SparseAD: Sparse Query-Centric Paradigm for Efficient End-to-End Autonomous Driving

• DOI: 10.48550/arxiv.2404.06892
• Publication Date: 2024-04-10
• Authors: D.Y. Zhang; Guoan Wang; Runwen Zhu; Jianbo Zhao; Xiwu Chen; Siyu Zhang; Jiahao Gong; Qibin Zhou; Wenyuan Zhang; Ningzi Wang; Feiyang Tan; Hangning Zhou; Ziyao Xu; Haotian Yao; Chengqi Zhang; Xiaojun Liu; Xiaoguang Di; Bin Li
• Keywords: End-to-end principle

End-to-End paradigms use a unified framework to implement multi-tasks in an autonomous driving system. Despite simplicity and clarity, the performance of end-to-end methods on sub-tasks is still far behind single-task methods. Meanwhile, widely used dense BEV features previous make it costly extend more modalities or tasks. In this paper, we propose Sparse query-centric paradigm for Autonomous Driving (SparseAD), where sparse queries completely represent whole scenario across space, time tasks without any representation. Concretely, design architecture perception including detection, tracking, online mapping. Moreover, revisit motion prediction planning, devise justifiable planner framework. On challenging nuScenes dataset, SparseAD achieves SOTA full-task among significantly narrows gap between Codes will be released soon.