Vehicle-to-everything decision optimization and cloud control based on deep reinforcement learning
Abstract To address the challenges of decision optimization and road segment hazard assessment within complex traffic environments, and to enhance the safety and responsiveness of autonomous driving, a Vehicle-to-Everything (V2X) decision framework is proposed. This framework is structured into thre...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-08-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-12772-3 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract To address the challenges of decision optimization and road segment hazard assessment within complex traffic environments, and to enhance the safety and responsiveness of autonomous driving, a Vehicle-to-Everything (V2X) decision framework is proposed. This framework is structured into three modules: vehicle perception, decision-making, and execution. The vehicle perception module integrates sensor fusion techniques to capture real-time environmental data, employing deep neural networks to extract essential information. In the decision-making module, deep reinforcement learning algorithms are applied to optimize decision processes by maximizing expected rewards. Meanwhile, the road segment hazard classification module, utilizing both historical traffic data and real-time perception information, adopts a hazard evaluation model to classify road conditions automatically, providing real-time feedback to guide vehicle decision-making. Furthermore, an autonomous driving cloud control platform is designed, augmenting decision-making capabilities through centralized computing resources, enabling large-scale data analysis, and facilitating collaborative optimization. Experimental evaluations conducted within simulation environments and utilizing the KITTI dataset demonstrate that the proposed V2X decision optimization method substantially outperforms conventional decision algorithms. Vehicle decision accuracy increased by 9.0%, rising from 89.2 to 98.2%. Additionally, the response time of the cloud control system decreased from 178 ms to 127 ms, marking a reduction of 28.7%, which significantly enhances decision efficiency and real-time performance. The introduction of the road segment hazard classification model also results in a hazard assessment accuracy of 99.5%, maintaining over 95% accuracy even in high-density traffic and complex road conditions, thus illustrating strong adaptability. The results highlight the effectiveness of the proposed V2X decision optimization framework and cloud control platform in enhancing the decision quality and safety of autonomous driving systems. |
|---|---|
| ISSN: | 2045-2322 |