Planning a path for automatically parking a motor vehicle, training a path-planning agent, and automatically parking a motor vehicle using reinforcement learning

Abstract

In order to plan a path for automatically parking a motor vehicle (1), surroundings data which represents the surroundings of the motor vehicle (1) is received, and a vehicle target state for the motor vehicle (1) is determined on the basis of the surroundings data. A first path for the motor vehicle (1) from the current vehicle state of the motor vehicle (1) to the vehicle target state is determined, wherein an initial driving maneuver according to the first path is a forward driving maneuver of the motor vehicle (1), a second path for the motor vehicle (1) from a current vehicle state of the motor vehicle (1) to the vehicle target state is determined, and an initial driving maneuver according to the second path is a reverse driving maneuver of the motor vehicle (1). On the basis of at least one specified rule, a path to be used in the process of automatically parking the motor vehicle (1) is selected from the first path and the second path.

Description

[0001] 2024PF01883

[0002] 1

[0003] Path planning for the automatic parking of a motor vehicle and training a path planning agent and automatic parking of a motor vehicle through reinforcement learning

[0004] The present invention relates to a computer-implemented method for path planning for the automatic parking of a motor vehicle, a computer-implemented training method for training a path planning agent for the automatic parking of a motor vehicle by reinforcement learning, and a method for the automatic parking of a motor vehicle, wherein such a computer-implemented method is carried out. The invention further relates to a data processing system for carrying out such a computer-implemented method or training method, an electronic vehicle guidance system, and corresponding computer program products.

[0005] In automated driving systems, numerous well-known algorithms exist for path planning, also known as trajectory planning. One class of such path planning algorithms is based on so-called agents, which have been trained using reinforcement learning.

[0006] The publication Y. Chen et al.: “Deep Reinforcement Learning in Autonomous Car Path Planning and Control: A Survey” (arXiv:2404.00340v1) provides an overview of literature on Deep Reinforcement Learning (DRL) in the field of path planning and control for vehicles.

[0007] Known algorithms for path planning in automatic parking can sometimes exhibit reduced accuracy, especially in difficult or complex parking scenarios. Another disadvantage is that they may plan complicated or long paths, which can negatively impact user convenience.

[0008] It is an object of the present invention to overcome at least some of the aforementioned disadvantages of automatic parking of a motor vehicle. 2024PF01883

[0009] 2

[0010] This problem is solved by the subject matter of the independent claim. Further embodiments and preferred developments are the subject matter of the dependent claims.

[0011] The invention is based on the concept of determining a target vehicle state during path planning based on environmental data and providing two paths from the current vehicle state to the target vehicle state, the initial driving maneuvers of which are directed in opposite directions, i.e., one forward and one backward. One of the two paths is then selected for actual use.

[0012] According to one aspect of the invention, a computer-implemented method for path planning for the automatic parking of a motor vehicle is provided. Environmental data representing the vehicle's surroundings are received, in particular from at least one environmental sensor system of the vehicle. A target vehicle state is determined based on this environmental data. A first path for the vehicle from its current state to the target state is determined, wherein an initial maneuver along the first path is a forward maneuver. A second path for the vehicle from its current state to the target state is determined, wherein an initial maneuver along the second path is a reverse maneuver.Based on at least one predefined rule, a path for use in the automatic parking of the motor vehicle is selected from the first path and the second path; in particular, depending on the at least one rule, either the first path or the second path is selected for use in the automatic parking of the motor vehicle.

[0013] Unless otherwise specified, all steps of the computer-implemented method can be performed by a data processing system comprising at least one data processing device, in particular by a data processing system of the motor vehicle. Specifically, the at least one additional data processing device is configured or adapted to perform the steps of the computer-implemented method. For this purpose, the at least one data processing device may, for example, store another computer program containing instructions which, when executed by the 2024PF01883

[0014] 3

[0015] at least one data processing device to cause at least one data processing device to execute the computer-implemented procedure.

[0016] All data processing devices of the at least one data processing device can be part of the motor vehicle. However, it is also possible that all data processing devices of the at least one data processing device are part of an external computing system outside the motor vehicle, for example, a mobile electronic device, a backend server, or a cloud computing system. It is also possible that the at least one data processing device comprises both at least one vehicle data processing device of the motor vehicle and at least one external data processing device of the external computing system.The at least one vehicle data processing device may, for example, comprise one or more electronic control units (ECUs), and / or one or more zone control units (ZCLIs), and / or one or more domain control units (DCLIs) of the motor vehicle.

[0017] In the event that the at least one data processing device comprises two or more data processing devices, certain steps performed by the at least one data processing device can be understood, for example, as different data processing devices performing different steps or different parts of a step. In particular, it is not necessary for each data processing device to perform the steps completely. In other words, the execution of the steps can be distributed among the two or more data processing devices.

[0018] Each embodiment of the computer-implemented method results in a corresponding embodiment of a method for path planning for the automatic parking of a motor vehicle, which is not purely computer-implemented, by including corresponding steps for generating the environmental data using the at least one environmental sensor system.

[0019] Receiving or receiving data or information can, for example, involve receiving or receiving the data, in particular through the relevant data processing system from a sending entity, or reading the data from a data storage device, or receiving or receiving a data stream containing the data, or extracting the data from the 2024PF01883

[0020] 4

[0021] This includes data streams and so on. Wired or wireless data transmission can be used for this purpose. Specifically, data transmission can occur between a hardware and / or software interface of the sending entity and a hardware and / or software interface of the data processing system.

[0022] An environmental sensor system can, for example, be understood as a sensor system that is capable of generating sensor data or sensor signals that map, represent, or reproduce an environment of the environmental sensor system.

[0023] In particular, the ability to detect electromagnetic or other signals from the environment is not sufficient to classify a sensor system as an environmental sensor system. For example, cameras, especially those operating in the visible or infrared range, radar systems, lidar systems, or ultrasonic sensor systems can be considered environmental sensor systems.

[0024] Depending on the implementation of the at least one environmental sensor system, the environmental data may therefore include, for example, one or more camera images, thermal images, lidar point clouds, lidar depth images, radar data, ultrasound data and / or corresponding images generated on the basis of the radar or ultrasound data.

[0025] A position for the motor vehicle can be understood here and in the following as the position of a predefined point, in particular on the motor vehicle.

[0026] For example, the center point of a front or rear axle of the motor vehicle can serve as a reference point for the position.

[0027] An orientation for the motor vehicle can be understood here and in the following as the direction of a vector originating from the point that defines the position for the motor vehicle, or originating from another predefined point.

[0028] Unless otherwise stated, a state for the motor vehicle can be understood here and in the following as being characterized at least by a position for the motor vehicle. Optionally, the state can be characterized by an orientation for the motor vehicle. A state, in particular a target state, for the motor vehicle can also be characterized by a position tolerance with respect to the position and / or an orientation tolerance with respect to the orientation. 2024PF01883

[0029] 5

[0030] A condition for the motor vehicle can also be characterized by a steering position, and so on.

[0031] In particular, the vehicle target state is characterized by a vehicle target position and / or a vehicle target orientation. For example, in addition to the vehicle target position, the vehicle target state is characterized by a position tolerance with respect to the vehicle target position and / or in addition to the vehicle target orientation, by an orientation tolerance with respect to the vehicle target orientation.

[0032] The vehicle target position corresponds in particular to a position in a parking space for the motor vehicle, and, for example, the vehicle target orientation corresponds to a direction or orientation of the parking space.

[0033] In particular, the current vehicle status is characterized by a current vehicle position and / or a current vehicle orientation.

[0034] The current vehicle position can be determined explicitly based on environmental data and / or using a self-localization algorithm. Similarly, the current vehicle orientation can be determined. Alternatively, the current vehicle position and orientation can be determined within a reference coordinate system, where the current vehicle position corresponds to a coordinate origin of the reference coordinate system. The current vehicle orientation can, for example, be at an angle of 0° with a predefined axis of the reference coordinate system, in particular a longitudinal axis of the reference coordinate system.The reference coordinate system can, for example, be a coordinate system fixed to the motor vehicle, the origin of which is located at the point mentioned above that defines the position of the motor vehicle, i.e., for example, in the center of the front axle or the rear axle of the motor vehicle.

[0035] The reference coordinate system can, for example, be a Cartesian coordinate system, where one axis of the reference coordinate system corresponds to a longitudinal axis of the vehicle and another axis of the reference coordinate system corresponds to a transverse axis of the vehicle perpendicular to it. 2024PF01883

[0036] 6

[0037] The first and second paths are each characterized, for example, by a respective final vehicle position for the motor vehicle and / or a final vehicle orientation for the motor vehicle.

[0038] The fact that the first path and the second path lead from the current vehicle state to the vehicle target state can be understood as follows: the first path and the second path each begin at the current vehicle position, in particular with the current vehicle orientation, and end at the respective final vehicle position, in particular with the respective final vehicle orientation, wherein the respective final vehicle position corresponds to the vehicle target position except for the position tolerance with respect to the vehicle target position, and the respective final vehicle orientation corresponds to the vehicle target orientation except for the orientation tolerance with respect to the vehicle target orientation.

[0039] Various criteria can be considered as a rule for selecting the path used in automatic vehicle parking. The selection can be random, for example. It is also possible to select the shorter path, the path with the fewest changes of direction, the path with the smallest deviation of the final vehicle orientation from the target orientation, the path with the smallest deviation of the final vehicle position from the target position, or the path with the greatest distance to obstacles, and so on. A more complex rule can also be implemented that considers two or more of these and / or additional criteria.It is also possible that the path is selected according to a defined specification, which may be application-specific or situation-dependent, or specified by a user or manufacturer of the motor vehicle.

[0040] The selected path can be considered the result of the path planning process. The vehicle can then, in particular, be automatically guided along the selected path to park.

[0041] The fact that the initial driving maneuver according to the first path is a forward driving maneuver of the motor vehicle can be interpreted as meaning that the first path begins with the motor vehicle moving forward. The fact that the initial driving maneuver according to the second path is a reverse driving maneuver of the motor vehicle can be interpreted as meaning that the first path begins with the motor vehicle moving backward. 2024PF01883

[0042] 7

[0043] According to the invention, two alternative paths are planned, each starting in a different direction of travel, and one of these is selected for use in the automatic parking of the motor vehicle. This allows the user to always select the path that is best suited according to the at least one predefined rule or is preferred for other reasons. This freedom of choice can increase the accuracy and / or reliability of the automatic parking and / or improve user convenience during automatic parking.

[0044] According to at least one embodiment, a weighted sum of two or more terms that characterize the respective path is calculated for both the first path and the second path.

[0045] At least one rule for selecting the path for use in automatic vehicle parking states, for example, that the path from the first and second paths whose weighted sum has the smaller value is selected. The signs of the corresponding weighting factors can be chosen depending on the sign of the respective term and whether a smaller or larger absolute value of the term is preferred. Alternatively, at least one rule for selecting the path for use in automatic vehicle parking can state, for example, that the path from the first and second paths whose weighted sum has the larger value is selected. In this case, the signs of the corresponding weighting factors must be chosen accordingly.

[0046] According to at least one embodiment, the first path and the second path are each characterized by the respective final vehicle position. The target vehicle state is characterized by the target vehicle position. For the first path and the second path, a misposition value is calculated, which corresponds to a deviation of the respective final vehicle position from the target vehicle position, in particular a distance of the respective final vehicle position from the target vehicle position. The path for use in the automatic parking of the vehicle is selected depending on the misposition values.

[0047] In other words, at least one rule depends on the misalignment values. Several variations are possible in how the rule can take the misalignment values ​​into account. In some embodiments, the rule states that the one 2024PF01883

[0048] 8

[0049] The path with the smaller misalignment value is selected from the first path and the second path. In other embodiments, the two or more terms of the respective weighted sum include the respective misalignment value.

[0050] This can increase the positioning accuracy when automatically parking the vehicle.

[0051] According to at least one embodiment, the first path and the second path are each characterized by the respective final vehicle orientation. The target vehicle state is characterized by the target vehicle orientation. For the first path and the second path, a misorientation value is calculated, corresponding to a deviation of the respective final vehicle orientation from the target vehicle orientation. The path used for automatic parking of the vehicle is selected based on the calculated misorientation values.

[0052] The deviation of the respective final vehicle orientation from the vehicle target orientation corresponds in particular to an angle between the two corresponding vectors.

[0053] In other words, at least one rule depends on the misorientation values. Several variations are possible for how the rule can take these misorientation values ​​into account. In some embodiments, the rule states that the path with the smaller misorientation value is selected from the first and second paths. In other embodiments, the two or more terms of the respective weighted sum include the respective misorientation value, for example, the respective misorientation value and the respective mispositioning value.

[0054] This can increase the orientation accuracy when automatically parking the vehicle.

[0055] According to at least one embodiment, the path for use in automatic parking of the motor vehicle is selected depending on the path length of the first path and the path length of the second path. 2024PF01883

[0056] 9

[0057] In other words, at least one rule depends on the path lengths. Several variations are possible for how the rule can take path lengths into account. In some embodiments, the rule states that the path with the shorter path length is selected from the first and second paths. In other embodiments, the two or more terms of the respective weighted sum include the respective path length. Additionally, the two or more terms of the respective weighted sum can, for example, include the respective misorientation value and / or the respective mispositioning value.

[0058] The path length can correspond to the total geometric length of the respective path.

[0059] In particular, this allows paths with shorter lengths to be favored. This increases the efficiency of path planning and can also improve user comfort. Furthermore, the accuracy and / or reliability of path planning can also be increased, since errors, variations in environmental perception, noise, and so on have less of an impact on shorter paths.

[0060] According to at least one embodiment, the first path and the second path are each characterized by a respective total number of path segments. The path used for automatic parking of the motor vehicle is selected depending on the total number of path segments of the first path and the total number of path segments of the second path.

[0061] In other words, at least one rule depends on the total number of path segments. Several variations are possible for how the rule can take the total number of path segments into account. In some embodiments, the rule states that the path with the smaller total number of path segments is selected from the first and second paths. In other embodiments, the two or more terms of the respective weighted sum include the respective total number of path segments.

[0062] Additionally, the two or more terms of the respective weighted sum can, for example, include the respective misorientation value and / or the respective mispositioning value and / or the respective path length.

[0063] A path segment can, for example, correspond to a section of the respective path along which the motor vehicle is intended to travel at a constant speed.

[0064] 10

[0065] direction, i.e. forwards or backwards, and in particular drives with a constant steering angle.

[0066] In particular, this allows paths with fewer changes in direction or steering angle to be prioritized. This increases the efficiency of path planning and can also improve user comfort. Furthermore, the accuracy and / or reliability of path planning can also be increased.

[0067] According to at least one embodiment, the first path and the second path are each characterized by a total number of changes in direction. The path used for automatic parking of the vehicle is selected depending on the total number of changes in direction of the first path and the total number of changes in direction of the second path.

[0068] In other words, at least one rule depends on the total number of direction changes. Several variations are possible for how the rule can take the total number of direction changes into account. In some embodiments, the rule states that the path with the lower total number of direction changes is selected from the first and second paths. In other embodiments, the two or more terms of the respective weighted sum include the respective total number of direction changes. Additionally, the two or more terms of the respective weighted sum can include, for example, the respective misorientation value and / or the respective mispositioning value and / or the respective path length and / or the respective total number of path segments.

[0069] In particular, this allows paths with fewer changes in direction to be prioritized. This increases the efficiency of path planning and can also improve user comfort. Furthermore, the accuracy and / or reliability of path planning can also be increased.

[0070] According to at least one embodiment, a robustness term is calculated for the first path and the second path, which corresponds to a minimum distance of the respective path from at least one obstacle in the environment and / or an average distance of the respective path from the at least one obstacle. The path for use in automatic parking of the motor vehicle is 2024PF01883

[0071] 11

[0072] selected depending on the robustness term of the first path and the robustness term of the second path.

[0073] In other words, at least one rule depends on the robustness terms. Several variations are possible for how the rule can consider the robustness terms. In some embodiments, the rule states that the path with the larger robustness term is selected from the first and second paths. In other embodiments, the two or more terms of the respective weighted sum include the respective robustness term. Additionally, the two or more terms of the respective weighted sum can include, for example, the respective misorientation value and / or the respective mispositioning value and / or the respective path length and / or the respective total number of path segments and / or the respective total number of direction changes.

[0074] This prioritizes paths where a collision with obstacles is particularly unlikely, thereby increasing the safety of path planning.

[0075] The respective distance of at least one obstacle can be determined in particular based on the environmental data.

[0076] According to at least one embodiment, a trained machine learning model (MLM) is used to determine the first path and the second path.

[0077] Training the MLM is not necessarily part of the computer-implemented path planning method according to the invention, but it can be in some embodiments.

[0078] A trained MLM can be understood as an algorithm, particularly a computer-implemented algorithm, that can replicate concrete functions or, more broadly, functions possible through human cognitive processes. A trained MLM can also be referred to as a "trained function." A trained MLM can be implemented in software and / or hardware.

[0079] When training an MLM, parameters of the MLM are generally adjusted or updated. Training can be supervised, semi-supervised, or unsupervised. Training can also include reinforcement learning or representation learning and / or other methods.

[0080] 12

[0081] Known training methods exist. In particular, the parameters of the MLM can be iteratively adjusted over several training steps. Specifically, a predefined loss function can be minimized for training purposes. If the MLM is an artificial neural network (ANN), a backpropagation algorithm can be used, for example, to adjust the parameters.

[0082] An MLM can, in particular, include an ANN, a support vector machine, a k-means cluster algorithm, a decision tree, and so on. Specifically, an ANN can be or include a deep neural network and / or a convolutional neural network, a CNN (especially a deep CNN), a recurrent neural network, a RNN (especially a recurrent CNN), a transformer network, and / or a generative adversarial network (GAN).

[0083] According to at least one embodiment, the MLM includes an agent trained using reinforcement learning, which can also be called a path planning agent.

[0084] It should be noted that the path planning agent can be a single module trained to determine both paths, or it can have two sub-modules, which in turn are also agents trained according to reinforcement learning, with one of the sub-modules being trained to determine the first path and one of the sub-modules being trained to determine the second path.

[0085] According to a further aspect of the invention, a computer-implemented training method for training a path planning agent for the automatic parking of a motor vehicle by reinforcement learning based on a plurality of training scenarios is provided. Each training scenario of the plurality of training scenarios is characterized by a training start state for the motor vehicle and a training target state for the motor vehicle. Based on at least one reinforcement learning strategy, the path planning agent determines a simulated first path for the motor vehicle from the training start state to the training target state, wherein an initial driving maneuver according to the simulated first path is a forward driving maneuver of the motor vehicle. Based on the at least one reinforcement learning strategy, the path planning agent determines a simulated second path for the motor vehicle from the training start state to the training target state.

[0086] 13

[0087] The training target state is determined, where an initial driving maneuver, corresponding to the simulated second path, is a reversing maneuver of the vehicle. At least one first value of at least one reward function for the simulated first path is calculated. At least one second value of the at least one reward function for the simulated second path is calculated. The at least one reinforcement learning strategy is updated depending on the at least one first value and the at least one second value.

[0088] Unless otherwise specified, all steps of the computer-implemented training procedure can be performed by another data processing system that includes at least one additional data processing device. In particular, the at least one additional data processing device is configured or adapted to perform the steps of the computer-implemented training procedure. For this purpose, the at least one additional data processing device can, for example, store another computer program containing further instructions which, when executed by the at least one additional data processing device, cause the at least one additional data processing device to execute the computer-implemented training procedure.

[0089] In the event that the at least one additional data processing device includes two or more further data processing devices, certain steps performed by the at least one further data processing device can be understood, for example, as different further data processing devices performing different steps or different parts of a step. In particular, it is not necessary for each further data processing device to perform the steps completely. In other words, the performance of the steps can be distributed among the two or more further data processing devices.

[0090] In reinforcement learning, the so-called environment comprises a set of states, particularly for the vehicle, and a set of possible actions, also particularly for the vehicle. The path-planning agent can interact with the environment at discrete points in time. At each point in time, one of the environment's states exists, and the path-planning agent selects one of the actions.

[0091] A reinforcement learning strategy, also called a policy, is a collection of probability measures, each of which represents a preference of the path planning agent for a particular action at a given time under a given state.

[0092] 14

[0093] The environment then outputs a subsequent state depending on the selected action, and the aforementioned steps are repeated until the entire path for a training scenario has been simulated. At least one reward function is evaluated, particularly when the entire path has been simulated. However, depending on the specific implementation of the reinforcement learning method, an evaluation of at least one reward function, or a portion thereof, can also be performed after each repetition.

[0094] The actions may include, for example, maintaining or changing the direction of travel of the motor vehicle and / or maintaining or changing the steering position or wheel position of the motor vehicle.

[0095] A reward function can be positive or negative, or it can contain positive and / or negative terms. Positive terms can be understood as positive rewards or rewards in the narrower sense, whereas negative terms can be understood as negative rewards or punishments.

[0096] By using a large number of training scenarios, the at least one reward function can be optimized by varying the at least one reinforcement learning strategy across the multitude of training scenarios, so that the trained path planning agent, which acts according to the trained at least one reinforcement learning strategy, can then be used for productive path planning, in particular in a computer-implemented method according to the invention for path planning for the automatic parking of a motor vehicle. The path planning agent generates, in particular, an action that corresponds to a part of one of the paths for the motor vehicle. The path planning agent can then generate another action that corresponds to another part of the path, and so on, until the vehicle target position and vehicle target orientation are reached. The resulting path corresponds, in particular, to a sequence of all parts generated in this way.Known methods of reinforcement learning can be used to optimize or update at least one reinforcement learning strategy depending on the value of at least one reward function.

[0097] In some embodiments, a training scenario may include additional content beyond that mentioned, in particular the positions, orientations, shapes, and / or dimensions of obstacles, other vehicles, walls, curbs, and so on. A training scenario may, in particular, include the position, orientation, shape, and / or extent of a motor vehicle.

[0098] 15

[0099] Represent or model a perceptible environment, in particular one perceptible by means of at least one environmental sensor system.

[0100] Since the path planning agent is to be trained for the automatic parking of the vehicle, the training scenarios are parking scenarios. These are characterized in particular by the fact that they each specify a training start position, a training start orientation, a training target position, and a training target orientation for the vehicle, as well as the positions and, if applicable, shapes of obstacles and / or a drivable area for the vehicle. The various training scenarios differ primarily in one or more of these aspects. Therefore, a large number of training scenarios can be generated and used without significant computational effort, thus achieving high reliability and accuracy in path planning by the trained path planning agent.In practice, the scenario can vary over time, for example due to errors in environmental perception and / or odometry. Because the path planning agent was exposed to a large number of training scenarios during its training, it can still reliably handle the current scenario.

[0101] According to at least one embodiment, the at least one reinforcement learning strategy consists of a joint reinforcement learning strategy, and the at least one reward function consists of a joint reward function. The at least one first value of the joint reward function consists of a first value of the joint reward function, and the at least one second value of the joint reward function consists of a second value of the joint reward function. A module of the path planning agent determines the simulated first path for the vehicle and the simulated second path based on the joint reinforcement learning strategy. The joint reinforcement learning strategy is updated depending on the first and second values.

[0102] According to at least one embodiment, the at least one reinforcement learning strategy consists of a first reinforcement learning strategy and a second reinforcement learning strategy, and the at least one reward function consists of a first reward function and a second reward function. The at least one first value of the at least one reward function consists of a first value of the 2024PF01883

[0103] 16

[0104] The first reward function and at least one second value of the at least one reward function consist of a second value of the second reward function. A first submodule of the path planning agent determines the simulated first path based on the first reinforcement learning strategy, and a second submodule of the path planning agent determines the simulated second path based on the second reinforcement learning strategy. The first reinforcement learning strategy is updated depending on the first value, and the second reinforcement learning strategy is updated depending on the second value.

[0105] According to at least one embodiment, the simulated first path and the simulated second path are each characterized by a respective simulated final vehicle position for the motor vehicle and / or a respective simulated final vehicle orientation for the motor vehicle.

[0106] According to at least one embodiment, the training target state is characterized by a training target position and / or a training target orientation.

[0107] According to at least one embodiment, the at least one reward function depends on a deviation of the simulated final vehicle position of the simulated first path from the training target position and on a deviation of the simulated final vehicle position of the simulated second path from the training target position.

[0108] Accordingly, simulated paths that end as close as possible to the respective training target position are preferred. This further increases the accuracy of path planning.

[0109] For example, the joint reward function depends on the deviation of the simulated final vehicle position of the simulated first path from the training target position and on the deviation of the simulated final vehicle position of the second path from the training target position. In other embodiments, the first reward function depends on the deviation of the simulated final vehicle position of the simulated first path from the training target position, and the second reward function depends on the deviation of the simulated final vehicle position of the simulated second path from the training target position. 2024PF01883

[0110] 17

[0111] According to at least one embodiment, the at least one reward function depends on a deviation of the simulated final vehicle orientation of the simulated first path from the training goal orientation and on a deviation of the simulated final vehicle orientation of the simulated second path from the training goal orientation.

[0112] Accordingly, simulated paths that end as close as possible to the respective training goal are preferred. This further increases the accuracy of path planning.

[0113] For example, the joint reward function depends on the deviation of the simulated final vehicle orientation of the simulated first path from the training goal orientation and on the deviation of the simulated final vehicle orientation of the simulated second path from the training goal orientation. In other embodiments, the first reward function depends on the deviation of the simulated final vehicle orientation of the simulated first path from the training goal orientation, and the second reward function depends on the deviation of the simulated final vehicle orientation of the simulated second path from the training goal orientation.

[0114] According to at least one embodiment, the at least one reward function depends on a path length of the simulated first path and a path length of the simulated second path.

[0115] This increases the efficiency of path planning and the resulting paths for parking the vehicle.

[0116] For example, the joint reward function depends on the path length of the simulated first path and on the path length of the simulated second path. In other embodiments, the first reward function depends on the path length of the simulated first path and the second reward function depends on the path length of the simulated second path.

[0117] According to at least one embodiment, the simulated first path and the simulated second path are each characterized by a respective total number of path segments. The at least one reward function depends on the total number of path segments.

[0118] 18

[0119] from the path segments of the simulated first path and the total number of path segments of the simulated second path.

[0120] This increases the efficiency of path planning and the resulting paths for parking the vehicle.

[0121] For example, the joint reward function depends on the total number of path segments of the simulated first path and on the total number of path segments of the simulated second path. In other embodiments, the first reward function depends on the total number of path segments of the simulated first path, and the second reward function depends on the total number of path segments of the simulated second path.

[0122] According to at least one embodiment, the simulated first path and the simulated second path are each characterized by a total number of changes in direction. The at least one reward function depends on the total number of changes in direction of the simulated first path and the total number of changes in direction of the simulated second path.

[0123] This increases the efficiency of path planning and the resulting paths for parking the vehicle.

[0124] For example, the joint reward function depends on the total number of direction changes of the simulated first path and on the total number of direction changes of the simulated second path. In other embodiments, the first reward function depends on the direction changes of path segments of the simulated first path, and the second reward function depends on the total number of direction changes of the simulated second path.

[0125] According to at least one embodiment, each training scenario of the plurality of training scenarios is characterized by at least one obstacle in the environment. For the simulated first path and the simulated second path, a robustness term is calculated, which corresponds to a minimum distance of the respective simulated path from the at least one obstacle and / or an average distance of the respective simulated path from the at least one obstacle, and the 2024PF01883

[0126] 19

[0127] at least one reward function depends on the robustness term of the simulated first path and the robustness term of the simulated second path.

[0128] This prioritizes simulated paths where a collision with obstacles is particularly unlikely, further increasing the safety of path planning.

[0129] For example, the training scenarios can each be characterized by the position, size, and / or orientation of at least one obstacle. Alternatively or additionally, the training scenarios can each be characterized by an area accessible to the vehicle. The minimum distance and / or the average distance can be determined based on one or more of these pieces of information.

[0130] For example, the joint reward function depends on the minimum distance of the first simulated path from the at least one obstacle and / or an average distance of the first simulated path from the at least one obstacle, and the joint reward function depends on the minimum distance of the second simulated path from the at least one obstacle and / or an average distance of the second simulated path from the at least one obstacle.In other embodiments, the first reward function depends on the minimum distance of the first simulated path from the at least one obstacle and / or an average distance of the first simulated path from the at least one obstacle, and the second reward function depends on the minimum distance of the second simulated path from the at least one obstacle and / or an average distance of the second simulated path from the at least one obstacle.

[0131] Further embodiments of the computer-implemented training method according to the invention follow directly from the various configurations of the computer-implemented method according to the invention, and vice versa. In particular, individual features and corresponding explanations as well as advantages relating to the various configurations of the computer-implemented method according to the invention can be transferred analogously to corresponding configurations of the computer-implemented training method according to the invention. 2024PF01883

[0132] 20

[0133] According to at least one embodiment of the computer-implemented method for path planning for the automatic parking of a motor vehicle, the first path and the second path are determined by a path planning agent trained by means of a computer-implemented training method according to the invention.

[0134] In some such embodiments, the computer-implemented method includes the execution of the computer-implemented training procedure.

[0135] According to a further aspect of the invention, a method for automatically parking a motor vehicle is described. In this method, a computer-implemented path planning method according to the invention is carried out for automatically parking a motor vehicle, and the motor vehicle is parked automatically by being guided along the selected path.

[0136] For example, at least one control signal is generated for automatically guiding the vehicle along the selected path. To automatically guide the vehicle along the selected path, at least one actuator of the vehicle is controlled depending on this at least one control signal.

[0137] The at least one actuator can, for example, include one or more brake actuators and / or one or more steering actuators and / or one or more drive motors of the vehicle. Based on the at least one control signal, the one or more actuators can influence the longitudinal and / or lateral steering of the vehicle to automatically guide it along the selected path.

[0138] According to another aspect of the invention, a data processing system is specified which is configured to carry out a computer-implemented method for path planning according to the invention.

[0139] The terms "data processing system" and "at least one data processing device" may be used interchangeably within the scope of this disclosure. For example, in this disclosure, a data processing device may be understood as a device with processing circuits for processing data. A data processing device can thus perform arithmetic operations to process data. An indexed access to a data structure, such as a lookup table (LUT) or a database, may also be considered an arithmetic operation.2024PF01883

[0140] 21

[0141] Data processing that is partially or fully implemented in hardware can also be considered a computational operation.

[0142] A data processing device may, in particular, comprise one or more computers, one or more microcontrollers, and / or one or more integrated circuits, for example, one or more application-specific integrated circuits (ASICs), one or more field-programmable gate arrays (FPGAs), and / or one or more systems-on-a-chip (SoCs). A data processing device may also comprise one or more processors, for example, one or more microprocessors, one or more central processing units (CPUs), one or more graphics processing units (GPUs), and / or one or more signal processors, in particular one or more digital signal processors (DSPs). The data processing device may also comprise a physical or virtual cluster of computers or other devices of the aforementioned type.

[0143] A data processing device may also include one or more hardware and / or software interfaces, for example for receiving and / or providing data.

[0144] A data processing device may also include one or more storage devices. A storage device may be implemented as volatile memory, such as dynamic random access memory (DRAM) or static random access memory (SRAM), or as non-volatile data storage, such as read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or flash EEPROM, ferroelectric random access memory (FRAM), magnetoresistive random access memory (MRAM), or phase-change random access memory (PCRAM).

[0145] According to a further aspect of the invention, a further data processing system is specified which is configured to carry out a computer-implemented training method according to the invention for planning a path to train a path planning agent.

[0146] According to another aspect of the invention, an electronic vehicle guidance system for a motor vehicle is provided. The electronic vehicle guidance system includes a data processing system configured to implement a computer-implemented path planning method according to the invention.

[0147] 22

[0148] to carry out the process, as well as a control system designed to automatically guide the vehicle along the selected path. In particular, the data processing system stores the trained path planning agent for this purpose.

[0149] The control system can, for example, be part of the data processing system or an additional data processing system.

[0150] An electronic vehicle control system (EVS) can be understood as an electronic system designed to control a vehicle fully automatically or autonomously, in particular without requiring any intervention from a driver. The vehicle automatically performs all necessary functions, such as steering, braking, and / or acceleration maneuvers, monitoring and recording road traffic, and reacting accordingly. Specifically, the EVS can implement a fully automatic or fully autonomous driving mode of the vehicle according to Level 5 of the SAE J3016 classification. An EVS can also be understood as an advanced driver assistance system (ADAS), which supports the driver during partially automated or semi-autonomous driving.In particular, the electronic vehicle guidance system can implement a partially automated or semi-autonomous driving mode according to levels 1 to 4 of the SAE J3016 classification. Here and in the following, "SAE J3016" refers to the corresponding standard in the April 2021 version.

[0151] At least partially automated vehicle control can therefore include driving the vehicle in accordance with a fully automated or fully autonomous driving mode of Level 5 according to SAE J3016. At least partially automated vehicle control can also include driving the vehicle in accordance with a partially automated or semi-autonomous driving mode according to Levels 1 to 4 of SAE J3016.

[0152] According to at least one embodiment, the electronic vehicle guidance system has at least one environmental sensor system configured to generate environmental data.

[0153] Further embodiments of the electronic vehicle guidance system according to the invention follow directly from the various configurations of the computer-implemented method according to the invention, and vice versa. 2024PF01883

[0154] 23

[0155] In particular, individual features and corresponding explanations as well as advantages relating to the various embodiments of the method according to the invention can be transferred analogously to corresponding embodiments of the electronic vehicle guidance system according to the invention. In particular, the electronic vehicle guidance system according to the invention is designed or programmed to carry out a computer-implemented method according to the invention. In particular, the electronic vehicle guidance system according to the invention carries out the computer-implemented method according to the invention.

[0156] According to another aspect of the invention, a first computer program with first instructions is specified. When the first instructions are executed by a data processing system, the first instructions cause the data processing system to carry out a computer-implemented method according to the invention.

[0157] The initial instructions can be provided as program code. This program code can be, for example, binary code or assembly language, and / or source code in a programming language such as C, and / or a program script such as Python.

[0158] According to a further aspect of the invention, a second computer program with second instructions is provided. When the second instructions are executed by a data processing system, they cause the data processing system to perform a computer-implemented training method according to the invention.

[0159] The second set of instructions can be provided as program code. This program code can be, for example, binary code or assembly language, and / or source code in a programming language such as C, and / or a program script such as Python.

[0160] According to a further aspect of the invention, a third computer program with third commands is specified. When the third commands are executed by an electronic vehicle guidance system according to the invention, in particular by the data processing system of the electronic vehicle guidance system, the third commands cause the electronic vehicle guidance system to

[0161] 24

[0162] to carry out a method according to the invention for automatically parking a vehicle.

[0163] The third set of instructions can be provided as program code. This program code can be, for example, binary code or assembly language, and / or source code in a programming language such as C, and / or a program script such as Python.

[0164] According to a further aspect of the invention, a computer-readable storage medium is provided which stores a first computer program according to the invention and / or a second computer program according to the invention and / or a third computer program according to the invention.

[0165] The first computer program, the second computer program, the third computer program, and the computer-readable storage medium are each computer program products containing the first instructions and / or the second instructions and / or the third instructions.

[0166] Further features of the invention are evident from the claims, the figures, and the description of the figures. The features and combinations of features mentioned above in the description, as well as those mentioned below in the description of the figures and / or illustrated in the figures, may be encompassed by the invention not only in the combinations specified, but also in other combinations. In particular, embodiments and combinations of features that do not include all the features of an originally formulated claim may also be encompassed by the invention. Furthermore, embodiments and combinations of features that go beyond or deviate from the combinations of features mentioned in the claims may also include the invention.

[0167] The invention is explained in more detail below with reference to specific exemplary embodiments and corresponding schematic drawings. Identical or functionally equivalent elements in the drawings may be provided with the same reference numerals. The description of identical or functionally equivalent elements is not necessarily repeated with respect to the different figures. 2024PF01883

[0168] 25

[0169] The figures show

[0170] Fig. 1 shows a schematic representation of a motor vehicle with an exemplary embodiment of an electronic vehicle guidance system according to the invention;

[0171] Fig. 2 shows a schematic representation of a motor vehicle during automatic parking according to an exemplary embodiment of a method according to the invention for automatically parking the motor vehicle;

[0172] Fig. 3 shows a schematic representation of a motor vehicle during automatic parking according to a further exemplary embodiment of a method according to the invention for automatically parking the motor vehicle; and

[0173] Fig. 4 shows a schematic representation of an exemplary embodiment of a computer-implemented training method for training a path planning agent to automatically park a motor vehicle by reinforcement learning.

[0174] Fig. 1 shows a schematic representation of a motor vehicle 1 with an exemplary embodiment of an electronic vehicle guidance system 3 according to the invention. The vehicle guidance system 3 has a data processing system 2 which is configured to carry out a computer-implemented method according to the invention for path planning for the automatic parking of the motor vehicle 1.

[0175] The motor vehicle 1, for example the electronic vehicle guidance system 3, has at least one environmental sensor system 4, for example one or more cameras, radar systems, lidar systems and / or ultrasonic sensor systems. The at least one environmental sensor system 4 is configured to generate environmental data representing the environment of the motor vehicle 1 and to provide it to the data processing system 2. Based on the environmental data, the data processing system 2 determines a target vehicle state for the motor vehicle 1. The data processing system 2 determines a first path for the motor vehicle 1 from a current vehicle state of the motor vehicle 1 to the target vehicle state, wherein an initial driving maneuver according to the first path is a forward driving maneuver of the motor vehicle 1, and a second path for the motor vehicle 1 from the current

[0176] 26

[0177] The vehicle state of motor vehicle 1 is compared to the target vehicle state, where an initial driving maneuver according to the second path is a reversing maneuver of motor vehicle 1. Based on at least one predefined rule, the data processing system 2 selects a path for use in the automatic parking of motor vehicle 1 from the first path and the second path.

[0178] The data processing system 2 is configured to automatically guide the motor vehicle 1 along the selected path. For this purpose, the data processing system 2 can, for example, generate at least one control signal and control at least one actuator of the motor vehicle 1, in particular at least one brake actuator and / or one or more steering actuators and / or one or more drive motors of the motor vehicle 1, depending on the at least one control signal.

[0179] The first path and the second path are each characterized, for example, by a respective final vehicle position and / or a respective final vehicle orientation for vehicle 1. The vehicle target state is characterized, for example, by a vehicle target position and / or a vehicle target orientation for vehicle 1.

[0180] Figure 2 schematically depicts the motor vehicle 1 as it is guided along the first path according to an exemplary embodiment of the method. SO corresponds to the current vehicle state. Starting from this, the motor vehicle is moved forward to state S1, then backward to state S2, and finally forward to a final vehicle state corresponding to the final vehicle position and orientation of the first path.

[0181] Figure 3 schematically depicts the motor vehicle 1 as it is guided along the second path according to an exemplary embodiment of the method. SO corresponds to the current vehicle state. From there, the motor vehicle is moved backwards to state ST and from there finally forwards to a final vehicle state corresponding to the final vehicle position and orientation of the second path.

[0182] The first path according to Fig. 2 therefore involves a total of two changes of direction for the vehicle 1, while the second path according to Fig. 3 involves only one change of direction. However, the first path may be safer or more robust because 2024PF01883

[0183] 27

[0184] The second path takes the motor vehicle 1 less close to obstacles 6a, 6b, 6c than the second path.

[0185] Depending on the specifications, the data processing system 2 can therefore select the first path for automatically parking the motor vehicle 1 if safety is to be maximized, or the second path if the number of changes of direction is to be minimized.

[0186] Depending on the use case, other and / or more complex rules can be applied to select the path. For example, a misposition value, corresponding to the deviation of the respective final vehicle position from the target vehicle position, and / or a misorientation value, corresponding to the deviation of the respective final vehicle orientation from the target vehicle orientation, can be calculated for both the first and second paths. The path for use in the automatic parking of vehicle 1 can then be selected, for example, based on the misposition values ​​and / or the misorientation values. Alternatively or additionally, the path can be selected based on the path length of the first path and the path length of the second path.

[0187] Fig. 4 schematically shows an exemplary embodiment of a computer-implemented training method for training a path planning agent 7 for the automatic parking of a motor vehicle 1 by reinforcement learning. The trained path planning agent 7 can then be used, for example, in a method as described with respect to Figs. 1 to 3, in particular for determining the first path and the second path.

[0188] A variety of training scenarios 9, 10, 11 are provided for training the path planning agent 7. These can be created, in particular, through simulation. Each training scenario 9, 10, 11 is characterized by at least one training start state for the vehicle 1 and one training target state for the vehicle 1. The training target state is characterized, in particular, by a training target position and a training target orientation, and the training start state, in particular, by a training start position and a training start orientation. Furthermore, the training scenarios 9, 10, 11 can be distinguished, for example, by different positions and / or shapes of obstacles 6a, 6b, 6c and, if applicable, by a 2024PF01883

[0189] 28

[0190] different numbers of obstacles 6a, 6b, 6c are distinguished and accordingly by different drivable areas for the motor vehicle 1.

[0191] For each training scenario 9, 10, 11, the path planning agent 7 determines a simulated first path for the vehicle 1 from the training start state to the training target state based on at least one reinforcement learning strategy, wherein an initial driving maneuver according to the simulated first path is a forward driving maneuver of the vehicle 1, and a simulated second path for the vehicle 1 from the training start state to the training target state, wherein an initial driving maneuver according to the simulated second path is a reverse driving maneuver of the vehicle 1.

[0192] At least one first value of at least one reward function for the simulated first path and at least one second value of at least one reward function for the simulated second path are calculated. The at least one reinforcement learning strategy is updated depending on the at least one first value and the at least one second value.

[0193] At least one reward function depends, for example, on a deviation of the simulated final vehicle position of the simulated first path from the training target position and on a deviation of the simulated final vehicle position of the simulated second path from the training target position, and / or on a deviation of the simulated final vehicle orientation of the simulated first path from the training target orientation and on a deviation of the simulated final vehicle orientation of the simulated second path from the

[0194] Training goal orientation.

[0195] The at least one reward function can also depend on a path length of the simulated first path and a path length of the simulated second path. The at least one reward function can also depend on a total number of path segments of the simulated first path and the total number of path segments of the simulated second path. The at least one reward function can also depend on a total number of changes of direction of travel of the simulated first path and the total number of changes of direction of travel of the simulated second path. The at least one reward function can also depend on a robustness term of the simulated first path and a robustness term of the simulated second path, where the respective 2024PF01883

[0196] 29

[0197] Robustness term corresponds to a minimum distance of the respective simulated path from the at least one obstacle 6a, 6b, 6c and / or an average distance of the respective simulated path from the at least one obstacle 6a, 6b, 6c.

[0198] Apart from the aspects described, seven well-known reinforcement learning techniques can be used to train the path planning agent, particularly using a so-called environment 8. The environment 8 can be considered a simulator that generates the various training scenarios 9, 10, and 11. The path planning agent 7 evaluates a current situation, which is represented as state s. t As described, an action leads to a t It exits and receives feedback from the environment after each action, for example. 8. The feedback is used as a reward. t This refers to something that can be positive or negative.

[0199] In some embodiments, the difficulty of the training scenarios 9, 10, 11 may increase during the training, the difficulty being, for example, due to the width of a parking space 5 and / or the positions and / or sizes of the obstacles 6a, 6b, 6c and / or corresponding tolerances to be observed.

[0200] In training scenario 9, for example, parking lot 5 is particularly wide, and there are only two obstacles, 6a and 6b, located far apart. Therefore, it is relatively easy for the path planning agent 7 to simulate a path to parking lot 5. However, training scenario 10 is already more demanding and thus presents a higher level of difficulty. The number of obstacles, 6a, 6b, and 6c, is increased by three or four, and parking lot 5 is narrower because the obstacles 6a, 6b, and 6c are closer together compared to training scenario 18. In training scenario 11, the difficulty is increased even further because the obstacles 6a, 6b, and 6c are wider and even closer together, making parking lot 5 even narrower.

[0201] An output from the trained path planning agent 7 can, for example, be converted into a steering angle profile 12 along the respective simulated path. The steering angle profile 12 is shown as a curve in Fig. 4. The horizontal axis shows a distance d from the current vehicle position, and the vertical axis shows a steering angle α of a steering system of the vehicle 1. The steering angle profile 12 here comprises, by way of example, three segments 13, 14, 15, each of which can be interpreted as the vehicle 1 traveling a certain distance with a constant or changing steering angle α. 2024PF01883

[0202] 30

[0203] The steering angle profile 12 can, for example, be provided to the data processing system 2 of the electronic vehicle guidance system 3, so that it can generate at least one control signal as described.

[0204] As described, the invention can increase the accuracy of path planning and / or user comfort in the automatic parking of a motor vehicle.

Claims

2024PF01883 31 Patent claims 1. Computer-implemented method for path planning for automatic parking of a motor vehicle (1), wherein Environmental data representing the environment of the motor vehicle (1) are received; a vehicle target state for the motor vehicle (1) is determined based on the environmental data; a first path for the motor vehicle (1) is determined from a current vehicle state of the motor vehicle (1) to the vehicle target state, wherein an initial driving maneuver according to the first path is a forward driving maneuver of the motor vehicle (1); a second path for the motor vehicle (1) is determined from the current vehicle state of the motor vehicle (1) to the vehicle target state, wherein an initial driving maneuver according to the second path is a reversing driving maneuver of the motor vehicle (1); based on at least one predefined rule, a path for use in the automatic parking of the motor vehicle (1) is selected from the first path and the second path.

2. Computer-implemented method according to claim 1, wherein the first path and the second path are each characterized by a respective final vehicle position for the motor vehicle (1); The vehicle target state is characterized by a vehicle target position; a mispositioning value is calculated for the first path and the second path, which corresponds to a deviation of the respective final vehicle position from the vehicle target position; The path for use in automatic parking of the motor vehicle (1) is selected depending on the mispositioning values. 2024PF01883 32 3. Computer-implemented method according to any one of the preceding claims, wherein the first path and the second path are each characterized by a respective final vehicle orientation for the motor vehicle (1); The vehicle target state is characterized by a vehicle target orientation; a misorientation value is calculated for the first path and the second path, which corresponds to a deviation of the respective final vehicle orientation from the vehicle target orientation; the path for use in automatic parking of the motor vehicle (1) is selected depending on the calculated misorientation values.

4. Computer-implemented method according to one of the preceding claims, wherein the path for use in automatic parking of the motor vehicle is selected depending on a path length of the first path and a path length of the second path.

5. Computer-implemented method according to any one of the preceding claims, wherein the first path and the second path are each characterized by a respective total number of path segments, and the path for use in automatic parking of the motor vehicle is selected depending on the total number of path segments of the first path and the total number of path segments of the second path; and / or the first path and the second path are each characterized by a respective total number of changes in direction, and the path for use in automatic parking of the motor vehicle is selected depending on the total number of changes in direction of the first path and the total number of changes in direction of the second path.

6. Computer-implemented method according to any one of the preceding claims, wherein For the first path and the second path, a robustness term is calculated, which corresponds to a minimum distance of the respective path of at least ei-2024PF01883 33 an obstacle (6a, 6b, 6c) in the vicinity and / or an average distance of the respective path from which at least one obstacle (6a, 6b, 6c) corresponds; and The path for use in automatic parking of the motor vehicle is selected depending on the robustness term of the first path and the robustness term of the second path.

7. Computer-implemented training method for training a path planning agent (7) for automatically parking a motor vehicle (1) by reinforcement learning based on a variety of training scenarios (9, 10, 11), wherein each training scenario (9, 10, 11) is characterized by a training start state for the motor vehicle (1) and a training target state for the motor vehicle (1); by the path planning agent (7) based on at least one reinforcement learning strategy, a simulated first path for the motor vehicle (1) from the training start state to the training target state is determined, wherein an initial driving maneuver according to the simulated first path is a forward driving maneuver of the motor vehicle (1); by the path planning agent (7) based on at least one reinforcement learning strategy, a simulated second path for the motor vehicle (1) from the training start state to the training target state is determined, wherein an initial driving maneuver according to the simulated second path is a reversing driving maneuver of the motor vehicle (1); at least one first value of at least one reward function for the simulated first path is calculated; at least a second value of at least one reward function is calculated for the simulated second path; and which at least one reinforcement learning strategy is updated depending on at least one first value and at least one second value.

8. Computer-implemented training method according to claim 7, wherein the simulated first path and the simulated second path are each characterized by a respective simulated final vehicle position for the motor vehicle (1); the training target state is characterized by a training target position; 2024PF01883 34 which at least one reward function depends on a deviation of the simulated final vehicle position of the simulated first path from the training target position and on a deviation of the simulated final vehicle position of the simulated second path from the training target position.

9. Computer-implemented training method according to one of claims 7 or 8, wherein the simulated first path and the simulated second path are each characterized by a respective simulated final vehicle orientation for the motor vehicle (1); the training target state is characterized by a training target orientation; which depends on at least one reward function on a deviation of the simulated final vehicle orientation of the simulated first path from the training target orientation and on a deviation of the simulated final vehicle orientation of the simulated second path from the training target orientation.

10. Computer-implemented training method according to any one of claims 7 to 9, wherein which at least one reward function depends on a path length of the simulated first path and a path length of the simulated second path; and / or the simulated first path and the simulated second path are each characterized by a respective total number of path segments and at least one reward function depends on the total number of path segments of the simulated first path and the total number of path segments of the simulated second path; and / or the simulated first path and the simulated second path are each characterized by a respective total number of changes in direction, and at least one reward function depends on the total number of changes in direction of the simulated first path and the total number of changes in direction of the simulated second path; and / or Each training scenario (9, 10, 11) is characterized by at least one obstacle (6a, 6b, 6c) in the environment, and a robustness term is calculated for each of the simulated first path and the simulated second path, which corresponds to a minimum distance of the respective simulated path from the at least one obstacle. 35 dernis (6a, 6b, 6c) and / or an average distance of the respective simulated path from the at least one obstacle (6a, 6b, 6c) and the at least one reward function depends on the robustness term of the simulated first path and the robustness term of the simulated second path.

11. Computer-implemented method according to any one of claims 1 to 6, wherein the first path and the second path are determined by a path planning agent (7) trained by means of a computer-implemented training method according to any one of claims 7 to 11.

12. Method for automatically parking a motor vehicle (1), wherein a computer-implemented method according to one of claims 1 to 6 or 11 is carried out and the motor vehicle (1) is automatically guided along the selected path.

13. Method according to claim 12, wherein at least one control signal is generated for the automatic guidance of the motor vehicle (1) along the selected path; and for automatically guiding the motor vehicle (1) along the selected path, at least one actuator of the motor vehicle (1) is controlled depending on which at least one control signal is applied.

14. Data processing system (2) configured to perform a computer-implemented method according to any one of claims 1 to 6 or 11 and / or a computer-implemented training method according to any one of claims 7 to 10.

15. Electronic vehicle guidance system (3) for a motor vehicle (1) comprising a data processing system (2) configured to perform a computer-implemented method according to any one of claims 1 to 6 or 11, and a control system configured to automatically guide the motor vehicle (1) along the selected path. 2024PF01883 36 16. Electronic vehicle guidance system (3) according to claim 15, wherein the electronic vehicle guidance system (3) comprises at least one environment sensor system (4) configured to generate the environment data.

17. computer program product first commands which, when executed by a data processing system (2), cause the data processing system to perform a computer-implemented method according to one of claims 1 to 6 or 11; and / or second commands which, when executed by a data processing system, cause the data processing system to perform a computer-implemented training method according to any one of claims 7 to 10; and / or third commands which, when executed by an electronic vehicle guidance system (3) according to claim 15, cause the electronic vehicle guidance system (3) to perform a method according to any one of claims 12 or 13.

Images