Method and device for generating at least one training travel traffic for training a driving

Abstract

The invention relates to a method for generating at least one training driving trajectory for training a driving mode of an autonomous vehicle (105), the method comprising the steps of: reading a possible driving trajectory of the vehicle (105) and a safety time parameter representing a time period, the vehicle (105) should comply with a safe distance from the assumed preceding vehicle (110) again within the time period after a deviation in position and / or speed on the travel trajectory; determining position deviation data and / or speed deviation data in order to change an assumed position and / or speed of the vehicle (105) on the travel trajectory at a predefined point in time; and determining a training driving trajectory by using the driving trajectory, the safe time parameter, the positional deviation data and / or the speed deviation data, wherein the training driving trajectory is determined such that a safe distance to the assumed preceding vehicle (110) is performed at least after a duration corresponding to the safe time parameter has elapsed.

Description

Technical Field

[0001] The invention is based on a method or an arrangement according to the preambles of the independent claims. A subject of the invention is also a computer program. Background Art

[0002] In imitation learning, the training data is augmented, i.e., supplemented with reduced distribution mismatches. In order to continue to have realistic training data, the existing data is usually modified within narrow limits. For example, an effective and popular method is used in which the driven trajectory is modified by additional, time-limited acceleration profiles (longitudinal or lateral). The training data is thus supplemented with states that deviate from the demonstration but are adjusted back to the target state again. In this way, the model should learn stable behavior so that it can compensate for deviations from the demonstrated behavior itself.

[0003] However, in more precise terms, this type of longitudinal perturbation is not ideal. If the training data is to be supplemented with fast driving, the additional acceleration profile should first accelerate (phase 1) and then brake again (phase 2). The perturbed trajectory (i.e. deliberately partially "distorted" or "perturbed" by perturbation parameters) has a higher average speed than the unperturbed trajectory. In the case of follow-up driving, there are two possibilities for synchronizing the two trajectories:

[0004] - Close following is achieved by aligning the perturbed and non-perturbed trajectories in position at the beginning of phase 2. The target behavior that the strategy should learn (phase 2) is therefore braking, but the autonomous vehicle (English: self-driving vehicle = SDV) follows the person in front much more closely than was the case in the undisturbed data. The learned behavior remains unsafe.

[0005] - Alternatively, the overspeed is completely adjusted to a safe state, i.e. the perturbed and undisturbed trajectories coincide at the end of phase 2. The states seen in the training then have higher speeds, but also greater safety distances, than the undisturbed data. The training data set is only inadequately supplemented with critical situations.

[0006] In both cases, it is impossible to achieve a truly unsafe state that includes safe continuations in the training data.

[0007] In a completely different approach to distribution mismatch, unsafe or critical situations are created in simulation using trained planning or forecasting models, which are then corrected again by expert models. However, these approaches can only be applied conditionally in practice, since both realistic simulations and expert models must be available. Summary of the invention

[0008] Against this background, the solution proposed here provides a method, a device and / or a control device using the method, and finally a corresponding computer program according to the independent claims. Advantageous expansions and improvements of the device described in the independent claims are made possible by the measures listed in the dependent claims.

[0009] The present invention provides a method for generating at least one training driving trajectory for training a driving style of an autonomous vehicle, wherein the method comprises the following steps:

[0010] - reading a possible driving trajectory of the vehicle and a safety time parameter, which indicates the time period within which the vehicle must again maintain a safe distance to the assumed vehicle ahead after a deviation in position and / or speed on the driving trajectory;

[0011] - determining position deviation data and / or speed deviation data in order to change an assumed position and / or speed of the vehicle on the driving trajectory at a specific and / or predefined point in time; and

[0012] - determining a training driving trajectory by using the driving trajectory, the safety time parameter, the position deviation data and / or the speed deviation data, wherein the training driving trajectory is determined so that a safe distance to an assumed preceding vehicle is observed at least after a time period corresponding to the safety time parameter has elapsed.

[0013] The driving trajectory of a vehicle can be understood as a motion path with time information or a trajectory representing the assumed driving path of the vehicle. In the present case, the safety time parameter can be understood as a parameter within which the vehicle, after deviating from the driving trajectory in terms of speed and / or position, reaches a state again in which the vehicle continues to travel at a safe distance from the assumed vehicle traveling ahead. However, the vehicle traveling ahead (which can also be referred to as the vehicle ahead) should only be virtually assumed, so that for the training driving trajectory to be determined in the end and used to train the driving behavior of the autonomous vehicle in the following method, the vehicle traveling ahead does not need to exist or be taken into account. On the contrary, the assumption of the position of the vehicle traveling ahead is described in the order of the individual positions of the vehicle in the training driving trajectory. Position deviation data can be understood as position data, at each position, the orientation (Orientierung) and / or position of the vehicle on the driving trajectory can be changed or moved by as much as the position data, so that the disturbance of the position on the driving trajectory can be depicted (abbilden) or simulated for generating the training driving trajectory. Similarly, speed deviation data can be understood as speed data, to the extent of which the speed of the vehicle at various positions on the driving trajectory can be changed, so that disturbances of the speed at different positions of the driving trajectory can also be depicted or simulated for the purpose of generating a training driving trajectory. In the present case, a safety distance can be understood as a distance that represents a safety distance from an assumed vehicle ahead, which the vehicle moving on the training driving trajectory should comply with after a time period corresponding to the safety time parameter has passed. Therefore, the training driving trajectory should be determined in such a way that, for example, in the case of driving closely behind an assumed vehicle ahead, the desired or required safety distance is fulfilled or complied with again after a short period of time.

[0014] The solution proposed here is based on the recognition that by using position deviation data and / or speed deviation data when modifying the driving trajectory of the vehicle taking into account the safety distance parameter, it is possible to find a possibility: even in critical situations, for example when driving closely behind the vehicle in front, the driving behavior of the vehicle can be described, without actually having to record the data of the vehicle in front in the training driving trajectory, and on the other hand, it is possible to describe the driver's driving on the road very close to reality. This close to reality driving can be described, for example, so that now also driving situations can be described in the training driving trajectory, in which the vehicle is driving more closely behind or closer to the vehicle in front at a slower driving speed, while still driving in a compliant and safe manner in traffic. The training driving trajectory to be determined can thus make it possible to expand the driving curve to be trained, so that the driving behavior can not only be statically adjusted to a fixed predetermined safety distance, but also this safety distance can be changed as a function of the speed and thus the precise training of the automatic driving algorithm is significantly improved, especially in critical driving situations.

[0015] The proposed approach enables the determination of training driving trajectories by generating perturbations of the (driving) trajectory, which on the one hand contain critical situations but on the other hand can also be safely corrected again. The disadvantages of conventional perturbations are thus overcome without having to rely on interactive simulations or expert models.

[0016] An embodiment of the proposed solution is particularly advantageous in which, in the determination step, the position deviation data and / or the speed deviation data are determined by using a probability distribution for the position deviation data and / or the speed deviation data. As probability distribution, for example, a normal distribution, a Poisson distribution, etc. can be used. Such an embodiment offers the advantage that a quasi-random selection of the position deviation data and / or the speed deviation data is generated in such a way that the randomness of the following disturbances is generated, by which the driving trajectory for the training driving trajectory is changed. As a result, it is achieved that the driving behavior of the autonomous vehicle is improved and more robust.

[0017] An embodiment of the solution proposed here can also be envisaged, in which, in the reading step, a time period parameter is read, which time period parameter indicates the time required for the vehicle to reach the assumed preceding vehicle at the current assumed speed, and in which, in the determination step, the training driving trajectory is determined by using the time period parameter. Alternatively or additionally, in the reading step, a safety distance parameter is read, which safety distance parameter indicates the safety distance to the assumed preceding vehicle that the vehicle should observe when stationary, and in which, in the determination step, the training driving trajectory is determined by using the safety distance parameter. Such a time period parameter can, for example, be understood as the time or time interval required to follow the preceding vehicle in a known assumed safety area up to a fixed predefined safety distance. The explicit selection of such a time period parameter enables a flexible setting of the driving behavior that can subsequently be trained to the autonomous vehicle.

[0018] According to another embodiment of the solution proposed here, in the determination step, the training driving trajectory can be determined by using an acceleration, wherein the acceleration is determined by using the current assumed position of the vehicle on the driving trajectory, the current assumed speed of the vehicle on the driving trajectory, the position deviation data, the speed deviation data, the safety time parameter and the time period parameter. In particular, the temporally flexible and variable acceleration determination proposed here enables very precise changes in the assumed position of the vehicle and / or determination of the speed of the vehicle at these positions, so that the above advantages can be effectively implemented or realized.

[0019] An implementation of the scheme proposed here can be implemented particularly accurately and adapted to the desired requirements. In the implementation, in the determination step, a training driving trajectory is determined by using an assumed position of the vehicle at a specific point in time, wherein the assumed position of the vehicle is determined by using an assumed position of the vehicle on the driving trajectory at that point in time, a current assumed speed of the vehicle on the driving trajectory at that point in time, position deviation data, speed deviation data, safety time parameters and time period parameters.

[0020] Similarly, an embodiment of the scheme proposed here can also be used to precisely adapt to the desired framework conditions, in which embodiment, in a determination step, a training driving trajectory is determined by using an assumed speed of the vehicle at a specific point in time, wherein the assumed speed of the vehicle is determined by using an assumed position of the vehicle on the driving trajectory at that point in time, a current assumed speed of the vehicle on the driving trajectory at that point in time, position deviation data, speed deviation data, safety time parameters and time period parameters.

[0021] In order to be able to depict the respective position and / or speed of the vehicle over a longer period of time and in this case to comply with the desired boundary conditions, according to one embodiment of the proposed solution, in the determination step, the respective assumed position and / or speed of the vehicle in the training driving trajectory can be determined by each determination of these positions and / or speeds on the training driving trajectory for different points in time. For this purpose, for example, each (assumed) position and / or (assumed) speed of the vehicle is determined for each point in time or each time interval and is stored in the training driving trajectory for the corresponding point in time or corresponding time interval.

[0022] In order to obtain a training driving trajectory that is as close to reality as possible, according to one embodiment of the solution proposed here, in the determination step, the training driving trajectory can be determined in such a way that the assumed speed of the vehicle is non-negative at all positions in the training driving trajectory. In this way, certain technically unrealistic or physically meaningless values ​​may be immediately discarded.

[0023] Similarly, in order to depict a driving behavior of the vehicle that is as close to reality as possible, according to another embodiment of the solution proposed here, the training driving trajectory can be determined in the determination step so that the lateral acceleration acting on the vehicle moving according to the training trajectory does not exceed the lateral acceleration threshold value. In the present case, the lateral acceleration threshold value can be understood as a value that represents the following acceleration acting on the vehicle, for example in the lateral direction, under which the vehicle will drive out of the curve.

[0024] In particular, in order to optimize the learned behavior of an autonomous vehicle over longer driving distances or longer time periods, according to another embodiment of the scheme proposed here, at least one of the steps of the method can be iteratively repeated, in particular in the case where the training driving trajectory represents the vehicle driving behavior within a time interval that is larger than the safety time parameter.

[0025] The method can be implemented, for example, in the form of software or hardware or in the form of a mixture of software and hardware, for example in a control unit.

[0026] The solution proposed here also creates a device, which is designed to perform, control or implement the steps of the variant of the method proposed here in a corresponding device. This embodiment variant of the invention in the form of an apparatus can also quickly and effectively achieve the object on which the invention is based.

[0027] For this purpose, the device can have at least one computing unit for processing signals or data, at least one storage unit for storing signals or data, at least one interface to a sensor or actuator for reading a sensor signal from a sensor or for outputting data or control signals to an actuator and / or at least one communication interface for reading or outputting data embedded in a communication protocol. The computing unit can be, for example, a signal processor, a microcontroller, etc., wherein the storage unit can be a flash memory or a magnetic storage unit. The communication interface can be designed to read or output data wirelessly and / or wired, wherein a communication interface that can read or output wired data can, for example, read the data electrically or optically from a corresponding data transmission line or output it to a corresponding data transmission line.

[0028] In the present case, the device can be understood to mean an electrical device that processes sensor signals and outputs control and / or data signals according to the sensor signals. The device can have interfaces that can be designed in hardware and / or software. In the case of hardware design, these interfaces can be, for example, part of a so-called system ASIC, which contains various functions of the device. However, these interfaces may also be their own integrated circuits or at least partially composed of discrete devices. In the case of software design, these interfaces can be, for example, software modules that exist on a microcontroller together with other software modules.

[0029] Also advantageous is a computer program product or a computer program with a program code, which can be stored on a machine-readable carrier or storage medium, such as a semiconductor memory, a hard disk memory or an optical memory, and is used to execute, implement and / or control the steps of a method according to one of the above-described embodiments, in particular when the program product or the program is executed on a computer or a device. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] An exemplary embodiment of the solution proposed here is shown in the drawings and explained in more detail in the following description.

[0031] Figure 1 A schematic diagram depicting the creation of a training driving trajectory for a motion path of a vehicle is shown;

[0032] Figure 2 A flow chart of a method according to an embodiment is shown;

[0033] Figure 3 A flow chart illustrating an embodiment of a method for generating at least one training driving trajectory to train a driving style of an autonomous vehicle;

[0034] Figure 4 A block diagram showing an embodiment of an apparatus for generating at least one training driving trajectory to train a driving style of an autonomous vehicle; and

[0035] Figure 5 A flow chart of an embodiment of a method 500 for training an autonomous driving vehicle is shown. DETAILED DESCRIPTION

[0036] In the following description of advantageous embodiments of the invention, identical or similar reference numerals are used for elements which are illustrated in different figures and have a similar function, wherein a repeated description of these elements is omitted.

[0037] Figure 1 A schematic diagram of the creation of a training driving trajectory 100 is shown, which depicts the movement path of a vehicle 105, which is driving, for example, behind a front vehicle 110. The training driving trajectory 100 to be determined using the approach proposed here is used to train the driving behavior of an autonomous vehicle so that the driving behavior of the autonomous vehicle ultimately behaves like that of vehicle 105. To determine the training driving trajectory 100, a driving trajectory 115 is first provided or read into a corresponding generation unit, wherein the driving trajectory 115 depicts the vehicle 105 at a specific position p at a time point i=1, ... n. i and has a specific speed v i In this case, however, it is assumed that the vehicle 105 is always at a certain safety distance Δ behind the vehicle 110 in front, wherein the safety distance Δ consists of a fixed safety distance λ at rest and a speed-dependent safety distance, wherein the speed-dependent safety distance is composed of the product of the (assumed) current speed v of the vehicle 105 and the safety time period τ. Therefore, an important aspect of the solution proposed here can be seen in the definition of the safety distance Δ to the (imaginary) person 110 in front, which can be defined according to the following relationship:

[0038] Δ=λ+τ·v

[0039] Where applicable:

[0040] λ: Safety distance when stationary

[0041] τ: a parameter, such as 1 to 2 seconds in following travel, and

[0042] v: current speed.

[0043] In order to now obtain the training driving trajectory 100 from the driving trajectory 115 , one or more individual time points in the training data of the driving trajectory 115 are perturbed, for example, in the following manner:

[0044] The position of the vehicle 105 moves forward or backward along the traveled (driving) trajectory 115 by the position deviation data δ p So much that position pi The change in the position of the relevant time point

[0045] Alternatively or additionally, the scalar speed of the vehicle 105 due to the speed deviation data δ v and increase or decrease, so that the speed v i Find the speed at the relevant time point

[0046] The future trajectory as training driving trajectory 100 represents the desired behavior of the autonomous vehicle and is modified by the constant speed profile so that after a set time interval θ, a safety distance is again present, i.e., a safety distance Δ to the (imaginary) person in front 115. This is in accordance with the reasonable assumption that the safety distance is always observed in the undisturbed data.

[0047] This consideration makes it possible, for example, to: Instead, a safety distance Δ is adjusted, which is smaller than the distance in the undisturbed data, but more slowly.

[0048] Figure 2 A diagram is shown, which shows the driving trajectory 115 (from Figure 2 The safety distance in the original situation in the figure above) is relative to the safety distance from the training driving trajectory 100 (from Figure 2 Here it is assumed that: due to the speed deviation data δ v , now assuming a lower speed of vehicle 105 This allows the safety distance Δ to be reduced without risking endangering the vehicle 105 .

[0049] The following assumptions may also be used to generate the training driving trajectory 100:

[0050] - the expert whose driving behavior is to be imitated always maintains a safe distance ahead; and

[0051] The model Δ=λ+τ·v describes the safety distance sufficiently well.

[0052] For example, the algorithm described below may be used to determine the various positions and / or velocities. In this case, the initial state of the autonomous vehicle 105SDV (at time i with a velocity v i At position p along the trajectory i ), whose driving behavior is distributed according to probability, e.g., as position deviation data The interval [-2; 2]m and / or as speed deviation data The disturbed state may violate the safety distance Δ, which is expected in the training data of the training driving trajectory 100 .

[0053] In the case of disturbances of this state, care must also be taken to avoid a collision with a (hypothetical) person in front or with vehicle 110 in front, taking into account the maximum acceleration a. Disturbances that violate this assumption can be discarded.

[0054] The future trajectory or training driving trajectory 100 is now changed with a constant acceleration curve a in such a way that a safe state is reached again at the time θ.

[0055] At this point in time, the (imaginary) person in front follows trajectory p f +Δ=p f +λ+τ·v f Therefore, the disturbed state should satisfy the following equation:

[0056]

[0057] Here it applies (constant additional acceleration a):

[0058]

[0059] These equations can have the following solutions:

[0060]

[0061] Additional restrictions can be easily integrated, for example to prevent reverse driving in the target trajectory, ie a non-negative speed should be required, or kinematic boundary conditions such as a maximum lateral acceleration in a curve should be observed, ie a maximum lateral acceleration threshold should not be exceeded.

[0062] If the target trajectory is to cover a greater time range than the safety time parameter θ, the algorithm can be applied iteratively, ie one or more steps can be repeatedly performed. The perturbed trajectory or training driving trajectory 100 is then increasingly adapted to the undisturbed driving trajectory 100 .

[0063] The probability distribution for disturbing the initial state as well as the time period parameter τ and the safety time parameter θ are parameters of the exemplary algorithm presented here and are to be selected as a function of the training data or driving trajectory 115 . This enables the algorithm to be used flexibly and in a variety of ways.

[0064] Figure 3A flow chart of an embodiment of a method 300 for generating at least one training driving trajectory for training a driving style of an autonomous vehicle is shown. The method 300 comprises a step 310 of reading a possible driving trajectory of the vehicle and a safety time parameter, the safety time parameter representing a time period within which the vehicle should again comply with a safety distance to an assumed vehicle ahead after a deviation in position and / or speed on the driving trajectory. In addition, the method 300 comprises a step 320 of determining position deviation data and / or speed deviation data in order to change an assumed position and / or speed of the vehicle on the driving trajectory at a specific and / or predefined time point. Finally, the method 300 comprises a step 330 of determining a training driving trajectory by using the driving trajectory, the safety time parameter, the position deviation data and / or the speed deviation data, wherein the training driving trajectory is determined such that the safety distance to the assumed vehicle ahead is complied with at least after a time period corresponding to the safety time parameter has elapsed.

[0065] Figure 4 A block diagram of a device 400 for generating at least one training driving trajectory for training the driving style of an autonomous vehicle is shown. The device 400 comprises a reading unit 410 for reading a possible driving trajectory of the vehicle and a safety time parameter, the safety time parameter representing the following time period, within which the vehicle should again comply with the safety distance to the assumed front vehicle after a deviation in the position and / or speed on the driving trajectory. In addition, the device 400 also comprises a unit 420 for determining position deviation data and / or speed deviation data in order to change the assumed position and / or speed of the vehicle on the driving trajectory at a specific and / or predefined time point. Finally, the device 400 comprises a unit 430 for determining the training driving trajectory by using the driving trajectory, the safety time parameter, the position deviation data and / or the speed deviation data, wherein the training driving trajectory is determined in such a way that the safety distance to the assumed front vehicle is complied with at least after a period of time corresponding to the safety time parameter has passed.

[0066] Figure 5 A flow chart of an exemplary embodiment of a method 500 for training an autonomously driven vehicle is shown, wherein the method has a step 510 of adapting a driving algorithm by using a training driving trajectory, wherein the training driving trajectory is generated according to a variant of the method proposed here for generating a training driving trajectory.

[0067] In summary, it should be noted that data-based learning methods in modern vehicles are increasingly being used in highly automated driving. Current research areas here are particularly the planning step, which uses an environment model to generate a target behavior for an automated vehicle (also called SDV = Self-Driving vehicle). One problem here is the "covariant shift" or "distribution mismatch": if the policy (Policy) enters a state in the application that was not adequately represented in the training data, New situations that are not represented or not represented at all usually indicate incorrect behavior. Even simple following driving is particularly obvious, for example: the expert whose behavior generates the training data maintains a reasonable safety distance from the person in front. Critical situations in which following driving closely are therefore missing from the training data. As a result, the strategy does not learn how to react when the safety distance is less than the safety distance. The approach proposed here creates a very concise and robust training method for this problem, so that training scenarios that cannot be described in advance can also be covered, thereby achieving a significant improvement in the robustness of the autonomous driving algorithm.

[0068] If an embodiment includes an “and / or” association between a first feature and a second feature, this should be understood to mean that the embodiment has both the first feature and the second feature according to one embodiment, and has either only the first feature or only the second feature according to another embodiment.

Claims

1. A method (300) for generating at least one training driving trajectory (100) for training a driving style of an autonomous vehicle (105), wherein the method (300) comprises the following steps: - reading (310) a possible driving trajectory (115) of the vehicle (105) and a safety time parameter (θ), the safety time parameter (θ) representing the time period during which the position (p) of the vehicle (105) on the driving trajectory (115) i ) and / or speed (v i ) after the deviation occurs, a safety distance (Δ) from the assumed preceding vehicle (110) should be observed again within the said time period; - Determine (320) position deviation data (δ p ) and / or speed deviation data (δ v ) to change the assumed position of the vehicle (105) on the driving trajectory (115) at a specific and / or predefined time point and / or speed and - by using the driving trajectory (115), the safety time parameter (θ), the position deviation data (δ p ) and / or the speed deviation data (δ v ) is used to determine (330) the training driving trajectory (100), wherein the training driving trajectory (100) is determined so that a safe distance (Δ) from the assumed leading vehicle (110) is maintained at least after a period of time corresponding to the safety time parameter (θ).

2. The method (300) according to claim 1, wherein: In the determination step (320), by using the position deviation data (δ p ) and / or speed deviation data (δ v ) is used to determine the position deviation data (δ p ) and / or the speed deviation data (δ v ).

3. The method (300) according to any one of the preceding claims, wherein: In the reading step (310), a time period parameter (τ) is read, wherein the time period parameter represents: the vehicle (105) at the current assumed speed (v i ) to reach the assumed leading vehicle (115), and wherein in the determining step (330), the training driving trajectory (100) is determined by using the time period parameter (τ), and / or wherein in the reading step (310), a safety distance parameter (λ) is read, the safety distance parameter representing the safety distance from the assumed leading vehicle (115) that the vehicle (105) should comply with when stationary, and wherein in the determining step (330), the training driving trajectory (100) is determined by using the safety distance parameter (λ).

4. The method (300) according to claim 3, wherein in the determining step (330) the training driving trajectory (115) is determined by using an acceleration (a), wherein the acceleration (a) is determined by using a current assumed position (p) of the vehicle (105) on the driving trajectory (115). f ), the current assumed speed (v) of the vehicle (105) on the driving trajectory (115) i ), the position deviation data (δ p ), the speed deviation data (δ v ), the safety time parameter (θ) and the time period parameter (τ).

5. The method (300) according to any one of the preceding claims 3 or 4, wherein: In the determining step (330), by using the assumed position of the vehicle (105) at a particular point in time (i) to determine the training driving trajectory (100), wherein the assumed position of the vehicle By using the assumed position (p) of the vehicle at the time point (i) on the driving trajectory (115) i ), the current assumed speed (v) of the vehicle at the time point (i) on the driving trajectory (115) i ), the position deviation data (δ p ), the speed deviation data (δ v ), the safety time parameter (θ) and the time period parameter (τ).

6. The method (300) according to any one of the preceding claims, wherein: In the determining step (330), by using the assumed speed of the vehicle (105) at a particular point in time (i) to determine the training driving trajectory (100), wherein the assumed speed of the vehicle (105) By using the assumed position (p) of the vehicle at the time point (i) on the driving trajectory (115) i ), the current assumed speed (v) of the vehicle at the time point (i) on the driving trajectory (115) i ), the position deviation data (δ p ), the speed deviation data (δ v ), the safety time parameter (θ) and the time period parameter (τ).

7. The method (300) according to any one of the preceding claims, wherein in the determination step (330) the respective assumed positions of the vehicle (105) in the training driving trajectory (100) are determined by each determination of the position and / or speed on the training driving trajectory (100) for different time points (i). and / or speed 8. The method (300) according to any one of the preceding claims, wherein: In the determining step (330), the training driving trajectory (100) is determined such that the assumed speed of the vehicle (105) All positions and / or time points (i) in the training driving trajectory (100) are non-negative.

9. The method (300) according to any one of the preceding claims, wherein: In the determination step (330), the training driving trajectory (100) is determined in such a way that a lateral acceleration acting on the vehicle (105) moving along the training trajectory (100) does not exceed a lateral acceleration threshold value.

10. The method (300) according to any one of the preceding claims, wherein: In particular, when the training driving trajectory (100) represents the driving behavior of the vehicle (105) within a time interval greater than the safety time parameter (θ), at least one of the steps (320, 330) of the method (300) is iteratively repeated.

11. A device (400) configured to execute and / or control the steps (310, 320, 330) of the method (300) according to any one of the preceding claims in a corresponding unit (410, 420, 430).

12. A method (500) for training an autonomous driving vehicle, wherein the method comprises: A step (510) of adapting a driving algorithm by using a training driving trajectory (100) generated according to a method according to any one of claims 1 to 10.

13. Computer program, which is configured to execute and / or control the steps of the method (300) according to any one of the preceding claims.

14. A machine-readable storage medium having stored thereon a computer program according to claim 12.

Images