WO2024068125A1

WO2024068125A1 - Method for assessing trajectories of a trajectory group, and method for operating a vehicle

Info

Publication number: WO2024068125A1
Application number: PCT/EP2023/072436
Authority: WO
Inventors: Julian Jordan; Julian SCHMIDT; Franz Gritschneder
Original assignee: Mercedes-Benz Group AG
Priority date: 2022-09-27
Filing date: 2023-08-15
Publication date: 2024-04-04
Also published as: DE102022003559B3

Abstract

The invention relates to a method for assessing trajectories (T1 to Tn) of a specified trajectory group (TS) for a vehicle (1). According to the invention, for all trajectories (T1 to Tn) of the trajectory group (TS), at specified scanning points (A1 to Az) lying ahead along the relevant trajectory (T1 to Tn), a situation to be expected at a particular scanning point (A1 to Az) is predicted in a grid map (RK, RK1 to RKx). The grid map (RK) has a plurality of layers and has one cost plane each for various specified tasks of the vehicle (1), in the grid cells (RZ) of which cost plane costs for the relevant task are entered. The costs specify a measure indicating to what extent the relevant task is being fulfilled at the location of the relevant grid cell (RZ), with lower costs being allocated to a fulfillment of the relevant task to a greater extent than to a fulfillment of the task to a lesser extent. For each scanning point (A1 to Az) on each cost plane, a cost value for the predicted situation is determined, the determination of the cost values being based on: a convolution of grid cells (RZ) of the relevant cost plane with grid cells (RZ), correspondingly equally positioned, of a grid presentation of the vehicle that is approximated for the particular scanning point; and a summary of the results of the convolution by means of a reduction formula. A total cost value is determined for the relevant trajectory (T1 to Tn) from the cost values determined via a prediction horizon, and is allocated to the relevant trajectory (T1 to Tn), the trajectory (T1 to Tn) having the lowest total cost value being identified as the optimum trajectory (T1 to Tn) from the trajectory group (TS).

Description

Method for evaluating trajectories of a set of trajectories and method for operating a vehicle

The invention relates to a method for evaluating trajectories of a given set of trajectories for a vehicle.

The invention further relates to a method for operating an automated vehicle.

From "Stiller, C. and Ziegler, J.: Fast Collision Checking for Intelligent Vehicle Motion Planning; In: 2010 IEEE Intelligent Vehicles Symposium, University of California, San Diego, CA, USA, June 21, 2010 to June 24, 2010" a method for fast collision checking for an application in motion planning for intelligent vehicles is known. In order to explicitly take the orientation of the vehicle into account, a vehicle shape is broken down into several disc-shaped basic shapes, whereby the collision check is broken down into several collision tests.

From US 2021/0 347 382 A1, a method for planning a trajectory for an autonomous vehicle is known, wherein the planning is carried out using a multi-layer grid map, which includes as layers an occupancy map with static objects, an occupancy map with dynamic objects predicted for a future period, and a cost map created using environmental data. The trajectory planning is based on the assignment of costs to a series of position points in advance and the selection of a most cost-effective set of position points.

From “Lu., DV, Hershberger, D., Smart, WD: Layered Costmaps for Context-Sensitive Navigation; IEEE/RSJ International Conference on Intelligent Robots and Systems, Chicago, IL, USA, 2014, pp. 709-715; lEEE Xplore [online], DOI: 10.1109/IROS.2014.6942636; In: IEEE” a method for trajectory planning for a robot is known, whereby it is provided that a plurality of cost maps are created for different functionalities of the robot, that the data of the various cost maps are aggregated in a master cost map and that the master cost map is used as a basis for trajectory planning.

From US 2020/0 159 225 A1 a movement planning model for an autonomous vehicle is known, it being provided that costs for future positions of the vehicle are determined using a model based on machine learning, that several possible trajectories are evaluated with the determined costs and that the trajectory with the lowest cost is selected.

The invention is based on the object of specifying a novel method for evaluating trajectories of a given set of trajectories for a vehicle and a novel method for operating an automated vehicle.

The object is achieved according to the invention by a method for evaluating trajectories of a predetermined set of trajectories for a vehicle, which has the features specified in claim 1, and by a method for operating an automated vehicle, which has the features specified in claim 10.

Advantageous embodiments of the invention are the subject of the subclaims.

In the method for evaluating trajectories of a given set of trajectories for a vehicle, according to the invention, a situation expected at a respective sampling point is predicted in a grid map for all trajectories of the set of trajectories at predetermined preceding sampling points on the respective trajectory. The grid map is constructed in several layers and has a cost level for various predetermined tasks of the vehicle, in whose grid cells costs for the respective task are entered. The costs indicate a measure of the extent to which, i.e. how well or poorly, the respective task is fulfilled at the location of the respective grid cell, whereby fulfilling the respective task to a greater extent is assigned lower costs than fulfilling the task with a smaller extent. Furthermore, a cost value for the predicted situation is determined for each sampling point on each cost level, whereby the determination of the cost values is based on a convolution of grid cells of the respective cost level with correspondingly equally positioned Grid cells of a grid representation of the vehicle approximated for the respective sampling point and a summary of the results of the convolution using a reduction formula. From the cost values determined over a forecast horizon, a total cost value for the respective trajectory is determined and assigned to the respective trajectory, and the trajectory with the lowest total cost value is identified as the optimal trajectory from the family of trajectories.

Alternatively, higher costs can be assigned to fulfilling the respective task on a larger scale than to fulfilling the task on a smaller scale, in which case the trajectory with the highest total cost value is then identified as the optimal trajectory from the set of trajectories.

An evaluation of a cost function is an essential component for motion planning of an automated, particularly highly automated or autonomous vehicle. The cost function defines in particular what the optimal behavior and thus the optimal trajectory for the vehicle is.

However, finding this optimal trajectory is not trivial due to the high dimensionality of the problem. For example, a behavior planning module of an automated vehicle creates a set of hypotheses for various options for the vehicle to act. These hypotheses are represented in the form of trajectories and form a family of trajectories.

The present method enables a particularly efficient and reliable selection of the optimal trajectory from such a set of trajectories. Furthermore, a particularly flexible evaluation is made possible.

Determining the cost values for each sampling point at each cost level using the above-mentioned convolution enables a particularly efficient and reliable determination of the cost values.

In one possible embodiment of the method, the raster representation of the vehicle itself is also multi-layered. By using the multi-layer raster map and corresponding multi-layer raster representations of the vehicle, the flexibility of the evaluation can be further increased. In particular, the resolution of the raster map and the raster representation of the vehicle is chosen to be identical. In another possible embodiment of the method, different layers of the raster representation of the vehicle are used for the convolution for different cost levels of the raster map. This means that the raster representations of the vehicle are not limited to a binary mask, so that, similar to a classic convolution matrix, any values can be used. This enables accelerated evaluation.

In a further possible embodiment of the method, a grid representation corresponding to an orientation of the vehicle at this sampling point is selected at the respective sampling point and the selected grid representation is transformed to an integer position on the grid map that is closest to the real position of the vehicle. The result of this is that no interpolation of the values is required, which further speeds up the process.

In a further possible embodiment of the method, a sum function and/or an average value function and/or a maximum value function are used as the reduction function. Such functions can be easily implemented into the method and deliver reliable results.

In another possible embodiment of the method, the cost values of the cost levels are weighted when determining the total costs. The weights can be used, for example, to approximate a probabilistic description of the uncertainty.

In another possible embodiment of the method, the tasks of the vehicle are derived from a static world and/or a dynamic world and/or traffic rules. This enables full consideration of all objects and events present in the vehicle's environment and relevant to its automated operation.

In order to be able to realize an optimized automated ferry operation of the vehicle, a further possible embodiment of the method provides that safety-relevant tasks and/or tasks relevant to compliance with traffic regulations and/or tasks concerning the comfort of vehicle occupants and/or tasks concerning the efficiency of the vehicle are specified as tasks of the vehicle. In the method according to the invention for operating an automated vehicle, a set of trajectories with several potential trajectories for reaching a predetermined destination of the vehicle is formed depending on the environmental situation of the vehicle, wherein in an automated ferry operation of the vehicle an optimal trajectory for reaching the destination is selected from the set of trajectories according to the previously described method for evaluating trajectories of a predetermined set of trajectories for a vehicle. The method enables a particularly efficient, fast and reliable selection of the optimal trajectory. The automated operation of the vehicle can thus be optimized.

Exemplary embodiments of the invention are explained in more detail below with reference to drawings.

Show:

Fig. 1 schematically shows an object-based representation of a vehicle’s environment,

2 schematically shows a grid representation of lane boundaries,

3 schematically shows a grid representation of lane directions,

4 schematically shows a grid representation of open spaces,

5 schematically shows a grid representation of lane occupancies,

Fig. 6 schematically shows a grid representation of navigable areas of

lanes,

7 schematically shows a grid representation of target lanes,

8 schematically shows a grid representation of free space semantics,

9 schematically shows a grid representation of lane occupancy semantics,

10 shows a schematic representation of a motion prediction of a scene, 11 schematically shows a grid-based representation of a vehicle, an environment of the vehicle and a set of trajectories,

12 schematically shows a multi-layer grid map with a multi-layer grid representation of a vehicle,

Fig. 13 schematically shows a plan view of a vehicle,

Fig. 14 schematically shows a grid representation of the vehicle according to Figure 13,

15 schematically shows a grid representation of the vehicle according to FIG. 13 with a changed orientation,

16 schematically shows an expanded grid representation of a vehicle,

17 schematically shows an expanded grid representation of a vehicle,

18 schematically shows a grid-based representation of a vehicle and an environment of the vehicle,

Fig. 19 schematically shows an enlarged section of a grid-based representation of a vehicle and an environment of the vehicle,

20 shows schematically the grid-based representation of the vehicle and the surroundings of the vehicle according to FIG. 18 as well as sampling points of a trajectory of the vehicle at a first point in time,

Fig. 21 schematically shows the grid-based representation according to Figure 20 at a later second point in time and

22 shows schematically the grid-based representation according to FIG. 20 at a later third point in time.

Corresponding parts are provided with the same reference numbers in all figures. Figure 1 shows schematically an object-based representation of an environment of a vehicle 1.

The vehicle 1 is designed in particular for automated ferry operations. To carry out the automated ferry operation, a trajectory T1 to Tn, shown in more detail in FIG. 11, must be found, which the vehicle 1 follows during its automated ferry operation. In order to determine this trajectory T1 to Tn, the surroundings of the vehicle 1 must be recorded and evaluated.

Based on the detection of the environment by vehicle sensors and subsequent processing of environmental data detected by the sensors, a model of the environment can be created which describes a current state of the vehicle 1 and the environment.

For planning a trajectory T1 to Tn, an expected temporal development of the environment is also determined. The representation of such a temporally changing environment is done, for example, using an object-based representation according to Figure 1. In this object-based representation, each object 01 to Om has its own representation, which describes a state of the object 01 to Om, i.e., for example, a type, a position, a speed and an acceleration of the object 01 to Om as well as other characteristics. The object-based representation enables a precise description of the individual objects 01 to Om, but the runtime of subsequent processing steps depends on the number of objects 01 to Om.

In order to find an optimal trajectory T1 to Tn, a behavior planning module of the vehicle 1 (not shown in detail) creates a set of hypotheses for various options for action of the vehicle 1. These hypotheses are shown, for example, in the form of trajectories T1 to Tn and form a family of trajectories TS, also shown in Figure 11. The movement planning of the vehicle 1 is carried out, among other things, by evaluating a cost function, which defines an optimal behavior of the vehicle 1 and thus the optimal trajectory T1 to Tn to be driven.

The decisive factor for selecting a trajectory T1 to Tn is the static world with static objects 01 to Om and obstacles, a road topology, road markings and a destination. The dynamic World with other road users, traffic rules, traffic lights and right of way rules.

These elements must be taken into account by the cost function, with other road users, obstacles, road topology and markings being represented in the object-related representation as simplified geometric objects, for example polygons. Traffic lights and traffic rules, for example, are modeled as stop lines.

The cost function describes expected costs for a future state of the vehicle 1 and evaluates, for example, safety-relevant tasks and/or tasks relevant to compliance with traffic rules and/or tasks relating to the comfort of vehicle occupants and/or tasks relating to the efficiency of the vehicle 1. Safety-relevant tasks include, for example, avoiding collisions with other road users and other objects. In order to fulfill tasks relevant to compliance with traffic rules, regulations specified by traffic lights, zebra crossings, speed limits, etc. are taken into account. Tasks relating to the comfort of vehicle occupants include, for example, minimizing accelerations acting on vehicle occupants and tasks relating to the efficiency of the vehicle 1 include, for example, minimizing the energy required by the vehicle 1 for various driving maneuvers to reach a destination.

To evaluate the cost function, for example to check whether there is a risk of collisions between the vehicle 1 and the objects 01 to Om on the corresponding trajectory T1 to Tn, a geometry of the vehicle 1 must be compared each time with a geometry of the objects 01 to Om become.

Costs determined using the cost function indicate the extent to which the respective task is fulfilled when traveling the respective trajectory T1 to Tn, whereby fulfilling the respective task to a greater extent is assigned lower costs than fulfilling the task with a smaller scope. The aim of trajectory planning is then to find the trajectory T1 to Tn that has the lowest costs. Alternatively, higher costs can be assigned to completing the respective task on a larger scale than to completing the task on a smaller scale, in which case the trajectory T1 to Tn with the highest costs is identified as the optimal trajectory T1 to Tn.

Another model of the environment of the vehicle 1, which enables the evaluation of such a cost function, is a grid-based representation of the environment of the vehicle 1. Here, all objects 01 to Om are entered into a common, discretized world representation. Such a grid-based representation enables a constant runtime in the subsequent processing steps, with examples of grid-based representations being shown in Figures 2 to 9.

Figure 2 shows a raster representation of lane boundaries FB (= raster map RK1), Figure 3 shows a raster representation of lane directions FR1, FR2 (= raster map RK2), Figure 4 shows a raster representation of free spaces FS

(= grid map RK3) and Figure 5 shows a grid representation of lane occupancies FO (= grid map RK4). In Figure 6 is a grid representation of drivable areas DS of lanes (= grid map RK5), in Figure 7 a grid representation of target lanes ZF (= grid map RK6), in Figure 8 a grid representation of a free space semantics FSS (= grid map RK7) and in Figure 9 one Raster representation of a lane occupancy semantics FOS (= raster map RK8). Each grid representation represents a possible cost term for the cost function.

Figure 10 shows a raster representation of a movement prediction of a scene (= raster map RK9). For each sampling time t1 to t5, grid cells occupied by objects 01 to Om, for example other road users such as vehicles, are estimated. This is required to represent the dynamic world as a raster map RK9.

The costs for a specific position and orientation of vehicle 1 at a specific point in time can be determined from these grid representations or

Derive grid maps RK, RK1 to RKx. The vehicle 1 should, for example, drive in the right direction within a lane (= taking the static world into account), get closer to a destination (= taking the static world into account) and not use any grid cells that are probably already used by other objects 01 to Om, for example others Vehicles are occupied (= taking the dynamic world into account). Figure 11 shows a grid map RK10 with a vehicle 1, an object 01, for example another vehicle, and a set of trajectories TS with several trajectories T1 to Tn, which represent hypotheses for various options for action of the vehicle 1 and, for example, by means of a behavior planning module of the vehicle, not shown in more detail 1 can be generated. The grid card RK10 has a plurality of grid cells RZ.

Both to represent the different cost types, as shown for example in Figures 2 to 9, and to represent the surroundings of the vehicle 1 at different points in time to represent the dynamic world, a multi-layer raster map RK is used here, which has several individual raster maps RK1 to RKx.

Such a multi-layer, time-varying raster map RK is shown in FIG. 12, through which a cost function is approximated and which thus represents a multi-layer cost map of the surroundings of the vehicle 1. This means that the grid map RK has a grid map RK1 to RKx designed as a cost level for various predetermined tasks of the vehicle 1, in whose grid cells RZ costs for the respective task are entered. The costs indicate the extent to which the respective task at the location of the respective grid cell RZ is fulfilled by the vehicle 1.

The tasks of the vehicle 1 are derived from the static world, the dynamic world and/or traffic rules and concern, for example, safety-relevant tasks, tasks relevant to compliance with traffic rules, tasks concerning the comfort of vehicle occupants and/or tasks concerning the efficiency of the vehicle 1.

For example, good performance of the respective task is rewarded with low costs, while poor performance is sanctioned with high costs. This means that the costs entered in a grid cell RZ are higher, the more undesirable the stay of the vehicle 1 is at the location of the respective grid cell RZ. Staying at the location of a grid cell RZ is, for example, undesirable if it is associated with poor performance of the task, for example if it is associated with lane departure, a high risk of collision, a violation of traffic rules, uncomfortable driving or energetically inefficient driving. For an efficient evaluation of a cost function on such a multi-layer grid map RK, detailed grid representations RD of the vehicle 1, shown in more detail in FIG. 13, are used. The grid representation RD of the vehicle 1 shows the space occupied by the vehicle 1 at a specific location. The vehicle 1 is advantageously approximated by a multi-layer grid representation RD, which has several individual grid representations RD1 to RDx of the vehicle 1 with grid cells RZ. The resolution of the raster maps RK1 to RKx and the associated raster representations RD1 to RDx of vehicle 1 are identical.

For each layer of the raster map RK forming the cost map, that is, for each individual raster map RK1 to RKx, an individual raster representation RD1 to RDx of vehicle 1 can be used. The raster representations RD1 to RDx of vehicle 1 are not limited to a binary mask. Any values can be used, similar to a classic convolution matrix.

A top view of a vehicle 1 is shown in FIG.

Figure 14 shows a grid representation RD1 of the vehicle 1 according to Figure 13. The grid representation RD1 can be generated for different orientations of the vehicle 1, in particular for 360 ° with configurable angular resolution, and stored in a memory. By storing the data in a memory, the evaluation can be accelerated. -

Figure 15 shows such a raster representation RD2 of vehicle 1 with a changed orientation.

Figures 16 and 17 show various extended raster representations RD3, RD4 of a vehicle 1. Grayscales in the raster representations RD3, RD4 describe different values of the raster representation RD3, RD4, each designed as a convolution mask. This can be used to model safety distances and probabilities, for example. The brighter a raster cell RZ is, the higher the costs of the cost cell charged with it are weighted. This can ensure that the vehicle 1 maintains a certain distance from obstacles, which may only be exceeded in critical situations. Figure 18 shows a raster-based representation of a vehicle 1 and an environment of the vehicle 1, in particular a raster map RK10 according to Figure 11.

The raster map RK10 represents a cost map, with grayscale describing the costs for driving the vehicle 1 within a lane. The brighter the raster cells RZ are, the higher the costs. The object 01, which is, for example, a predicted other vehicle 1, is also approximated with associated costs in the raster map RK10. The vehicle 1 itself is also approximated at its corresponding vehicle position in the raster map RK10 by its transformed raster representation RD10, which is designed in particular as a convolution mask. The size of the raster cells RZ is, for example, 13 cm x 13 cm.

If the cost function is now to be evaluated for a specific position and orientation of the vehicle 1, a grid representation RD1 to RDx of the vehicle 1 corresponding to the orientation of the vehicle 1 is first selected and then transformed to the next integer position on the corresponding grid map RK1 to RKx.

This is shown schematically and by way of example in FIG. 19 for a grid map RK1 and a grid representation RD1 of the vehicle 1.

First, the raster representation RD1 of vehicle 1 is transformed to a real-value position on the raster map RK1 of the environment. This is shown in the form of a dashed representation of the raster representation RD1. The raster representation RD1 is then rounded to the next integer position, shown with solid lines. This means that no interpolation of the values is required, which speeds up the process.

Figures 20 to 22 show the raster map RK10 of the vehicle 1 and the surroundings of the vehicle 1 according to Figure 18 as well as sampling points A1 to Az of a trajectory T 1 of the trajectory family TS of the vehicle 1 at different times.

To determine the costs, for each trajectory T1 to Tn of the trajectory family TS, the situation expected at the respective sampling point A1 to Az is predicted in the raster map at predetermined preceding sampling points A1 to Az on the respective trajectory T1 to Tn. This means that the respective trajectory T1 to Tn is sampled at a certain frequency, represented by the sampling points A1 to Az, for example every 0.1 seconds up to a given look-ahead horizon, and the corresponding cost function is evaluated at the corresponding positions. The trajectory T1 to Tn to be evaluated represents an input parameter.

In the exemplary embodiment shown, the costs for maintaining the lane (gray levels) and the costs for a collision with the object 01 are offset for all grid cells RZ within the grid representation RD1, in particular a convolution mask. For each sampling point A1 to Az of the trajectory T1 to Tn, a prediction of the movement of the other objects 01 to Om is required.

This is carried out for each trajectory T1 to Tn of the trajectory family TS at each cost level of the raster map RK, i.e. on each raster map RK1 to RKx, at the same position. For example, to determine the costs, the cost values in the raster cells RZ in the respective raster representation RD1 to RDx are multiplied by the cost values in the raster cells RZ of the respective raster map RK1 to RKx. The results of the individual cost levels are then combined with different weights.

The costs determined are calculated using an aggregation or reduction function, for example a maximum function, mean value function or sum function, to obtain a cost value for the entire trajectory T1 to Tn. This means that for each sampling point A1 to Az, a cost value for the predicted situation is determined using the cost levels. A total cost value is then determined from the cost values determined over the forecast horizon and assigned to the respective trajectory T1 to Tn. The total cost value is the result of the trajectory evaluation.

This is carried out for each trajectory T1 to Tn from the set of trajectories TS, whereby the trajectory T1 to Tn with the lowest total cost value is identified as the optimal trajectory T1 to Tn and is selected from the set of trajectories TS in the automated ferry operation of the vehicle 1 to achieve the destination.

In summary, the course of the procedure described above can be briefly described as follows:

- Selection of the corresponding grid representation RD1 to RDx for a requested orientation of vehicle 1. - Transformation of the selected raster representation RD1 to RDx to the integer position closest to the actual position on the corresponding raster map RK1 to RKx.

- Selection of time-dependent cost levels according to a timestamp.

- Folding the grid cells RZ of the cost levels with the grid cells within the transformed grid representations RD1 to RDx of vehicle 1.

- Apply the reduction function to the individual convolution results.

- The weighted sum of the results of the reduction functions gives the costs.

Claims

Claims Method for evaluating trajectories (T 1 to T n) of a predetermined set of trajectories (TS) for a vehicle (1), where

- for all trajectories (T 1 to T n) of the trajectory family (TS), a situation expected at a respective sampling point (A1 to Az) is predicted in a raster map (RK, RK1 to RKx) at predetermined preceding sampling points (A1 to Az) on the respective trajectory (T1 to Tn),

- the grid map (RK) is constructed in multiple layers and has a cost level for various specified tasks of the vehicle (1), in whose grid cells (RZ) costs for the respective task are entered,

- the costs indicate the extent to which the respective task is fulfilled at the location of the respective grid cell (RZ),

- lower costs are allocated to the performance of the respective task on a larger scale than to the performance of the task on a smaller scale,

- a cost value for the predicted situation is determined for each sampling point (A1 to Az) at each cost level, the cost values being determined on a convolution of grid cells (RZ) of the respective cost level with correspondingly equally positioned grid cells (RZ) for the respective sampling point (A1 to Az) approximated grid representation (RD1 to RDx) of the vehicle (1) and a summary of the results of the folding using a reduction formula,

- a total cost value for the respective trajectory (T1 to Tn) is determined from the cost values determined over a look-ahead horizon for the sampling points (A1 to Az) of the trajectories (T1 to Tn) and the respective one

trajectory (T1 to Tn) and - the trajectory (T1 to Tn) with the lowest total cost value is identified as the optimal trajectory (T1 to Tn) from the trajectory family (TS). Method according to claim 1, characterized in that the raster representation (RD) of the vehicle (1) is designed in multiple layers. Method according to claim 2, characterized in that for different cost levels of the raster map (RK), different layers of the raster representations (RD1 to RDx) of the vehicle (1) are used in the folding. Method according to one of the preceding claims, characterized in that

- at the respective sampling point (A1 to Az) a raster representation (RD1 to RDx) corresponding to an orientation of the vehicle (1) at this sampling point (A1 to Az) is selected and

- the selected raster representation (RD1 to RDx) to an integer position on the map closest to the real position of the vehicle (1)

Raster map (RK1 to RKx) is transformed. Method according to one of the preceding claims, characterized in that a sum function and/or a mean value function and/or a maximum value function are used as the reduction function. Method according to one of the preceding claims, characterized in that when determining the total costs, the cost values of the cost levels are each weighted. Method according to one of the preceding claims, characterized in that the tasks of the vehicle (1) consist of

- a static world and/or

- a dynamic world and/or

- Traffic rules be derived. Method according to one of the preceding claims, characterized in that the tasks of the vehicle (1)

- security-related tasks and/or

- tasks relevant to compliance with traffic rules and/or

- tasks concerning the comfort of vehicle occupants and/or

- tasks relating to the efficiency of the vehicle (1) are specified. Method for operating an automated vehicle (1), wherein

- Depending on an environmental situation of the vehicle (1), a set of trajectories (TS) with several potential trajectories (T1 to Tn) is formed to achieve a predetermined goal of the vehicle (1) and

- in an automated ferry operation of the vehicle (1), an optimal trajectory (T1 to Tn) for reaching the destination is selected from the set of trajectories (TS) according to a method according to one of the preceding claims.