Layered vehicle driving decision-making method based on multiple time scales

By adopting a multi-time scale stratified vehicle driving decision-making method in autonomous vehicles, combined with deep stratified reinforcement learning decision model and TD3/OC algorithm, the problem of degraded decision performance of autonomous vehicles in dynamic traffic scenarios is solved, and more efficient and reliable driving decisions are achieved.

CN119749589AActive Publication Date: 2025-04-04NINGBO UNIV
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411733247.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-29
Publication Date
2025-04-04
Estimated Expiration
2044-11-29

AI Technical Summary

Technical Problem

When existing autonomous vehicles face dynamic and uncertain traffic scenarios, it is difficult for them to make timely and accurate driving decisions, resulting in a degradation in decision-making performance and low system stability.

Method used

A layered vehicle driving decision-making method based on multi-time scales is adopted. By constructing a deep hierarchical reinforcement learning decision model, combining high-level models and low-level models, the high-level models are used to plan driving behaviors, the low-level models are used for attitude control, and trained through TD3 and OC algorithms to ensure the stability of the decision system.

Benefits of technology

It improves the adaptability of autonomous vehicles to the environment and the reliability of decision-making, realizes smarter and safer driving decisions, and enhances the stability and driving efficiency of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119749589A_ABST
    Figure CN119749589A_ABST
Patent Text Reader

Abstract

The invention relates to a hierarchical vehicle driving decision-making method based on multiple time scales, which adopts a deep hierarchical reinforcement learning decision-making model formed by combining a high-level model and a low-level model, macroscopically plans the driving behavior of a target vehicle by the high-level model according to environment observation information, then selects a corresponding low-level model based on the driving behavior, and finally determines the driving behavior of the target vehicle according to the selected low-level model. The vehicle attitude is adjusted according to the driving behavior decision and environment observation information, a complex driving decision task is decomposed into diversified driving behaviors to adapt to different traffic road environments, hierarchical control of the automatic driving vehicle is achieved, and the diversified requirements of automatic driving on safety and efficiency are better met; meanwhile, a training mode of pre-training the low-level model and then combining the high-level model and the low-level model is adopted, so that the problem of option collapse occurring in the combined training process of the high-level model and the low-level model is effectively avoided, and the stability of the whole decision-making system is ensured.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Automatic driving vehicle lane keeping method based on TD3 algorithm improved by exploration strategy

    CN114997048A

  • Automatic driving lane changing decision control method based on rule fusion reinforcement learning

    CN115257745A

  • Behavior decision-making and planning method for autonomous vehicle

    CN117284327A

  • Automatic driving vehicle trajectory planning method and device based on hierarchical reinforcement learning

    CN118243125A

  • Automatic driving non-signalized intersection decision generation method based on hierarchical reinforcement learning

    CN118790290A