CN-121989999-A - Driving track determining method and device

CN121989999ACN 121989999 ACN121989999 ACN 121989999ACN-121989999-A

Abstract

The application provides a driving track determining method and device, the determining method comprises the steps of obtaining at least one initial planning track output by a decision planning model aiming at a target vehicle, conducting smoothing processing on the initial planning track to obtain at least one model output track, generating a safety bottom planning track of at least one target vehicle through a Toren track generating method based on an obtained local map related to the target vehicle, conducting arbitration evaluation on the model output track and the safety bottom planning track, and taking the evaluated optimal track as an executable track of the current target vehicle. By the method and the device, the safety, the comfort and the trafficability of automatic driving can be greatly improved.

Inventors

WANG YU
ZHANG YUE
CHEN SHIYUAN
WANG GENG
WANG YE

Assignees

中国第一汽车股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260330

Claims (10)

1. A method for determining a driving trajectory, the method comprising: Acquiring at least one initial planning track output by a decision planning model aiming at a target vehicle; Smoothing the initial planning track to obtain at least one model output track; Generating a safety bottom planning track of at least one target vehicle by a Tonlon track generation method based on the acquired local map related to the target vehicle; And carrying out arbitration evaluation on the model output track and the safety bottom planning track, and taking the evaluated optimal track as an executable track of the current target vehicle.
2. The method according to claim 1, wherein the step of smoothing the initial planned trajectory to obtain at least one model output trajectory comprises: Filtering each initial planning track to obtain a target sampling point corresponding to the initial planning track; and optimizing the target sampling points corresponding to the initial planning track by using a pre-constructed punishment function to obtain a model output track corresponding to the model output track.
3. The method of determining according to claim 1, wherein the step of generating at least one safety-spam planned trajectory by a homotopy trajectory generation method based on the acquired local map related to the target vehicle comprises: Acquiring a local map related to a target vehicle, wherein the local map comprises static and dynamic obstacle information; Based on the drivable area divided by the obstacle in the local map, making a transverse decision and a longitudinal decision; fusing the transverse decision and the longitudinal decision to generate at least one homotopy traffic decision; And calling a mathematical optimization algorithm to generate a safety bottom planning track of at least one target vehicle based on the Tonlun traffic decision.
4. The method according to claim 1, wherein the step of performing arbitration evaluation on the model output trajectory and the safety bottom planning trajectory, and taking the evaluated optimal trajectory as an executable trajectory of the current target vehicle includes: Respectively inputting the model output track and the safety bottom planning track into a target evaluator to obtain evaluation scores of each model output track and each safety bottom planning track; Determining whether the track meets the execution requirement according to the evaluation score; If so, determining the track with the minimum evaluation score in the qualified tracks as the executable track of the current target vehicle.
5. The determination method according to claim 4, characterized in that the determination method further comprises: If the current running data of the target vehicle is not met, fitting the executable track determined at the current previous sampling moment with the current running data of the target vehicle, and determining the track obtained after fitting as the executable track of the current target vehicle.
6. A determination device of a driving trajectory, characterized in that the determination device comprises: the acquisition module is used for acquiring at least one initial planning track output by the decision planning model aiming at the target vehicle; the smoothing module is used for carrying out smoothing treatment on the initial planning track to obtain at least one model output track; the generation module is used for generating a safety bottom planning track of at least one target vehicle through a homotopy track generation method based on the acquired local map related to the target vehicle; And the evaluation module is used for carrying out arbitration evaluation on the model output track and the safety bottom planning track, and taking the evaluated optimal track as the executable track of the current target vehicle.
7. The determining device according to claim 6, wherein the smoothing module is specifically configured to: Filtering each initial planning track to obtain a target sampling point corresponding to the initial planning track; and optimizing the target sampling points corresponding to the initial planning track by using a pre-constructed punishment function to obtain a model output track corresponding to the model output track.
8. The determination device according to claim 6, wherein the generating module is specifically configured to: Acquiring a local map related to a target vehicle, wherein the local map comprises static and dynamic obstacle information; Based on the drivable area divided by the obstacle in the local map, making a transverse decision and a longitudinal decision; fusing the transverse decision and the longitudinal decision to generate at least one homotopy traffic decision; And calling a mathematical optimization algorithm to generate a safety bottom planning track of at least one target vehicle based on the Tonlun traffic decision.
9. An electronic device comprising a processor, a storage medium and a bus, the storage medium storing machine-readable instructions executable by the processor, the processor and the storage medium in communication over the bus when the electronic device is operating, the processor executing the machine-readable instructions to perform the steps of the determining method according to any one of claims 1 to 5.
10. A computer-readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, performs the steps of the determination method according to any of claims 1 to 5.

Description

Driving track determining method and device Technical Field The application relates to the technical field of automatic driving, in particular to a driving track determining method and device. Background With the development of artificial intelligence technology, the automatic driving field starts to widely adopt a data driving method based on machine learning/deep learning, in particular an end-to-end decision planning model. Generally, the model is trained through massive drive test data, can directly output a planning track according to perception map information, and has the advantages of light code, large planning space, good imitation of human driving behaviors and the like. However, such data-driven decision-making planning models also suffer from inherent drawbacks in that, first, their performance is highly dependent on the amount, quality and scene coverage of training data, making it difficult to effectively incorporate a priori knowledge of vehicle dynamics, traffic rules, etc. Second, in real-vehicle applications, the model may output trajectories that are abrupt in curvature, not smooth, and not in line with the vehicle dynamics constraints. Furthermore, such models may produce unreasonable plans that violate traffic rules or violate security wisdom in the face of rare or extreme scenarios where the training data coverage is inadequate. The safety, comfort and traffic efficiency of autonomous vehicles are severely affected by the adverse behavior resulting from these irrational plans. Currently, post-processing and safety enhancement technologies for output results of decision planning models are not mature. Simple trajectory smoothing methods may not be able to handle the fundamental errors of model output, but rely entirely on traditional rule methods and lose the advantages of decision-making models. Disclosure of Invention Accordingly, embodiments of the present application provide a method and apparatus for determining a driving track, so as to overcome at least one of the above-mentioned drawbacks. In a first aspect, an exemplary embodiment of the present application provides a method for determining a driving trajectory, the method comprising obtaining at least one initial planned trajectory output by a decision-making planning model for a target vehicle; Smoothing the initial planning track to obtain at least one model output track; Generating a safety bottom planning track of at least one target vehicle by a Tonlon track generation method based on the acquired local map related to the target vehicle; And carrying out arbitration evaluation on the model output track and the safety bottom planning track, and taking the evaluated optimal track as an executable track of the current target vehicle. Optionally, the step of smoothing the initial planned trajectory to obtain at least one model output trajectory includes: Filtering each initial planning track to obtain a target sampling point corresponding to the initial planning track; and optimizing the target sampling points corresponding to the initial planning track by using a pre-constructed punishment function to obtain a model output track corresponding to the model output track. Optionally, the step of generating at least one safety bottom planning track by using the homotopy track generating method based on the acquired local map related to the target vehicle includes: Acquiring a local map related to a target vehicle, wherein the local map comprises static and dynamic obstacle information; Based on the drivable area divided by the obstacle in the local map, making a transverse decision and a longitudinal decision; fusing the transverse decision and the longitudinal decision to generate at least one homotopy traffic decision; And calling a mathematical optimization algorithm to generate a safety bottom planning track of at least one target vehicle based on the Tonlun traffic decision. Optionally, the step of performing arbitration evaluation on the model output track and the safety spam planning track, and taking the evaluated optimal track as the executable track of the current target vehicle includes: Respectively inputting the model output track and the safety bottom planning track into a target evaluator to obtain evaluation scores of each model output track and each safety bottom planning track; Determining whether the track meets the execution requirement according to the evaluation score; If so, determining the track with the minimum evaluation score in the qualified tracks as the executable track of the current target vehicle. Optionally, the determining method further comprises the steps of fitting an executable track determined at the current previous sampling moment with current running data of the target vehicle if the executable track is not met, and determining the track obtained after fitting as the executable track of the current target vehicle. In a second aspect, an embodiment of the present application further prov