Automatic driving decision method and system based on maximum entropy layered reinforcement learning

By constructing a multi-layer network system using the maximum entropy hierarchical reinforcement learning method, the problem of single strategy for autonomous vehicles in complex environments is solved, and efficient autonomous driving decision-making for autonomous vehicles under different road conditions is achieved.

CN116257065BActive Publication Date: 2026-06-09NANJING UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
NANJING UNIV
Filing Date
2023-04-12
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing reinforcement learning algorithms struggle to adapt to complex and ever-changing road conditions in autonomous driving environments, resulting in a limited range of driving strategies for driverless vehicles and impacting autonomous driving performance.

Method used

We employ a maximum entropy-based hierarchical reinforcement learning approach to construct a driving policy value network, a driving policy selection policy network, and a driving policy group network. Through end-to-end training, the autonomous vehicle selects the optimal driving policy under different road conditions, and information entropy is used as an internal reward to improve learning efficiency.

Benefits of technology

Autonomous vehicles can efficiently learn multiple driving strategies, reduce training costs, adapt to complex autonomous driving environments, and improve the robustness and efficiency of autonomous driving decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116257065B_ABST
    Figure CN116257065B_ABST
Patent Text Reader

Abstract

The application discloses an automatic driving decision-making method and system based on maximum entropy layered reinforcement learning. With the help of layered reinforcement learning, multiple driving strategies are learned by using limited data when learning an automatic driving strategy. Through a maximum entropy reinforcement learning method, a driving strategy selection strategy is synchronously learned, so that an unmanned vehicle can select an optimal strategy from multiple learned driving strategies according to an automatic driving scene to make an automatic driving decision, can adapt to more complex and changeable automatic driving scenes, and has good generalization and practicability.
Need to check novelty before this filing date? Find Prior Art