Autonomous Driving Core Tasks: Object Detection and Path Planning

Engineering notes

Key topics: autonomous driving, autonomous vehicle, path planning, object detection, reinforcement learning, deployment, perception, planning. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

With the continual improvement of artificial intelligence, sensing, and computational technologies, autonomous vehicles have become an important part of intelligent transportation systems research. Environment perception being accurate and safe trajectory creation is the basis of autonomous driving, and object detection and path planning are the two must-have basic tasks. A detailed review of two key Autonomous Driving technologies, Object Detection and Path Planning, is given by this article. On objects, it looks at the YOLO-based single stage, the Transformer-based global approach, and two-stage based or multi-modal approaches. It discusses each models principles, advantages and disadvantages for path planning, traditional algorithms, end to end deep learning, and deep reinforcement learning methods will be introduced in terms of features and problems. The study points out that YOLO has high real-time, but is lacking on robustness in some cases; the transformers model brings in global context but adds much more computational cost; two-stage, multi-modal models provide high accuracy, however, they have difficulty in deployment. Path Planning traditional algorithms are more steady than not adaptive, deep leaning algorithms are adaptive than data dependant. Future works: Lightweighting the model, Multi-modal model, Online learning, and Interpretable model to improve safety, stability and practical application.

Links and sources

• Official / arXiv page

Comments