Method for determining a starting pose of a vehicle

Abstract

A method for determining a starting pose of a vehicle by a controller is disclosed, wherein measurement data determined by a GNSS sensor system and / or an odometry sensor system is received and analyzed for determining an approximate pose of the vehicle with a range of uncertainty, at least one trajectory of a traffic participant is extracted from a trajectory map for the determined range of uncertainty, check points are arranged along the extracted trajectory, and an optimization algorithm is implemented for each check point along the trajectories, by which an pose with a corresponding cost function is determined, a pose with the greatest cost function is determined from the poses determined by the optimization algorithm as the starting pose of the vehicle. Furthermore, a controller, a computer program and a machine-readable storage medium are disclosed.

Description

TECHNICAL FIELD

[0001] The present invention relates to a method for determining a starting pose of a vehicle. Furthermore, the present invention relates to a controller, a computer program and a machine-readable storage medium. BACKGROUND

[0002] The use of high-precision digital maps is necessary for implementing automated driving functions. Such maps enable a sensor-based environmental perception of an autonomously operable vehicle. A wide variety of digital maps are known, for example containing a geographical road course and typical trajectories of traffic participants. Typical trajectories can here be extracted from so-called swarm trajectories of a large number of vehicles.

[0003] In order to utilize digital maps, a vehicle must be able to determine its own pose, which consists of the position and orientation of the vehicle, precisely. The pose of a vehicle can be determined, for example, by analyzing GNSS signals from a corresponding satellite. However, GNSS-based positioning has insufficient availability and accuracy.

[0004] Iterative positioning methods are generally used for determining the pose, which require an accurate initial position or starting position. For determining the starting position or starting pose, the vehicle can be approximately positioned and the pose can then be refined by feature-based positioning, for example by wheel measurement data. In feature-based positioning, static features are extracted from radar measurement data and compared with features saved in a feature map. For this purpose, for example, a multidimensional hill climbing algorithm is used in order to obtain an optimal orientation of the features. However, the orientation of the features in multiple dimensions requires high computing power. SUMMARY

[0005] The task on which the present invention is based can be seen as presenting a method for determining a starting pose of a vehicle, which has a reduced requirement on computing power.

[0006] The task is solved by means of a method for determining a starting pose of a vehicle by a controller according to the invention, a controller according to the invention, a computer program product according to the invention and a machine-readable storage medium according to the invention. Advantageous configurations of the present invention are given below.

[0007] According to one aspect of the present invention, a method for determining an initial pose or starting pose of a vehicle by a controller is provided.

[0008] The method can in particular be implemented in the form of hardware and / or software by an initialization module of the controller.

[0009] In a step, measurement data determined by a GNSS sensor system and / or an odometry sensor system is received and analyzed for determining an approximate pose of the vehicle with an uncertainty range.

[0010] From the trajectory map, at least one trajectory of a traffic participant is extracted for the determined uncertainty range. The at least one extracted trajectory represents, for example, a typical trajectory of the traffic participant from historical data.

[0011] In a further step, check points are arranged along the extracted trajectories and an optimization algorithm is implemented for each check point along the trajectories. By implementing the optimization algorithm, a feature-based localization or a refinement of the approximate pose can be performed. For this purpose, it is possible to consider measurement data of a radar sensor system, a lidar sensor system and / or a camera sensor system and to compare these measurement data with data of a digital feature map. Here, the optimization algorithm can be performed continuously along the trajectories starting from each check point or can be implemented discretely for each specified check point.

[0012] Next, poses with a corresponding cost function are determined by the optimization algorithm. These poses can represent, for example, local maxima of the cost function.

[0013] In a further step, the pose with the maximum cost function is determined from the poses determined by the optimization algorithm as the starting pose of the vehicle.

[0014] By this method, the required computing power for determining the starting pose can be reduced, since the optimization algorithm can optimize the cost function with a smaller number of necessary function calls. In particular, based on the information from the trajectory map, a two-dimensional optimization problem is simplified to a one-dimensional optimization problem along the typically extracted trajectories of the trajectory map.

[0015] The optimization algorithm is used, in particular, for comparing static features extracted from measurement data of an environmental sensor system with features from a feature map and thereby determining an exact starting pose of the vehicle. For this purpose, measurement data of an environmental sensor system, for example a radar sensor system, a lidar sensor system and / or a camera sensor system, can be received and analyzed in advance in order to extract static features from the vehicle's environment. Based on the section of the trajectory from the trajectory map received within the uncertainty range of the approximately determined pose, these static features can be compared with the features saved in the feature map linearly or one-dimensionally along the trajectory. By this measure, the starting pose can be determined faster and with low computing power requirements.

[0016] According to another aspect of the application a controller is provided, wherein the controller is configured to implement the method. The controller can be for example a controller on the vehicle side, a controller outside the vehicle or a server unit outside the vehicle, for example a cloud system.

[0017] Restricting the optimization problem to the predefined trajectories is not limited here to the determination of the initial pose. Rather, this principle can also be applied to other methods, for example a lane keeping assistant.

[0018] In particular, the controller can have a localization module and / or an initialization module. The controller is thereby able to implement the method for determining the initial pose of the vehicle and / or the method for performing the localization.

[0019] Furthermore, according to an aspect of the application a computer program is provided comprising instructions which, when the computer program is executed by a computer or a controller, cause the computer or the controller to carry out the method according to the application. According to another aspect of the application a machine-readable storage medium is provided, on which the computer program according to the application is stored.

[0020] The vehicle can be assisted, partially automated, highly automated and / or fully automated or driverless according to the BASt (Bundesanstalt für Straβenwesen, German Federal Institute for the

[0021] The vehicle can be configured for example as a passenger car, a truck, a robot taxi or the like. The vehicle is not limited to operation on a road. Rather, the vehicle can also be configured as a waterway means of transport, an airway means of transport, for example a transport drone or the like.

[0022] In an embodiment, the orientation of the vehicle is determined from the determined approximate pose of the vehicle and compared to the driving direction of the extracted trajectories, wherein the filter function is executed to consider only trajectories having a driving direction, wherein trajectories are considered which are in line with the orientation of the vehicle.

[0023] By this measure all sections of trajectories within the uncertainty range which are not in line with the driving direction of the vehicle can be removed. Thereby the optimization algorithm can be additionally accelerated, since for example a comparison with features along trajectories of counter traffic is avoided.

[0024] According to a further embodiment, a hill climbing algorithm is implemented as optimization algorithm. Thereby, the comparison of the static features extracted from the measurement data of the environmental sensor system with the features from the feature map can be realized technically particularly simply and quickly. The cost function can preferably be greatest in the case of greatest agreement of the static features with the features from the feature map, whereby the possible objective of the optimization algorithm is to find at least one maximum of the cost function.

[0025] According to a further embodiment, the optimization algorithm is implemented along the at least one extracted trajectory within the uncertainty range determined. By restricting the trajectory to the uncertainty range of the determined pose, the length of the trajectory segment can be further reduced.

[0026] By restricting the length of the trajectory segment, the search range or orientation range of the optimization algorithm can be further reduced and the method accelerated.

[0027] According to a further embodiment, the method for determining the starting pose of the vehicle is repeatedly executed at defined time intervals. By this measure, the starting pose can be determined at regular time intervals and, for example, taken into account for a plausibility check or a parallel positioning function.

[0028] According to a further embodiment, a plurality of starting poses of the vehicle are determined at different points in time, wherein a lane deviation or a lane change of the vehicle is determined by detecting an inconsistency between the determined starting poses. If, for example, one of the plurality of determined poses deviates from the trajectory and / or from a previous pose, a difference or inconsistency can be determined.

[0029] In the case of a determined inconsistency, a lane change of the vehicle can be considered as a trigger for the inconsistency. By this measure, a lane change or an unintentional lane deviation can be detected.

[0030] Alternatively or additionally, the inconsistency can be determined by a deviation of the determined pose from the extracted trajectory. Here, a permissible deviation can be provided, which must be exceeded in order to determine the inconsistency.

[0031] According to a further embodiment, the measurement data determined by at least one lidar sensor and / or radar sensor are received and analyzed. Subsequently, features are determined from these measurement data and compared with the data of the feature map in order to determine the pose of the vehicle along the extracted trajectory in the feature map.

[0032] A maximum cost function for the pose of the vehicle is calculated by an optimization algorithm based on the minimum deviation of the features retrieved and the features saved in the feature map. In particular, one of the multiple potential poses can be selected which has the highest or best cost function and thus the best feature consistency.

[0033] By this measure an approximate or rough range can be determined which is further refined by the feature-based localization. Thereby, the initial pose can be implemented faster and with less computational operations, since the range considered for the feature-based localization is limited by the uncertainty range of the approximately retrieved pose of the vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0034] The preferred embodiments of the present application are explained in more detail below on the basis of highly simplified schematic drawings. Herein are shown:

[0035] Figure 1 : a schematic view of a driving road with an uncertainty range of an approximate pose and multiple points for implementing an optimization algorithm; and

[0036] Figure 2 : a schematic view of a driving road with an uncertainty range of a pose and multiple points for running an optimization algorithm along a trajectory. DETAILED DESCRIPTION

[0037] Figure 1 and Figure 2 A schematic diagram for illustrating a method for retrieving an initial pose A of a vehicle 2 is shown, which can be used for a further not illustrated localization of the vehicle 2. The initial pose A can be retrieved for example by a vehicle-side controller 4 implementing the method.

[0038] Figure 1 A schematic view of a driving road 1 is shown with an uncertainty range U of an approximate pose P and multiple check points 5 for implementing an optimization algorithm. The driving road 1 is driven through by the vehicle 2 in a driving direction F.

[0039] The vehicle 2 has an odometry sensor system and / or a GNSS sensor system 6 and an additional sensor system 8 for feature-based localization. The additional sensor system 8 can be for example a lidar sensor system, a radar sensor system and / or a camera sensor system.

[0040] In the shown embodiment, measurement data is collected in driving by the odometry sensor system and the GNSS sensor system 6. The approximate pose P is determined from the measurement data of the odometry sensor system and the GNSS sensor system 6. Since this approximate pose P is erroneous, an uncertainty range U of the approximate pose P is shown schematically.

[0041] The check points 5 can be distributed over the entire uncertainty range U. Starting from the positions of the check points 5, the optimization algorithm can be implemented. However, in order to speed up the optimization algorithm, trajectories 10 are extracted from the trajectory map, which trajectories extend through the uncertainty range U.

[0042] Next, the check points 5 are arranged along the extracted trajectories 10, so that the optimization algorithm is implemented for each check point 5 and along the trajectories 10. Thereby, the computational overhead of the optimization algorithm can be reduced from a two-dimensional problem to a one-dimensional problem. This step is illustrated in Figure 2 .

[0043] Figure 2 A schematic diagram of a driving road 1 is shown, with an uncertainty range U of a pose and a plurality of check points 5 for implementing the optimization algorithm along the trajectories 10.

[0044] Based on the historical data of the trajectory map, the probability of the vehicle 2 following one of the extracted trajectories 10 is the maximum probability. The optimization algorithm is therefore implemented along the extracted trajectories 10.

[0045] In implementing the optimization algorithm, it is preferable to receive and analyze the measurement data taken by the environmental sensor system 8.

[0046] Static features are taken from the measurement data and compared with the data of the feature map. The optimization algorithm is used, in particular, to compare the static features with the features of the feature map and to maximize the cost function.

[0047] When the extracted static features best match the features of the feature map, Figure 1 the cost function becomes maximum.

[0048] The pose of the vehicle 2 with the maximum cost function from the plurality of possible poses of the vehicle 2 is taken as the starting pose A and used for further localization of the vehicle 2.

Claims

1. Method for determining a starting pose (A) of a vehicle (2) by a controller (4), wherein - receiving measurement data determined by a GNSS sensor system and / or an odometry sensor system (6) and analyzing the measurement data for determining an approximate pose (P) of the vehicle (2) with an uncertainty range (U), - extracting at least one trajectory (10) of a traffic participant from a trajectory map for the determined uncertainty range (U), - arranging check points (5) along the extracted trajectories (10) and applying an optimization algorithm to each check point (5) along the trajectories (10), - determining a pose with a corresponding cost function by the optimization algorithm, - determining a pose with the largest cost function from the poses determined by the optimization algorithm as the starting pose (A) of the vehicle (2), - wherein measurement data determined by at least one lidar sensor and / or radar sensor (8) is received and analyzed, wherein features are determined from the measurement data and compared to data of a feature map in order to determine a pose of the vehicle (2) along at least one extracted trajectory (10) in the feature map, and wherein the largest cost function of a pose of the vehicle (2) is calculated by the optimization algorithm based on the smallest deviation of the determined features from features saved in the feature map.

2. The method of claim 1, wherein, - determining an orientation of the vehicle (2) from the determined approximate pose (P) of the vehicle (2) and comparing the orientation to a driving direction (F) of the extracted trajectories, wherein a filter function is executed for considering only trajectories (10) with a driving direction (F), wherein at least one trajectory (10) coinciding with the orientation of the vehicle (2) is considered.

3. The method of claim 1 or 2, wherein, - executing a hill climbing algorithm as the optimization algorithm.

4. The method of claim 1 or 2, wherein, - executing the optimization algorithm along at least one extracted trajectory (10) within the determined uncertainty range (U).

5. The method of claim 1 or 2, wherein, - repeating the method for determining a starting pose (A) of the vehicle (2) at defined time intervals.

6. The method of claim 5, wherein, - determining a plurality of starting poses (A) of the vehicle (2) at different points in time, wherein a lane deviation or a lane change of the vehicle (2) is decided by detecting inconsistencies between the determined starting poses (A).

7. A controller (4), wherein - the controller (4) is configured to execute the method according to any one of claims 1 to 6.

8. Computer program product, the computer program comprising instructions which, when the computer program is executed by a computer or a controller, cause the computer to carry out the method according to any one of claims 1 to 6.

9. Machine-readable storage medium on which instructions are stored which, when the computer program is executed by a computer or a controller, cause the computer to carry out the method according to any one of claims 1 to 6.

Images