corresponding map information g is extracted using map data.

[0058] In the second step, a machine learning algorithm is trained to learn the previously defined function f. This can be accomplished using standard regression methods (such as random forests) in simple designs.

[0059] In subsequent applications, for regions where there is no target velocity in two rounds, as mentioned earlier, a feature vector is formed using a map and four rounds of data. Then, the trained model is used to predict the target velocity v_target_2W' in two rounds.

[0060] In the previously described embodiment, a separate speed value was used and learned / estimated for each curve.

[0061] In a more general implementation, the velocity curve is learned and estimated, meaning the parameter v_target_2W is expanded from a one-dimensional scalar to a vector: v_target_2W=(v_1,v_2,...,v_N), Here, N represents the number of reference points for the curve. For example, good results can be obtained when N is chosen to distribute the reference points at equal intervals of 5 meters.

[0062] In another embodiment, instead of four-wheel data, only map data is used as input features for machine learning. While this may initially lead to a decrease in predicted speed, this decrease can be kept to an acceptable level through improved machine learning methods. The main advantage of 4W data is that it implicitly represents features such as "visibility" and road curvature (the driver's response to potential combinations of curves), which can also be represented using appropriate map data (e.g., contour lines, building information, etc.) and improved machine learning methods.

[0063] Finally, it should be noted that words such as "having" and "comprising" do not exclude other elements or steps, and words such as "an" or "a" do not exclude multiple elements. Reference numerals in the claims should not be considered limiting.

Claims

1. A method for simulating the driving behavior (100) of a motorcycle in a curve (102), wherein, Motorcycle data (110) and curve data (108) are provided to a trainable algorithm (104), wherein the motorcycle data (110) represents the actual detected motorcycle trajectory when the motorcycle passes through the curve (102), and the curve data (108) represents the geometry and constraints of the curve (102) so as to train the algorithm (104) based on the relationship between the geometry on the curve (102) and the motorcycle trajectory, wherein the motorcycle trajectory of each curve (102) contains at least one motorcycle target speed (106) for a target point (112) of the curve (102).

2. The method according to claim 1, wherein, Car data (114) is also provided to the trainable algorithm (104), wherein the car data (114) represents the actual detected car trajectory as the car passes through the curve (102), so as to train the algorithm (104) for the relationship between geometry, car trajectory and motorcycle trajectory, wherein the car trajectory of each curve (102) includes at least one car target speed for a target point of the curve (102).

3. The method according to any one of the preceding claims, wherein, The geometry of each curve (102) is represented by at least one of the following values: the curve radius of the curve (102), the direction change angle of the curve (102), the road features ahead of the curve (102), the lane width of the curve (102), the direction of the curve (102), the visibility distance in the curve (102), and the lane yaw angle in the curve (102).

4. A method for determining a target speed (106) of a motorcycle on a curve (102), wherein, The curve data (108) is provided to the trained algorithm (104), wherein the curve data (108) represents the geometry and constraints of the curve (102), and wherein the algorithm (104) outputs at least one target speed (106) of the motorcycle for a target point (112) of the curve (102).

5. The method according to claim 4, wherein, The algorithm (104) outputs a target speed curve for a motorcycle, which includes at least two target speeds (106) for different target points (112) of the curve (102).

6. The method according to claim 5, wherein, For the target point (112), the target speed of the motorcycle (106) is output at predetermined intervals.

7. The method according to any one of claims 4 to 5, wherein, The algorithm (104) is also provided with vehicle data (114), wherein the vehicle data (114) represents the actual vehicle speed curve detected when the vehicle passes through the curve (102).

8. A controller, wherein, The controller is configured to execute, implement, and / or control the method according to any one of the preceding claims in the corresponding device.

9. A computer program product configured to instruct a processor to perform, implement, and / or control the method according to any one of claims 1 to 3 or 4 to 7 when the computer program product is executed.

10. A machine-readable storage medium having stored thereon a computer program product according to claim 9.

Images