无人驾驶CVPR 20252025
UniAD++: Unified Autonomous Driving with World Models
Zhenyu Li, Wenhai Wang, Jifeng Dai, Yu QiaoShanghai AI Laboratory, CUHK
摘要
We present UniAD++, a unified autonomous driving framework that integrates perception, prediction, and planning with world models. UniAD++ achieves end-to-end optimization across all driving modules, demonstrating superior performance on the nuScenes benchmark with improved safety metrics.
unified drivingworld modelend-to-end optimizationperception-prediction-planningsafety
技术细节
模型骨架
Unified Transformer + World Model Head
编码器
BEVFormer Encoder + Temporal Module
解码器
Multi-Task Decoder: Tracking+Map+Motion+Planning (12 layers)
