DE102021214763A1 - Method and control device for controlling an automated vehicle - Google Patents

Abstract

A method for controlling an automated vehicle (100) is described, in which static and dynamic measurement data (510, 520) are provided by at least one on-board environment sensor (111, 112, 113), the static measurement data (510) being associated with static objects (310 ) are assigned in the area (300) of the automated vehicle (100), while the dynamic measurement data (520) are assigned to dynamic objects (320) in the area (300) of the automated vehicle (100). A localization (631) is then carried out, in which the features of static objects (310) in the environment (300) of the automated vehicle (100) determined using the static measurement data (510) are compared with features from a provided digital localization map (212), to determine a global pose of the automated vehicle (100). Object tracking (632) is also carried out, in which dynamic objects (320) in the environment (300) of the automated vehicle (100) are determined with their relative positions using the dynamic measurement data (520). The results (530) of the localization (631) and the results (540) of the object tracking (632) are then determined in a joint evaluation process (630) using both the static and the dynamic measurement data (510, 520). Finally, the determined results (530) of the localization (631) and/or the results (540) of the object tracking (632) are used to plan the behavior and/or the movement of the automated vehicle (100).

Description

The invention relates to a method for controlling an automated vehicle. The invention also relates to a control device for carrying out the method.

Current driver assistance systems (ADAS, Advanced Driver Assistance Systems) and highly automated vehicle systems for autonomous driving (AD) require detailed knowledge of the vehicle's surroundings and the current traffic situation. In order to be able to plan and carry out suitable driving maneuvers with sufficient accuracy, one's own position or pose as well as the relative positions and possibly also the speeds of neighboring road users must be known. The localization of the vehicle can be map-based, for which purpose an environment model, which is generated by detecting the vehicle environment using environment sensors and then extracting certain information about the environment, is compared with a digital localization map. A radar feature map, for example, which is based on static radar measurements of vehicles in a vehicle fleet (crowd-sourced radar feature map), can serve as the digital localization map. A position of the ego vehicle determined in this way within a navigation map is also used in navigation devices to create a route and to navigate the vehicle along the route.

In addition, some radar-based ADAS applications (e.g. adaptive cruise control (ACC), traffic jam pilot (TJP), highway pilot hands-free, etc.) use dynamic object tracking for detection dynamic objects (e.g. surrounding vehicles or pedestrians) in the vehicle environment. In contrast to the radar-based localization function, the radar-based ADAS applications mainly use dynamic radar measurements to determine the relative positions of the surrounding dynamic objects.

The localization and dynamic object tracking thus represent two separate components within the ADAS/AD architecture. While static radar measurements and the radar feature map are used to determine the vehicle position (own position), the dynamic radar measurements that take place independently of this are used to determine the positions and, if necessary, speeds of neighboring vehicles dynamic objects relative to the ego vehicle. Then the information provided by the localization component and the information provided by the dynamic object tracking component are combined as input for behavior generation and vehicle motion control. Since both components (localization and dynamic object tracking) have independent estimation errors, these estimation errors accumulate in behavior generation and vehicle motion control, which is accompanied by greater uncertainty in behavior generation and vehicle motion control. Furthermore, the fact that both components (localization and dynamic object tracking) do not use all of the measurement data from the radar sensor, but only part of it, leads to additional uncertainty in the information provided by the two components. This error also affects applications that only use information from one of the two components (localization or dynamic object tracking).

The object on which the invention is based can therefore be seen as providing a possibility for improving the accuracy of localization and/or dynamic object tracking. This object is solved by means of the respective subject matter of the independent claims. Advantageous configurations of the invention are the subject matter of the dependent subclaims.

According to the invention, a method for controlling an automated vehicle is provided in which static and dynamic measurement data are provided by at least one on-board environment sensor, the static measurement data being associated with static objects in the environment of the automated vehicle, while the dynamic measurement data are associated with dynamic objects in the environment of the automated vehicle are assigned. Localization is then carried out, in which features of static objects in the area surrounding the automated vehicle, determined using the static measurement data, are compared with features from a provided digital localization map in order to determine the global pose of the automated vehicle. Dynamic object tracking is also carried out, in which dynamic objects in the area surrounding the automated vehicle are determined with their relative positions using the dynamic measurement data. The results of the localization and the results of the dynamic object tracking are determined in a joint evaluation process using both the static and the dynamic measurement data. Finally, the results of the localization and/or the results of the dynamic object tracking are used to plan the behavior and/or the movement of the automated driving stuff used. The joint processing of the static and dynamic measurement data can improve the accuracy of the localization and dynamic object tracking results achieved thereby. Among other things, various errors and disturbances associated with the respective measurement method can be compensated for by jointly processing and optimizing the measurement data. For example, systematic measurement errors of the respective sensor can be better recognized and reduced by using the HD localization map as a reference. Furthermore, when using static and dynamic measurement data, more measurements can be available overall, and an improvement in the measurement accuracy can already be achieved from a statistical point of view. Those functions of the automated vehicle that only use the results of the localization or only the results of the dynamic object tracking also benefit from this improvement in accuracy. By using the measurement data of both static and dynamic objects in the vehicle environment, both localization and dynamic object tracking can be improved. For applications that do not require a localization mode localization map. Dynamic object tracking can be improved by taking static radar measurements and map information into account. On the other hand, the accuracy of the localization can also be improved by taking into account dynamic objects and possibly also the unmapped static objects, such as parked vehicles. An extended object tracking including classification can thus support the localization. The combined solution of the two problems (localization and dynamic object tracking) not only leads to an improvement in the accuracy of the results obtained therefrom (vehicle pose and relative position of surrounding dynamic objects), but is also more efficient than solving the two problems separately.

One embodiment provides for the results of the localization and dynamic object tracking to be determined using a graph-based optimization method in which boundary conditions derived from the static measurement data and boundary conditions derived from the dynamic measurement data are added to form a common optimization problem. Such a graph-based optimization procedure is a very effective method for determining both the ego pose using dynamic measurement data and an HD localization map, and the relative position of neighboring dynamic objects.

A further embodiment provides that the results of the localization and dynamic object tracking are determined using a filter-based state estimation method. The filter state of a filter used in the state estimation method includes both the unknown states of the localization and the unknown states of the dynamic object tracking. Filter-based state estimation methods are also particularly effective methods for determining unknown states of the ego vehicle and neighboring dynamic objects.

A further embodiment provides that the filter-based optimization process uses a Kalman filter or a particle filter. These are particularly effective filter-based methods with which a large number of measurement data can be processed effectively and precisely.

A further embodiment provides that, in addition to the static and dynamic objects, unmapped static objects are also used for the localization and/or dynamic object tracking. The unmapped static objects, which can be parked vehicles, for example, are also suitable for increasing the number of individual measurements for the statistical methods used. As a result, this procedure can lead to an increase in accuracy when estimating or optimizing the ego pose or the position of neighboring vehicles.

A further embodiment provides that only the results of the localization or only the results of the dynamic object tracking are used to plan the behavior and/or the movement of the automated vehicle. Even with such a configuration, in which only the results of the localization or the results of the dynamic object tracking are required for the corresponding function, the results required in each case can be increased by jointly evaluating the static and dynamic measurement data.

A further embodiment provides that at least one radar sensor, a lidar sensor and/or a video camera are used as the surroundings sensor. These sensors provide particularly reliable measurement data on static and dynamic objects and are therefore particularly well suited to sensing the vehicle's surroundings.

According to a further aspect, there is also a control device for an automated vehicle which is set up to carry out at least some of the steps of the method mentioned above. The advantages already mentioned in connection with the method result for the control device.

According to a further aspect, a computer program is provided comprising instructions which, when the computer program is executed on a computer, cause the latter to carry out one of the methods described above. Implementing the methods described above in the form of a computer program offers a particularly high degree of flexibility.

Finally, according to a further aspect, a computer-readable storage medium is provided, on which the above-mentioned computer program is stored.

• 1 schematisch eine Verkehrssituation, bei der ein automatisiertes Fahrzeug mithilfe bordeigener Sensoren seine Umgebung abtastet;
• 2 schematisch den Aufbau einer Anordnung zur separaten Durchführung einer Lokalisierung und dynamischen Objektverfolgung anhand statischer und dynamischer Messdaten;
• 3 ein Diagramm zur Verdeutlichung der Funktionsweise eines graphbasierten Optimierungsprozesses zu Ermittlung der Ego-Pose des Fahrzeugs;
• 4 ein Diagramm zur Verdeutlichung der Funktionsweise eines graphbasierten Optimierungsprozesses zur kombinierten Ermittlung der Ego-Pose des Fahrzeugs und der relativen Position wenigstens eines umgebenden dynamischen Objekts;
• 5 schematisch den Aufbau einer Anordnung zur Durchführung einer kombinierten Lokalisierung und dynamischen Objektverfolgung;
• 6 schematisch eine Variation der Anordnung aus 5,
• 7 schematisch eine weitere Variation der Anordnung aus 5; und
• 8 ein vereinfachtes Ablaufdiagramm zur Verdeutlichung des Verfahrens.

The invention is described in more detail below with reference to figures. show:

• 1 schematic of a traffic situation in which an automated vehicle scans its surroundings using on-board sensors;
• 2 Schematically the structure of an arrangement for separately carrying out localization and dynamic object tracking using static and dynamic measurement data;
• 3 a diagram to clarify the functioning of a graph-based optimization process for determining the ego pose of the vehicle;
• 4 a diagram to clarify the functioning of a graph-based optimization process for the combined determination of the ego pose of the vehicle and the relative position of at least one surrounding dynamic object;
• 5 schematic of the structure of an arrangement for carrying out a combined localization and dynamic object tracking;
• 6 schematically shows a variation of the arrangement 5 ,
• 7 schematically shows another variation of the arrangement 5 ; and
• 8th a simplified flow chart to clarify the procedure.

The concept described here provides that the static and dynamic measurement data of a radar sensor or another environment sensor of the automated vehicle are not evaluated separately as part of localization and dynamic object tracking, but are evaluated together.

The 1 1 schematically shows an automated vehicle 100 driving on a road 340. The automated vehicle 100 has an environment sensor system 110 for detecting objects and structures in its environment 300 and a control device 120 for controlling the automated vehicle 100. In the present example, the environment sensor system 110 includes several environment sensors, such as a radar sensor 111, a LiDAR sensor 112 and a video camera 113. In the present example, the on-board radar sensor 112 detects surrounding static objects 310 (e.g. traffic signs), dynamic objects 320 (e.g oncoming vehicles) and also unmapped static or semi-dynamic objects 330 (eg parked vehicles).

Automated vehicle 100 also includes sensors for detecting various measured variables and parameters, such as inertial sensors 114 for detecting the current state of motion of automated vehicle 100, a steering angle sensor 115, a GNSS device 116 for satellite-based determination of the global position of automated vehicle 100, and an odometry device 117 for determining the distance covered, for example based on the rotational movement of the vehicle wheels. The automated vehicle 100 also includes a wireless communication device 190 for establishing a wireless communication connection 201 to an external server 210. As in FIG 1 is shown, the automated vehicle 100 and the external server are part of a system 200, which can include other vehicles and servers and certain services (eg, cloud services) offers. In the present example, the server 210 has a storage device 211 in which digital localization maps 212 are stored. In operation, the automated vehicle 100 receives from the server 210 a suitable digital localization map 212 of the environment 300 in which the automated vehicle 100 in question is currently located. The received digital location map 212 may e.g. B. be stored in an on-board storage device 122.

How out 1 can also be seen that in the present example, in addition to storage device 122, control device 120 also includes a control device 121 for determining an optimal trajectory 301 and a movement control device 123 for controlling the movement of automated vehicle 100 along the determined trajectory 301.

In the 2 an arrangement 150 of a radar sensor 111 and two components 121, 122 of the control device 120 of an automated vehicle is shown in the form of a simplified block diagram. In the example shown here, the control device 121 comprises two separate components, namely a localization module 131 and an object tracking module 132. As indicated by the two arrows, the radar sensor 111 sends the static measurement data 510 to the localization module 131 and the dynamic measurement data 520 to the object tracking module 132. The static and dynamic measurement data 510, 520 are then processed or evaluated separately from one another using suitable evaluation methods, such as a filter-based state estimation method or a graph-based optimization method. As a result of this evaluation, the localization module 131 then provides the ego pose 530 of the automated vehicle 100, while the object tracking module 132 provides a list 540 of the dynamic objects 320 detected in the environment with their relative positions as a result. In addition to the static measurement data 510, the localization module 131 also uses the digital localization map 212 stored in the storage device 122 to determine the ego pose 530. The planning device 140 then uses the results 530, 540 provided by the localization module 131 and the object tracking module 132 to plan the Behavior and trajectory 301 of automated vehicle 100. For this purpose, planning device 140 has a behavior planner 141 and a trajectory planner 142.

The determination of the current ego pose of the automated vehicle 100 and the positions of surrounding dynamic objects 320 can in principle be carried out using various methods. In the examples described here, two approaches are used for this, which are also used in Simultaneous Localization and Mapping (SLAM). The first approach concerns a filter-based method that uses, for example, a Kalman filter or an extended Kalman filter (EKF) or a particle filter. In this approach, the unknown state of the respective object, for example the ego pose of the automated vehicle 100 or the relative positions of surrounding dynamic objects 320, is estimated. The estimate is made "on-the-go", i.e. continuously with the latest measurement data. The Kalman filter represents a particularly efficient recursive estimation algorithm that optimally estimates the state of a dynamic system from erroneous observations in the sense of the least squares. The Kalman filter is therefore a mathematical method for iteratively estimating parameters for describing system states on the basis of erroneous observations or measurements.

A so-called particle filter forms another suitable filter for use in the filter-based method. This is a Monte Carlo method, i.e. a stochastic method for estimating the state in a dynamic process, the dynamics of which are only known as statistical averages and which can only be observed incompletely. In the case of dynamic object tracking, an accurate and continuously updated determination of the location and speed of the object under consideration is made due to an inaccurate and erroneous measurement of the location.

The determination of the first-person pose of the vehicle 100 as well as the relative position of surrounding dynamic objects 320 can also be formulated as a pose graph optimization problem. Here, the entire trajectory of each object under consideration is estimated or optimized, taking into account all previous measurements. The aim of a graph-based optimization of the pose of an object is accordingly to estimate the trajectory of the respective object resulting from the sequence of successively measured or determined poses using relative measurements of the pose.

With the help of 3 and 4 The aim is to clarify the difference between a simple graph-based optimization process, which only uses the static measurement data 510, and an extended graph-based optimization process, which also uses the dynamic measurement data 520. For this is in the 3 a so-called sliding graph is shown, in which the ego poses 530 of the automated vehicle 100 determined using the dynamic measurement data 520 and the digital localization map 212 are plotted over a number of consecutive measurement cycles. Such a graph-based optimization process usually takes into account all previous measurement results at the same time, which are compared together with the features from the digital localization map.

The 4 on the other hand, illustrates the evaluation of the measurement data using an extended graph-based optimization method, which also uses a sliding graph in the present example. In contrast to the one in the 3 example shown, this extended sliding graph now includes both the static and the dynamic measurement data 510, 520. In addition to the static measurement data 510 global ego poses of automated vehicle 100 determined over several measurement cycles within the digital localization map, the positions of adjacent dynamic objects 320 relative to automated vehicle 100, also determined over several measurement cycles using dynamic measurement data 520, are also shown. In the extended graph-based optimization method, boundary conditions derived from the static measurement data 510 and boundary conditions derived from the dynamic measurement data 520 are thus added to form a common optimization problem. The combined solution of the two problems, namely the localization and the dynamic object tracking, adds all the boundary conditions specified by the static, dynamic measurement data to a single optimization problem. Compared to the one in the 3 The separate solution of the two problems shown above makes the optimization process more precise thanks to the additional information. Furthermore, the optimization method also becomes more efficient, since in this case only one optimization for the simultaneous solution of the two problems (localization and dynamic object tracking) is necessary.

In the 5 an improved arrangement 150 is shown, which also includes a radar sensor 111 and a control device 120 with a control device 121 and a memory device 122 . In contrast to the arrangement 150 from 2 the localization module 131 and the object tracking module 132 of the control device 121 shown here are combined in a common measurement data processing device 130 . Consequently, the static and dynamic measurement data 510, 520 provided by the radar sensor 111 are sent together to the measurement data processing device 130. After evaluating the measurement data, the results of this evaluation, namely the ego pose 530 of the automated vehicle 100 and a list 540 of the surrounding dynamic objects together with their relative position, are transferred to the planning device 140 . The planning device 41 uses its behavior planner 141 and trajectory planning 142 to determine the respective optimal trajectory 301 for the automated vehicle 100.

In the 6 is a variation of that in the 5 shown improved assembly 150 shown. In this case, an application of the automated vehicle, such as B. the automatic distance control (ACC, Adaptive Cruise Control) no position of the automated vehicle 100 on the map, but only the positions of the surrounding dynamic objects 320 relative to the automated vehicle. In this case, the system architecture and/or the data flow of the control device 121 can correspond to that in FIG 6 shown embodiment can be adapted. The measurement data processing device 130 only transmits the results 540 of the dynamic object tracking, namely the list of the surrounding dynamic objects 320 together with their relative positions, to the hardware or software component executing the respective application (eg the ACC control module).

In the 7 is another variation of the in the 5 shown improved assembly 150 shown. In this case, an application of the automated vehicle 100, such as a lane-level navigation system, does not require any knowledge of neighboring dynamic objects, but only the current pose of the automated vehicle 100. In this case, the joint optimization or filtering of the measurement data 510, 520 delivers as The result is only the current pose 530 of the automated vehicle, which is provided by the measurement data processing device 130 for the hardware or software component executing the corresponding application (eg entertainment system for extended recruiting at lane level).

The 8th finally shows a simplified flow chart of the method 600 described here. Accordingly, in a first step 610 a digital localization map 212 is provided by a storage device 122 of the vehicle 100 . In a further step 620, the surroundings 300 of the automated vehicle 100 are scanned using at least one on-board surroundings sensor 111, 112, 113. The static and dynamic measurement data 510, 520 obtained in this way are provided in step 621. In step 630 there is a joint evaluation of the static and dynamic measurement data 510, 520. This step includes the localization 631, ie the determination of the own pose of the automated vehicle 100 within the localization map 212, and the dynamic object tracking 632, ie the determination of surrounding dynamic Objects 320 and their respective positions relative to the automated vehicle. In the subsequent step 633 the results of the localization and the dynamic object tracking are provided. Then, in step 640, planning takes place, which includes behavior planning 641 and movement planning 642. In the subsequent method step 650, the vehicle is controlled using the results of the planning.

Although the invention has been illustrated and described in detail by the preferred embodiments, the invention is not limited by the disclosed examples. Rather, other variations can also be derived from this by a person skilled in the art without the protection leave the beginning of the invention. For example, depending on the application, the components, devices and modules described here can be implemented in the form of software or hardware or as a combination of software and hardware.

Claims (10)

Method for controlling or automated vehicle (100), wherein static and dynamic measurement data (510, 520) of at least one on-board environment sensor (111, 112, 113) are provided, wherein the static measurement data (510) are assigned to static objects (310) in the environment (300) of the automated vehicle (100). , while the dynamic measurement data (520) are assigned to dynamic objects (320) in the environment (300) of the automated vehicle (100), with a localization (631) being carried out in which the static measurement data (510) are used to determine characteristics of static objects (310) in the environment (300) of the automated vehicle (100) are compared with features from a provided digital localization map (212) in order to determine a global pose of the automated vehicle (100), wherein dynamic object tracking (632) is carried out, in which dynamic objects (320) in the environment (300) of the automated vehicle (100) are determined with their relative positions on the basis of the dynamic measurement data (520), wherein the results (530) of the localization (631) and the results (540) of the dynamic object tracking (632) are determined in a joint evaluation process (630) and using both the static and the dynamic measurement data (510, 520), and wherein the results (530) of the localization (631) and/or the results (540) of the dynamic object tracking (632) are used for planning the behavior and/or the movement of the automated vehicle (100). procedure after claim 1 , the results (530, 540) of the localization (631) and dynamic object tracking (632) being determined using a graph-based optimization method in which the static measurement data (510) derived boundary conditions and the dynamic measurement data (520) derived boundary conditions to one common optimization problem can be added. Method according to one of the preceding claims, wherein the results (530, 540) of the localization (631) and dynamic object tracking (632) are determined using a filter-based state estimation algorithm, wherein the filter state of a filter used in the state estimation algorithm includes both the unknown states of location (631) and the unknown states of dynamic object tracking (632). procedure after claim 3 , wherein the filter-based state estimation algorithm uses a Kalman filter algorithm or a particle filter algorithm. Method according to one of the preceding claims, in which, in addition to the static and dynamic objects (310, 320), unmapped static objects (330) are also used for localization (631) and/or dynamic object tracking (632). Method according to one of the preceding claims, wherein only the results (530) of the localization (631) or only the results (540) of the dynamic object tracking (632) are used for planning the behavior and/or the movement of the automated vehicle (100). Method according to one of the preceding claims, wherein at least one radar sensor (111), a lidar sensor (112) and/or a video camera (113) are used as the surroundings sensor (111, 112, 113). Control device (120) for an automated vehicle (100) which is set up, at least part of the steps of the method according to one of Claims 1 until 7 to execute. Computer program comprising instructions which, when the computer program is executed by a computer, cause the latter to carry out a method according to one of Claims 1 until 7 to execute. A computer-readable storage medium (122) on which a computer program claim 9 is saved.

Images