Surround Vision Autonomous Driving Based on Swin-Transformer and Trajectory Prediction Network

Engineering notes

Key topics: end-to-end autonomous driving, autonomous driving, bev, end-to-end, trajectory prediction, carla, perception, prediction, 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

Vision-based autonomous driving has attracted increasing attention due to its low cost, fast inference speed, broad applicability, and strong consistency with human driving perception. To further exploit the rich spatiotemporal information provided by surround-view visual inputs, this paper proposes an end-to-end autonomous driving framework based on surround-view vision. The proposed method leverages the sliding-window mechanism of the Swin Transformer to extract spatiotemporal features in the bird's-eye-view (BEV) space. Furthermore, a conditionally variable trajectory prediction network is introduced to model the probabilistic distributions of surrounding agents and predict their future trajectories. The predicted trajectories are subsequently mapped to ego-vehicle control commands via a multi-layer perceptron. Extensive experiments conducted in the CARLA simulator demonstrate the feasibility and effectiveness of the proposed vision-based autonomous driving approach, achieving competitive performance.

Links and sources

• Official / arXiv page
• PDF from original source

Comments