Granted Patent B2

US 12,637,111 · App. 18/828,095 · Granted May 26, 2026

Inventors: Daniel Grimm (Karlsruhe, DE); Alexander Naumann (Karlsruhe, DE); Felix Hertlein (Karlsruhe, DE); Juergen Luettin (Renningen, DE); Maximilian Zipfl (Karlsruhe, DE); Achim Rettinger (Trier, DE); Lavdim Halilaj (Leonberg, DE); Marius Zoellner (Karlsruhe, DE); Stefan Schmid (Waiblingen, DE); Steffen Thoma (Karlsruhe, DE)

Assignee: ROBERT BOSCH GMBH

B60W60/00274B60W50/0097G06N3/02B60W2552/53B60W2554/4041B60W2554/4045B60W2556/40

Abstract

Claims (45)

1 . A computer-implemented method for trajectory prediction, comprising the following steps:

receiving, by a prediction module, trajectory data of motion trajectories of road users arranged in a surrounding area of an ego vehicle, wherein the trajectory data are arranged in a graph representation, wherein nodes of the graph representation include position information of the road users and/or speed information of the road users, and wherein edges of the graph representation define temporal relations between the position information of the road users and/or the speed information of the road users;

receiving, by the prediction model, map data of a map representation mapping the surrounding area of the ego vehicle, wherein the map data are arranged in a graph, wherein nodes of the graph representation include position information of a roadway boundary element of a lane and/or lane, used by the ego vehicle and/or by other road users, and wherein edges of the graph representation define spatial relations between the position information of the roadway boundary element;

generating, by the prediction model, an interaction graph representation for the road users based on the trajectory data of the road users and the position information of the roadway boundary elements of the map representation, wherein each node of the interaction graph representation represents a road user positioned in the surrounding area of the ego vehicle and includes position information and/or speed information of the road user, and wherein each edge of the interaction graph representation defines an arrangement relation between two of the road users represented by the nodes; and

predicting, by the prediction module, a future motion trajectory to be executed for at least one other road user based on the trajectory data, the map data, and the interaction graph representation of the road users.

2 . The method according to claim 1 , further comprising the following:

generating, by the prediction module, anchor paths based on the trajectory data of the road users and the map data of the map representation, wherein the anchor paths define regions on roadways used by the road users in which possible motion trajectories can be arranged;

wherein the prediction of the future motion trajectory is effected based on the trajectory data, and/or the map data, and/or the interaction graph representation of the road users, and/or the anchor paths.

3 . The method according to claim 2 , wherein the generating of the anchor paths includes:

localizing, by the prediction module, at least one road user on a roadway and/or lane;

identifying a roadway segment and/or lane segment of the roadway and/or lane on which the road user is positioned, wherein the roadway segment and/or lane segment defines a partial region of the roadway and/or lane, wherein the roadway segment and/or the lane segment is bounded by roadway boundary elements of the roadway and/or lane boundary elements of the lane, wherein the roadway segment can be connected to further roadway segments and/or the lane segment can be connected to further lane segments, and wherein a plurality of roadway segments forms the roadway and/or a plurality of lane segments ( 233 ) forms the lane; and

connecting the roadway segment and/or the lane segment to further roadway segments and/or lane segments, each of which can be connected to one another, to form the anchor path, taking into account prevailing traffic rules and a course of the roadway and/or lane.

4 . The method according to claim 3 , wherein the anchor path includes a roadway change and/or a lane change, provided that a roadway change and/or a lane change is compatible with the course of the roadway and/or lane and with prevailing traffic rules.

5 . The method according to claim 2 , further comprising the following:

classifying the other road users based on environmental sensor data of at least one environmental sensor of the ego vehicle by defining an environmental detection as one of the following list: vehicle, bus, truck, streetcar, motorcyclist, cyclist, pedestrian, animal;

wherein anchor paths are generated only for road users, which are classified as vehicle, bus, truck, streetcar, motorcyclist.

6 . The method according to claim 1 , further comprising the following:

receiving, by the prediction module, further map data of a map representation mapping the surrounding area of the ego vehicle, wherein the further map data are image data; and

reading out, by the prediction module, roadway characteristics of roadways used by the road users from the further map data;

wherein the prediction of the future motion trajectories is effected based on the trajectory data, the map data, the interaction graph representation of the road users, and the roadway characteristics.

7 . The method according to claim 6 , wherein the roadway characteristics include the following: stop lines, traffic signs, traffic lights, crosswalks, roadway environments, sidewalks, parking lots.

8 . The method according to claim 6 , wherein arrangement relations of the edges of the interaction graph representation include lane information with respect to the road users, and wherein the lane information includes lane information:

that a road user is positioned in the same lane as another road user and/or the ego vehicle, and/or

that a road user is positioned in a lane adjacent to the lane of the other road user and/or the ego vehicle, and/or

that a road user is positioned in a lane that crosses the lane of the other road user and/or the ego vehicle; and

wherein the prediction of the future motion trajectories is effected based on the trajectory data, the map data, the interaction graph representation of the road users, the roadway characteristics, and the arrangement relations of the interaction graph representation.

9 . The method according to claim 8 , wherein the predicting of the motion trajectories includes the following:

ascertaining for road users in the surrounding area of the ego vehicle distances of the road users to other road users and/or to the ego vehicle, based on the position information and lane information of the interaction graph representation;

ascertaining probability values for collisions of road users among one another or with the ego vehicle based on the distances and speed information of the road users; and

wherein the motion trajectories are effected taking into account the ascertained probability values.

10 . The method according to claim 1 , wherein the nodes of the interaction graph representation include information with respect to a type of the road user, and wherein the type is defined as one of the following list: vehicle, bus, truck, motorcyclist, cyclist, pedestrian, animal.

11 . The method according to claim 1 , wherein the prediction module is an artificial intelligence that is trained to execute the steps of the method, wherein the prediction module includes a user level, a map level, a fusion level, a merger level, and at least one encoder element, wherein the user level is configured to process the trajectory data and/or to generate the interaction graph representation, wherein the map level is configured to process the map data, wherein the fusion level is configured to fuse results of the user level and the map level, wherein the at least one encoder element is configured to process further map data including mage data and to extract roadway characteristics and/or lane characteristics from the map data, and wherein the merger level is configured to predict and output at least one motion trajectory of at least one road user based on results of the fusion level and results of the encoder element.

12 . The method according to claim 11 , wherein the artificial intelligence includes at least one graph neural network, and/or wherein the encoder element is an autoencoder and is configured to generate a latent space representation in a latent space of the roadway characteristics based on the image data of the further map data.

13 . The method according to claim 1 , further comprising:

executing at least one control function of the ego vehicle taking into account the predicted driving trajectory of the vehicle.

14 . A computing unit for trajectory prediction, the computing unit configured to:

receive, by a prediction module, trajectory data of motion trajectories of road users arranged in a surrounding area of an ego vehicle, wherein the trajectory data are arranged in a graph representation, wherein nodes of the graph representation include position information of the road users and/or speed information of the road users, and wherein edges of the graph representation define temporal relations between the position information of the road users and/or the speed information of the road users;

receive, by the prediction model, map data of a map representation mapping the surrounding area of the ego vehicle, wherein the map data are arranged in a graph, wherein nodes of the graph representation include position information of a roadway boundary element of a lane and/or lane, used by the ego vehicle and/or by other road users, and wherein edges of the graph representation define spatial relations between the position information of the roadway boundary element;

generate, by the prediction model, an interaction graph representation for the road users based on the trajectory data of the road users and the position information of the roadway boundary elements of the map representation, wherein each node of the interaction graph representation represents a road user positioned in the surrounding area of the ego vehicle and includes position information and/or speed information of the road user, and wherein each edge of the interaction graph representation defines an arrangement relation between two of the road users represented by the nodes; and

predict, by the prediction module, a future motion trajectory to be executed for at least one other road user based on the trajectory data, the map data, and the interaction graph representation of the road users.

15 . A non-transitory computer-readable storage medium on which is stored a computer program including commands for trajectory prediction, the commands, when executed by a data processor, causing the data processor to perform the following steps:

receiving, by a prediction module, trajectory data of motion trajectories of road users arranged in a surrounding area of an ego vehicle, wherein the trajectory data are arranged in a graph representation, wherein nodes of the graph representation include position information of the road users and/or speed information of the road users, and wherein edges of the graph representation define temporal relations between the position information of the road users and/or the speed information of the road users;

receiving, by the prediction model, map data of a map representation mapping the surrounding area of the ego vehicle, wherein the map data are arranged in a graph, wherein nodes of the graph representation include position information of a roadway boundary element of a lane and/or lane, used by the ego vehicle and/or by other road users, and wherein edges of the graph representation define spatial relations between the position information of the roadway boundary element;

generating, by the prediction model, an interaction graph representation for the road users based on the trajectory data of the road users and the position information of the roadway boundary elements of the map representation, wherein each node of the interaction graph representation represents a road user positioned in the surrounding area of the ego vehicle and includes position information and/or speed information of the road user, and wherein each edge of the interaction graph representation defines an arrangement relation between two of the road users represented by the nodes; and

predicting, by the prediction module, a future motion trajectory to be executed for at least one other road user based on the trajectory data, the map data, and the interaction graph representation of the road users.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (14)