Method for generating a training dataset, training dataset, method for training a map generation module, and map generation module

Abstract

The invention relates to a computer-implemented method (100) for generating a training data set (301) for training a map generation module (303), comprising: Determining (105) a position value (315) of the environment sensor data (305), wherein the position value (315) of the environment sensor data (305) is defined by a pose of the moving unit (309); Determining (107) a different position value (317) of the environment sensor data (305), wherein the different position value (317) differs from the position value (315) of the environment sensor data (305) by a deviation value (319); Determining (109) a map section (321) of the electronic map (313) based on the divergent position value (317), wherein the map section (321) is arranged around the divergent position value (317) of the environmental sensor data (305); Grouping the environmental sensor data (305) and the determined map section (321) into a data set unit (323); and integrating (111) the data set unit (323) into the training data set (301). The invention further relates to a method (200) for training a map generation module (303).

Description

[0001] The present invention relates to a method for generating a training data set for training a map generation module to generate a map representation. The invention further relates to a corresponding method for training a map generation module, a map generation module, and a training data set. State of the art

[0002] Precise maps are essential for autonomous vehicle driving. Methods for generating such maps and using AI-based map generation modules are known in the prior art. Furthermore, methods for training such map generation modules are also known.

[0003] It is an object of the present invention to provide an improved method for generating a training data set for training a map generation module, an improved method for training a map generation module, an improved training data set and an improved map generation module.

[0004] The problem is solved by the methods, the training dataset, and the map generation module of the independent claims. Advantageous embodiments are the subject of the dependent claims.

[0005] According to one aspect, a computer-implemented method for generating a training dataset for training a map generation module is provided, comprising: Receiving environmental sensor data from an environmental sensor of a moving unit, wherein the environmental sensor data at least partially represent the environment of the moving unit; Receiving map data from an electronic map, wherein the electronic map at least partially represents the environment of the moving unit and / or includes information relating to the environment; Determining a position value from the environmental sensor data, wherein the position value of the environmental sensor data is defined by a pose of the moving unit; determining a deviating position value based on the position value, wherein the deviating position value differs from the position value of the environmental sensor data by a deviation value; Determining a map section of the electronic map based on the divergent position value, wherein the map section is arranged around the divergent position value of the environmental sensor data; Grouping the environmental sensor data and the determined map section into a single data set unit; and Integrating the data set unit into a training data set.

[0006] This provides the technical advantage of an improved method for generating a training dataset for training a map generation module. To generate the training dataset, environmental sensor data from at least one sensor of a moving unit is considered, with the environmental sensor data at least partially representing the environment of the respective moving unit. Furthermore, map data from an electronic map is also considered. The electronic map extends beyond the environment of the moving unit and includes various pieces of information about the environment. First, a position value is determined from the environmental sensor data. This position value is defined by the pose of the moving unit at the time the environmental sensor data was acquired.Based on this, a different position value is determined, which deviates from the position value of the environmental sensor data by a selectable but defined deviation value.

[0007] Based on the differing position value, a map section is generated from the electronic map. This map section is defined as a spatial area arranged around the differing position value and at least partially depicts the environment of the moving unit. The determined map section thus corresponds to the representation of the electronic map of the environment of the moving unit as depicted by the environmental sensor data.

[0008] The determined map section and the associated environmental sensor data are then combined into a data set unit and integrated as a unit into the training data set.

[0009] The environmental sensor data and the corresponding map section of the data set generated according to the steps above thus have different position values. The position value of the environmental sensor data corresponds to the actual position value, i.e., the actual pose, at the time the environmental sensor data was acquired. The position value of the map section corresponds to the differing position value, which deviates from the actual position value of the environmental sensor data by the specified deviation value. This allows inaccuracies in pose determination during the acquisition of the environmental sensor data to be simulated.

[0010] The intended deviation between the two position values ​​of the environmental sensor data and the corresponding map section allows the map generation module to be trained on corresponding inaccuracies in pose determination during the acquisition of environmental sensor data by the moving units during subsequent training based on the correspondingly generated training data set.

[0011] This allows for more precise training of the map generation module, enabling the later, appropriately trained map generation module to produce more precise maps even with inaccurate pose determinations.

[0012] This allows the appropriately trained map generation module to be used even with lower-quality moving units that may have lower sensor quality and therefore less precise pose determination during sensor data acquisition. Modifying the position value of the environmental sensors by a freely selectable but defined deviation value and generating the corresponding position value allows for the simple consideration of such pose determination inaccuracies during the map generation module's training.

[0013] According to one embodiment, the differing position value is determined by shifting the position value of the environmental sensor data along a displacement axis by a displacement value, and / or wherein the differing position value is determined by rotating the position value of the environmental sensor data about a rotation axis by a rotation value.

[0014] This allows for the technical advantage of enabling a simple deviation between the position value of the environmental sensor data and the alternative position value. The deviation between the position value and the alternative position value can include a displacement along the translational axes, thus simulating a different position of the moving unit relative to an absolute reference system. Alternatively or additionally, the deviation between the position value and the alternative position value can include a rotation around a rotational axis, thus representing a different orientation of the moving unit relative to the absolute reference system.

[0015] According to one embodiment, the deviating position value is determined based on a random distribution of the position value, wherein the random distribution is limited by a predefined limit value, and wherein the predefined limit value defines a maximum permissible deviation of the deviating position value from the position value.

[0016] This offers the technical advantage of easily determining the deviating position value by considering the random distribution. The random distribution can, for example, be a normal or uniform distribution centered around the actual position value of the environmental sensor data. This allows the simulation of inaccuracies in the position determination of the moving unit during the acquisition of the environmental sensor data. A predefined threshold limit can restrict the deviation between the position value and the deviating position value, thus preventing excessively divergent position values ​​and avoiding such data that would be unusable for training the map generation module. The predefined threshold is freely selectable and can be determined based on the required training precision.

[0017] According to one embodiment, the position value of the environment sensor data is based on data from a Global Navigation Satellite System, and the differing position value is determined based on the position value taking into account an error value of the position value determined based on the data of the Global Navigation Satellite System.

[0018] This offers the technical advantage of enabling a simple determination of the deviation value between the position value from the environmental sensor data and the deviating position value. The inaccuracies or errors of the global navigation satellite system, on whose data the position values ​​of the environmental sensor data are based, are used as the deviation value. The deviating position value can, for example, be obtained by adding the specified error values ​​of the global navigation satellite system to the actual position value.

[0019] In particular, when considering environmental sensor data from a multiple of different moving units, the deviations between the position values ​​of the environmental sensor data provided by the different moving units can be taken into account as a deviation value.

[0020] The identified deviations between the position values ​​relate to environmental sensor data provided by various moving units for a comparable pose.

[0021] In one embodiment, the map section is converted into a bird's-eye view representation.

[0022] This offers the technical advantage that the bird's-eye view allows for a more precise consideration of the map section's information for map display by the map generation module.

[0023] According to one embodiment, a plurality of data record units with different deviation values ​​between the respective position value and the deviating position value of the respective environment sensor data are generated.

[0024] This allows for the technical advantage of generating an improved training dataset. This dataset can contain various data units with different deviation values ​​between the position values ​​of the respective environmental sensor data and the corresponding position values ​​of the map sections. These varying deviation values ​​allow the training of the map generation module to account for different levels of inaccuracy in position determination during the generation of the environmental sensor data. This, in turn, enables improved training of the map generation module and, consequently, improved performance of the trained module.

[0025] According to one embodiment, the environmental sensor data is based on fleet data from a plurality of moving units.

[0026] This approach offers the technical advantage of providing a large volume of environmental sensor data by considering fleet data from the majority of mobile units. Furthermore, it enables a high degree of diversity in the environmental sensor data, thus largely reducing systematic errors in data acquisition, such as those that might occur with individual mobile units. Additionally, the environmental sensor data from the fleet of mobile units allows for the inclusion of diverse spatial areas and environments, as represented by the respective sensor data, in the training dataset and, consequently, in the training of the map generation module.

[0027] According to one embodiment, the movable unit is a vehicle, and the electronic map is designed as an electronic road map of a road traffic network.

[0028] This allows for the technical advantage of providing an improved training dataset for in-vehicle navigation, taking into account information from electronic road maps. The correspondingly trained map generation module is therefore suitable for use in vehicles for navigation in road traffic.

[0029] According to one embodiment, the environmental sensor data includes data from the following list: radar data, lidar data, ultrasound data, camera data, and / or wherein the electronic map information includes information relating to elements from the list: lane marking, road center line, road signs, topological features.

[0030] This allows for the technical advantage that precise environmental sensor data and meaningful information from the electronic road map can be taken into account in the training data set.

[0031] According to one aspect, a training data set is provided for training a map generation module, wherein the training data set comprises a plurality of data set units, each containing environmental sensor data and a corresponding map section, and wherein the training data set was generated according to the method for generating a training data set according to one of the preceding embodiments.

[0032] This allows for the technical advantage of providing an improved training dataset. The training dataset comprises data units generated according to the above-described process steps, each with varying deviation values ​​between the position values ​​of the environmental sensor data and the corresponding position values ​​of the map segments. This improved training dataset can then be used to train a map generation module, enabling the module to be trained based on intentionally incorporated inaccuracies in position determination during the generation of the environmental sensor data.

[0033] Following one aspect, a procedure for training a map generation module is provided, including: Providing a training data set according to the invention; training the map generation module to generate a map representation based on the data set units of the training data set, wherein the map representation includes information from the map display and information from the environment sensor data and at least partially depicts the environment of the moving unit.

[0034] This allows for the technical advantage of improved training of the map generation module. The improved training dataset enables the module to account for inaccuracies in pose determination during the acquisition of environmental sensor data by moving units. The corresponding map generation module can therefore also be used for moving units with correspondingly inaccurate pose determination.

[0035] According to one embodiment, the training of the map generation module comprises a first training phase and a subsequent second training phase, wherein in the first training phase the map generation module is trained on data record units of the training data set, each of which has a deviation value between the position value and the deviating position value that is less than or equal to a predefined limit, and wherein in the second training phase the map generation module is trained on data record units of the training data set, each of which has a deviation value between the position value and the deviating position value that is greater than the predefined limit.

[0036] This allows for the technical advantage of improved training of the map generation module. In an initial training phase, the map generation module is trained on data sets where the deviation between the position value of the environmental sensor data and the corresponding position value of the respective map segment is less than or equal to a predefined threshold. For example, the map generation module can initially be trained on data sets that consider only the actual position values ​​of the environmental sensor data without any corresponding deviation value.

[0037] After completion of the first training phase, in which the map generation module was trained based on the environmental sensor data and map sections with actual position values, the second training phase takes into account data set units in which there are deviation values ​​between the position values ​​of the environmental sensor data and the deviating position values ​​of the associated map sections.

[0038] By initially training the map generation module on unaltered data with original position values, and only after this pre-training considering data sets with substantial deviations between the position values ​​of the environmental sensor data and the differing position values ​​of the map segments, more precise training of the map generation module can be achieved. The appropriately trained map generation module is then able to perform map generation on both environmental sensor data with precise position determination and on environmental sensor data with inaccurate position determination.

[0039] According to one embodiment, in the second training phase a proportion of the data set units used for training with a deviation value greater than the predefined limit is gradually increased, and / or wherein, in the second training phase, data set units with gradually increasing deviation values ​​between the position value and the deviating position value are used for training.

[0040] This allows for the technical advantage of further improving the training of the map generation module. In the second training phase, as the training duration increases, the proportion of data sets with a substantial deviation between the position value of the environmental sensor data and the corresponding position value of the map segments is increased. By gradually increasing the proportion of data sets with a corresponding deviation value, the trained map generation module can be progressively brought closer to the environmental sensor data with its inaccurate position determination. Alternatively or additionally, in the second training phase, as the training duration increases, the deviation values ​​between the position values ​​of the environmental sensor data and the corresponding position values ​​of the map segments can be increased.For this purpose, data sets with increasingly larger deviation values ​​are used for training as the training duration increases. This allows the correspondingly trained map generation module to be gradually introduced to environmental sensor data with progressively greater inaccuracies in pose determination.

[0041] According to one aspect, a map generation module is provided, wherein the map generation module has been trained according to the method for training a map generation module according to one of the preceding embodiments, and wherein the map generation module is configured to generate a map representation of the environment of the moving unit based on environmental sensor data and an electronic map.

[0042] This allows for the technical advantage of providing an improved map generation module. This improved module is designed to generate maps based on environmental sensor data and map sections from an electronic road map, and is capable of performing this map generation based on environmental sensor data even with inaccurate position determination of the respective moving units.

[0043] According to one aspect, a computing unit is provided which is set up to execute the method for generating a training data set for training a map generation module according to one of the preceding embodiments and / or the method for training a map generation module according to one of the preceding embodiments.

[0044] According to one aspect, a computer program product is comprehensively provided with instructions which, when the program is executed by a data processing unit, cause it to execute the method for generating a training data set for training a map generation module according to one of the preceding embodiments and / or the method for training a map generation module according to one of the preceding embodiments.

[0045] Embodiments of the invention are described with reference to the following figures. The figures show: Fig. 1 a schematic representation of a method for generating a training data set for training a map generation module according to one embodiment; Fig. 2 a schematic representation of a method for training a map generation module according to one embodiment; Fig. 3 a flowchart of the procedure for generating a training data set for training a map generation module according to an embodiment; Fig. 4 a flowchart of the procedure for training a map generation module according to one embodiment; and Fig. 5. A schematic representation of a computer program product.

[0046] Fig. Figure 1 shows a schematic representation of a method 100 for generating a training data set 301 for training a map generation module 303 according to an embodiment.

[0047] In the embodiment shown, the system 300 comprises a computing unit 325 on which a data set generation module 327 is executed. The data set generation module 327 is configured to execute the inventive method 100 for generating a training data set 301 for training a map generation module 303.

[0048] To generate the training data set 301, the data set generation module 327 first receives environmental sensor data 305 from at least one environmental sensor 307 of at least one moving unit 309. The environmental sensor data 305 represent the environment of the respective moving unit 309.

[0049] In the embodiment shown, the environmental sensor data 305 provided originate from a plurality of different moving units 309.

[0050] In the embodiment shown, the movable units 309 are designed as vehicles. The environmental sensor data 305 can be, for example, radar data, lidar data, ultrasound data or camera data.

[0051] The environmental sensor data 305 each have a position value 315, which is defined by a pose of the respective moving unit 309 providing the environmental sensor data 305.

[0052] In the embodiment shown, the respective position value 315 is based on data from a global navigation satellite system 333.

[0053] In addition to the environmental sensor data 305, the data set generation module 327 also receives map data 311 from an electronic map 313. In the embodiment shown, the electronic map 313 is designed as an electronic road map and shows the course of several roads 331 of a road traffic network.

[0054] To generate the training data set 301, the data set generation module 327 first determines a position value 315 for each of the environmental sensor data 305. The position value 315 is defined by a pose of the respective moving unit 309 at the time the environmental sensor data 305 is acquired. In the embodiment shown, the position value 315 can, for example, be provided by the global navigation satellite system 333. The position value 315 could, for example, be a corresponding GPS value.

[0055] Based on the determined position value 315, a different position value 317 is subsequently generated by the data record generation module 327. This different position value 317 deviates from position value 315 by a selectable but defined deviation value 319.

[0056] The deviating position value 317 can be generated by shifting the position value 315 by a displacement value along a displacement axis. Alternatively or additionally, the deviating position value 317 can be generated by rotating the position value 315 about a rotation axis. The shift changes the positioning of the pose of the respective movable unit 309, while the rotation changes the orientation of the pose relative to a fixed reference frame.

[0057] Alternatively or additionally, the deviating position value 317 can be calculated from position value 315 using a random distribution. This random distribution can be a normal or uniform distribution with position value 315 at its center. A maximum deviation value 319 between the deviating position value 317 and position value 315 can be defined using a predefined limit, which could, for example, be defined by a width value of the random distribution.

[0058] Alternatively or additionally, the differing position value 317 can be determined based on errors or discrepancies between the data provided by the global navigation satellite system 333 from the multiple moving units 309. The corresponding inaccuracies or errors are added to or subtracted from the actual position values ​​315 according to a predefined rule.

[0059] Based on the determined deviation value 317, the data set generation module 327 subsequently determines a map section 321 based on the electronic map 313. Map section 321 is defined by a spatial area of ​​the electronic map 313 arranged around the deviation value 317. Map section 321 thus represents at least part of the environment of the respective moving unit 309. However, map section 321 is shifted or rotated relative to the environment sensor data 305 by the deviation value 319. The environment sensor data 305 of a moving unit 309, each centered around the position value 315, and the map section 321 associated with the environment sensor data 305 represent the same environment of the respective moving unit 309, but are shifted and / or rotated relative to each other by the deviation value 319.

[0060] This ensures that the environmental sensor data 305 with the position value 315 and the corresponding map section 321, which is aligned around the differing position value 317 determined based on position value 315, exhibit an inaccuracy relative to each other, represented by the deviation value 319. This allows for the simulation of an inaccuracy in the position determination by the respective moving unit 309 during the generation of the environmental sensor data 305. A corresponding inaccuracy in the position determination, in turn, leads to a deviation of the corresponding environmental sensor data 305 from the respective map section 321. Such inaccuracies in position determination are to be expected particularly with moving units 309 equipped with low-quality environmental sensors 307.

[0061] After the creation of map section 321, map section 321 can be displayed in a bird's-eye view.

[0062] Subsequently, the data set generation module 327 combines the environmental sensor data 305 and the respective corresponding map section 321 into a data set unit 323. The respective data set unit 323 is then integrated into the training data set 301.

[0063] Each data record unit 323 thus comprises a set of environmental sensor data 305, each assigned to the original position value 315, and a corresponding map section 321, which is assigned to the deviating position value 317 determined on the basis of the original position value 315.

[0064] To generate the training data set 301, a large number of such data set units 323 are generated. These data set units 323 can be generated with different deviation values ​​319 between the position value 315 of the respective environmental sensor data 305 and the differing position value 317 of the corresponding map section 321. The magnitude of each deviation value 319 is freely selectable and can be chosen with regard to the desired performance of the correspondingly trained map generation module 303.

[0065] Fig. Figure 2 shows a schematic representation of a method 200 for training a map generation module 303 according to one embodiment.

[0066] In the embodiment shown, for the training of the map generation module 303, a [various parameters] are first [various parameters] according to the [specific parameters]. Fig. The training dataset 301 generated by the aforementioned procedure steps is provided. Based on the dataset units 323 of the training dataset 301, the map generation module 303 to be trained is subsequently trained to generate a corresponding map representation 329 of the environment based on environmental sensor data 305 and map sections 321, each of which at least partially depicts the same environment of a corresponding moving unit 309. The map representation 329 includes information from the environmental sensor data 305, including information from the map section 321. The information from the map section 321 can, for example, include information regarding the following elements: lane markings, center line, road signs, traffic rules, and topological features of the road network, such as bus stops or parking spaces.

[0067] The training of the map generation module 303 can be performed according to training methods known from the prior art, for example, supervised or unsupervised. The map generation module 303 can be trained as a corresponding artificial intelligence.

[0068] In the embodiment shown, the training of the map generation module 303 comprises a first training phase P1 and a later second training phase P2.

[0069] In the first training phase P1, the training of the map generation module 303 is performed on data record units 323 whose deviation values ​​319 between the position values ​​315 of the environmental sensor data 305 and the differing position value 317 of the corresponding map section 321 reaches or falls below a predefined limit. In particular, the training of the map generation module 303 in the first training phase P1 can be performed primarily on data record units 323 in which both the environmental sensor data 305 and the map section 321 are arranged around the same original position value 315, so that the deviation value is zero.

[0070] In the later second training phase P2, the training of the map generation module 303 is then carried out on data record units 323, whose deviation values ​​319 between the position value 315 of the environment sensor data 305 and the deviating position value 317 of the respective map section 321 exceed the predefined limit value.

[0071] The map generation module 303 can therefore initially be trained in the first training phase P1 on the unaltered data with original position values ​​315. This allows the precision of the generation of the map display 329 to be achieved.

[0072] In the second training phase P2, the map generation module 303, which is pre-trained on the unadulterated data set units 323 and already functions with acceptable performance, can then be trained taking into account the inaccuracies of the position determination in the form of the data set units 323 with substantial deviation values ​​319 between the position value 315 of the environment sensor data 305 and the differing position value 317 of the respective associated map section 321.

[0073] According to one embodiment, the proportion of data record units 323 with deviation value 319 of the data record units 323 used for training can be gradually increased in the second training phase P2 as training progresses.

[0074] Alternatively or additionally, with ongoing training progress in the second training phase P2, data set units 323 with gradually increasing deviation values ​​319 can be taken into account.

[0075] This allows the map generation module 303, which is already pre-trained for the unadulterated data, to be gradually brought into consideration of inaccuracies in the position determination when generating environmental sensor data 305 by the respective moving units 309, which in the present procedure is taken into account by the deviation values ​​319 between the position values ​​315 of the environmental sensor data 305 and the differing position values ​​317 of the respective map sections 321 of the data record units 323.

[0076] By dividing the training into the first and second training phases P1 and P2, a precise consideration of the inaccuracies in pose determination can be achieved when generating the map display 329. The correspondingly trained map generation module 303 is thus able to generate map displays 329 with high precision based on environmental sensor data 305 and corresponding map sections 321, both for environmental sensor data 305 with high accuracy in pose determination and for environmental sensor data 305 with reduced accuracy in pose determination.

[0077] The appropriately trained map generation module 303 is specifically configured to perform online map generation. In operation, the trained map generation module 303 is specifically configured to take into account, while the vehicle is driving, the currently generated environmental sensor data 305 from the vehicle's environmental sensor 307 and the map data 311 from a pre-stored electronic road map 313, in order to generate map representations 329 based on this data that depict the current environment of the moving vehicle and take into account both the information from the environmental sensor data 305 and the information from the electronic road map 313, which is used as the map priority in the described procedure.

[0078] Fig. Figure 3 shows a flowchart of the method 100 for generating a training data set 301 for training a map generation module 303 according to an embodiment.

[0079] To generate the training data set, in a first process step 101, environmental sensor data 305 from the environmental sensor 307 of the moving unit 309 are received. The environmental sensor data 305 at least partially represent the environment of the moving unit 309.

[0080] In a further process step 103, the map data 311 of the electronic map 313 are received. The electronic map 313 at least partially depicts the environment of the mobile unit 309 or includes information regarding the respective environment.

[0081] In a further process step 105, a position value 315 of the environmental sensor data 305 is determined. The position value 315 is defined here by the pose of the respective moving unit 309 at the time the environmental sensor data 305 is recorded.

[0082] In a further process step 107, a different position value 317 is determined based on the position value 315 of the environmental sensor data 305. The different position value 317 deviates from the original position value 315 by a freely selectable but defined deviation value 319.

[0083] In a further process step 109, a map section 321 is determined based on the electronic map 313. The map section 321 is defined by a defined spatial area of ​​the electronic map 313 arranged around the differing position value 317.

[0084] In a further process step 111, the environmental sensor data 305 with the position value 315 and the map section 321 with the differing position value 317 are grouped into a data set unit 323.

[0085] In a further process step 113, the correspondingly generated data set unit 323 is integrated into the training data set 301.

[0086] Integrating data set unit 323 into training data set 301 can also include the respective data set unit 323 representing the first data set unit 323 of training data set 301.

[0087] Fig. Figure 4 shows a flowchart of the procedure 200 for training a map generation module 303 according to one embodiment.

[0088] To train the map generation module 303, a training data set 301, which was generated according to the above described procedure steps of procedure 100, is first provided in a first procedure step 201.

[0089] In a further process step 203, the corresponding map generation module 303 is trained to generate a map representation based on the training data set 301. The corresponding map representation 329 includes information from the electronic map 313 and information from the environmental sensor data 305.

[0090] The map generation module 303 is particularly capable of performing online map creation. During vehicle operation, the information from the environmental sensor data 305 and the information from the electronic offline road map are processed in real time, and a corresponding map display 329 is generated.

[0091] The electronic map 313 is in particular an offline map, while the map display 329 is an online map and represents the current state of the environment of the moving unit 309, i.e. the vehicle.

[0092] Fig.Figure 5 shows a schematic representation of a computer program product 400, comprising instructions which, when the program is executed by a data processing unit, cause it to execute the procedure 100 for generating a training data set 301 for training a map generation module 303 and / or the procedure 200 for training a map generation module 303.

[0093] In the embodiment shown, the computer program product 400 is stored on a storage medium 401. The storage medium 401 can be any storage medium known from the prior art.

Claims

[1] Computer-implemented method (100) for generating a training data set (301) for training a map generation module (303), comprising: Receiving (101) environmental sensor data (305) from an environmental sensor (307) of a moving unit (309), wherein the environmental sensor data (305) at least partially represent the environment of the moving unit (309); Receiving (103) map data (311) from an electronic map (313), wherein the electronic map (313) represents at least part of the environment of the mobile unit (309) and / or includes information relating to the environment; Determining (105) a position value (315) of the environment sensor data (305), wherein the position value (315) of the environment sensor data (305) is defined by a pose of the moving unit (309); Determining (107) a different position value (317) based on the position value (315), wherein the different position value (317) differs from the position value (315) of the environment sensor data (305) by a deviation value (319); Determining (109) a map section (321) of the electronic map (313) based on the divergent position value (317), wherein the map section (321) is arranged around the divergent position value (317) of the environmental sensor data (305); Grouping (111) the environmental sensor data (305) and the determined map section (321) into a data set unit (323); and integrating (113) the data set unit (323) into the training data set (301). [2] Method (100) according to claim 1, wherein the deviating position value (317) is determined by shifting the position value (315) of the environment sensor data (305) along a displacement axis by a displacement value, and / or wherein the deviating position value (317) is determined by rotating the position value (315) of the environment sensor data (305) about a rotation axis by a rotation value. [3] Method (100) according to one of the preceding claims, wherein the deviating position value (317) is determined based on a random distribution of the position value (315), wherein the random distribution is limited by a predefined limit value, and wherein the predefined limit value defines a maximum permissible deviation of the deviating position value (317) from the position value (315). [4] Method (100) according to one of the preceding claims, wherein the position value (315) of the environment sensor data (305) is based on data from a Global Navigation Satellite System, and wherein the differing position value (317) is determined based on the position value (315) taking into account an error value of the position value (317) determined based on the data from the Global Navigation Satellite System. [5] Method (100) according to one of the preceding claims, wherein the map section (321) is converted into a bird's-eye view representation. [6] Method (100) according to one of the preceding claims, wherein a plurality of data record units (323) with different deviation values ​​(319) between the respective position value (315) and the deviating position value (317) of the respective environment sensor data (305) are generated. [7] Method (100) according to one of the preceding claims, wherein the environmental sensor data (305) are based on fleet data of a plurality of mobile units (309). [8] Method (100) according to one of the preceding claims, wherein the movable unit (309) is a vehicle, and wherein the electronic map (313) is designed as an electronic road map of a road traffic network. [9] Method (100) according to any of the preceding claims, wherein the environmental sensor data (305) comprise data from the following list: Radar data, lidar data, ultrasound data, camera data, and / or wherein the electronic map information (313) includes information relating to elements from the list: lane marking, center line, road signs, topological features. [10] Training data set (301) for training a map generation module (303), wherein the training data set (301) comprises a plurality of data set units (323) each containing environmental sensor data (305) and a corresponding map section (321), and wherein the training data set (301) was generated according to the method for generating a training data set (301) according to any one of the preceding claims 1 to 10. [11] Method (200) for training a map generation module (303), comprising: Providing (201) a training data set (301) according to claim 10; training (203) the map generation module (303) to generate a map representation (329) based on the data set units (323) of the training data set (301), wherein the map representation (329) comprises information from the map representation (329) and information from the environment sensor data (305) and at least partially maps the environment of the moving unit (309). [12] Method (200) according to claim 11, wherein the training of the map generation module (303) comprises a first training phase (P1) and a subsequent second training phase (P2), wherein in the first training phase (P1) the map generation module (303) is trained on data record units (323) of the training data set (301), each of which has a deviation value (319) between the position value (315) and the deviating position value (317) that is less than or equal to a predefined limit, and wherein in the second training phase (P2) the map generation module (303) is trained on data record units (323) of the training data set (301), each of which has a deviation value (319) between the position value (315) and the deviating position value (317) that is greater than the predefined limit. [13] Method (200) according to claim 12, wherein in the second training phase (P2) a proportion of the data set units (323) used for training with a deviation value (319) greater than the predefined limit value is gradually increased, and / or wherein in the second training phase (P2) data set units (323) with gradually increasing deviation values ​​(319) between the position value (315) and the deviating position value (317) are used for training. [14] Map generation module (303), wherein the map generation module (303) was trained according to the method for training a map generation module (303) according to any one of the preceding claims 11 to 13, and wherein the map generation module (303) is configured to generate a map representation (329) of the environment of the moving unit (309) based on environment sensor data (305) and an electronic map (313). [15] Computing unit (325) configured to perform the method (100) for generating a training data set (301) for training a map generation module (303) according to any one of claims 1 to 9 above and / or the method for training a map generation module (303) according to any one of claims 11 to 13 above. [16] Computer program product (400) comprising instructions which, when the program is executed by a data processing unit, cause the data processing unit to execute the method (100) for generating a training data set (301) for training a map generation module (303) according to any one of claims 1 to 9 above and / or the method for training a map generation module (303) according to any one of claims 11 to 13 above.

Images