Autonomous Driving Behavior Decision-Making and Control Command Generation Based on Domain Generalization

Engineering notes

Key topics: end-to-end autonomous driving, autonomous driving, end-to-end, carla, planning, control. 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

Traditional end-to-end autonomous driving algorithms often require training data and test data to have the same distribution, which is difficult to achieve in practical application scenarios, and often failing to achieve expected results when facing unfamiliar new environments. We propose a method based on domain invariant feature extraction, which removes factors unrelated to driving such as weather and lighting in traffic scenarios, while retaining domain invariant features closely related to driving such as pedestrians, vehicles, and roads. Then, based on the behavior decision-making model after domain generalization, we propose an autonomous driving decision-making optimization method based on hierarchical proximal policy optimization and a control command generation method, aiming to achieve decision-making and planning tasks. Finally, the experiment based on the CARLA simulation platform, proved that the algorithm proposed still exhibits excellent generalization ability from virtual scene to real scene.

Links and sources

• Official / arXiv page

Comments