An automatic driving decision planning method based on a time sequence graph neural network and reinforcement learning

By constructing dynamic traffic scene graphs and temporal graph neural networks, and combining deep reinforcement learning and safety constraint mechanisms, the modeling problem of multi-agent interaction and temporal dependency relationships in autonomous driving is solved, realizing safe, efficient, and interpretable decision-making and planning, and improving the performance and reliability of autonomous driving systems.

CN122116673APending Publication Date: 2026-05-29JIANGSU HOPERUN SOFTWARE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
JIANGSU HOPERUN SOFTWARE CO LTD
Filing Date
2026-01-26
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing autonomous driving decision-making and planning methods suffer from several problems when dealing with complex scenarios such as multi-vehicle interaction and pedestrian avoidance, including insufficient modeling of multi-agent interactions, inadequate capture of temporal dependencies, limited scene representation capabilities, difficulty in balancing safety and exploratory capabilities, insufficient interpretability, and low sample efficiency.

Method used

A dynamic traffic scene graph is constructed, and interaction features are extracted using a temporal graph neural network. Deep reinforcement learning is combined with this feature for policy optimization. Safety constraints and interpretability mechanisms are introduced. Graph convolution and temporal modeling are used to capture the interaction relationships between multiple agents. A deep deterministic policy gradient algorithm is used to optimize the decision-making strategy. Safety constraints and attention visualization techniques are used to improve the safety and interpretability of the decision-making.

Benefits of technology

It enables safe, efficient, and interpretable autonomous driving decision-making and planning in complex traffic environments, improving decision quality and safety, reducing collision rate, increasing sample efficiency, and meeting real-time requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122116673A_ABST
    Figure CN122116673A_ABST
Patent Text Reader

Abstract

The application discloses an automatic driving decision planning method based on a time sequence graph neural network and reinforcement learning, and specifically comprises the following steps: S1, dynamic traffic scene graph construction; S2, time sequence graph neural network feature extraction; S3, deep reinforcement learning strategy optimization; S4, safety constraint mechanism; and S5, explainability mechanism; the application represents multi-agent interaction by constructing a dynamic traffic scene graph, extracts historical time sequence and spatial topology features by using a time sequence graph neural network, optimizes a decision strategy in combination with a deep reinforcement learning algorithm, and introduces a safety constraint mechanism and attention visualization technology, so that safe, efficient and explainable automatic driving decision planning in a complex traffic environment is realized.
Need to check novelty before this filing date? Find Prior Art