RoboDrive 挑战赛：在任意时间、任意地点、任意条件下驾驶
计算机视觉与模式识别
2024-05-31 v2 机器人学
摘要
在自动驾驶领域，分布外条件下的鲁棒感知对于车辆的安全部署至关重要。恶劣天气、传感器故障和环境不可预测性等挑战会严重影响自动驾驶系统的性能。2024 年 RoboDrive 挑战赛旨在推动能够承受并适应这些真实世界变化的驾驶感知技术的发展。本次竞赛聚焦于四项关键任务——BEV 检测、地图分割、语义占据预测和多视图深度估计——为创新和增强系统抵御典型与非典型扰动的韧性设立了挑战。今年的挑战赛包含五个不同的赛道，吸引了来自 11 个国家 93 个机构的 140 支注册队伍，通过我们的服务器评估了近千次提交。竞赛最终产生了 15 个表现最优的解决方案，这些方案引入了一系列创新方法，包括先进的数据增强、多传感器融合、用于纠错的自监督学习以及增强传感器鲁棒性的新算法策略。这些贡献显著推进了当前技术的前沿，特别是在处理传感器不一致性和环境多变性方面。参赛者通过协作努力，突破了现有技术的边界，展示了其在真实世界场景中的潜力。广泛的评估和分析提供了关于这些解决方案有效性的见解，突出了提升驾驶感知系统鲁棒性的关键趋势和成功策略。本次挑战赛在该领域树立了新的基准，提供了丰富的技术库，有望指导该领域的未来研究。
引用
@article{arxiv.2405.08816,
title = {The RoboDrive Challenge: Drive Anytime Anywhere in Any Condition},
author = {Lingdong Kong and Shaoyuan Xie and Hanjiang Hu and Yaru Niu and Wei Tsang Ooi and Benoit R. Cottereau and Lai Xing Ng and Yuexin Ma and Wenwei Zhang and Liang Pan and Kai Chen and Ziwei Liu and Weichao Qiu and Wei Zhang and Xu Cao and Hao Lu and Ying-Cong Chen and Caixin Kang and Xinning Zhou and Chengyang Ying and Wentao Shang and Xingxing Wei and Yinpeng Dong and Bo Yang and Shengyin Jiang and Zeliang Ma and Dengyi Ji and Haiwen Li and Xingliang Huang and Yu Tian and Genghua Kou and Fan Jia and Yingfei Liu and Tiancai Wang and Ying Li and Xiaoshuai Hao and Yifan Yang and Hui Zhang and Mengchuan Wei and Yi Zhou and Haimei Zhao and Jing Zhang and Jinke Li and Xiao He and Xiaoqiang Cheng and Bingyang Zhang and Lirong Zhao and Dianlei Ding and Fangsheng Liu and Yixiang Yan and Hongming Wang and Nanfei Ye and Lun Luo and Yubo Tian and Yiwei Zuo and Zhe Cao and Yi Ren and Yunfan Li and Wenjie Liu and Xun Wu and Yifan Mao and Ming Li and Jian Liu and Jiayang Liu and Zihan Qin and Cunxi Chu and Jialei Xu and Wenbo Zhao and Junjun Jiang and Xianming Liu and Ziyan Wang and Chiwei Li and Shilong Li and Chendong Yuan and Songyue Yang and Wentao Liu and Peng Chen and Bin Zhou and Yubo Wang and Chi Zhang and Jianhang Sun and Hai Chen and Xiao Yang and Lizhong Wang and Dongyi Fu and Yongchun Lin and Huitong Yang and Haoang Li and Yadan Luo and Xianjing Cheng and Yong Xu},
journal= {arXiv preprint arXiv:2405.08816},
year = {2024}
}
备注
ICRA 2024; 32 pages, 24 figures, 5 tables; Code at https://robodrive-24.github.io/