RoboSense挑战赛：感知万物、导航各处、跨平台自适应
机器人学
2026-01-09 v1
摘要
自主系统越来越多地部署在从城市街道到空中及室内空间等开放且动态的环境中，在这些环境中，感知模型必须在传感器噪声、环境变化和平台切换下保持可靠性。然而，即使是最先进的方法在未见过的条件下也经常发生性能退化，这凸显了对鲁棒且具泛化能力的机器人感知的需求。RoboSense 2025挑战赛旨在提升机器人感知在多种感知场景下的鲁棒性与自适应能力。该挑战赛统合了五个互补的研究赛道，涵盖基于语言锚定的决策、符合社会规范的导航、传感器配置泛化、跨视角与跨模态对应以及跨平台3D感知。这些任务共同构成了一个综合基准，用于评估在领域偏移、传感器故障和平台差异下真实世界感知的可靠性。RoboSense 2025提供了标准化数据集、基线模型和统一评估协议，从而实现对鲁棒感知方法的大规模且可复现的比较。本次挑战赛吸引了来自16个国家85个机构的143支队伍参加，反映了社区的广泛参与。通过汇总23个获胜方案的见解，本报告重点介绍了所有赛道中涌现的方法论趋势、共同的设计原则以及开放性挑战，标志着向构建能够在真实世界环境中可靠感知、鲁棒行动并跨平台自适应的机器人迈出了一步。
引用
@article{arxiv.2601.05014,
title = {The RoboSense Challenge: Sense Anything, Navigate Anywhere, Adapt Across Platforms},
author = {Lingdong Kong and Shaoyuan Xie and Zeying Gong and Ye Li and Meng Chu and Ao Liang and Yuhao Dong and Tianshuai Hu and Ronghe Qiu and Rong Li and Hanjiang Hu and Dongyue Lu and Wei Yin and Wenhao Ding and Linfeng Li and Hang Song and Wenwei Zhang and Yuexin Ma and Junwei Liang and Zhedong Zheng and Lai Xing Ng and Benoit R. Cottereau and Wei Tsang Ooi and Ziwei Liu and Zhanpeng Zhang and Weichao Qiu and Wei Zhang and Ji Ao and Jiangpeng Zheng and Siyu Wang and Guang Yang and Zihao Zhang and Yu Zhong and Enzhu Gao and Xinhan Zheng and Xueting Wang and Shouming Li and Yunkai Gao and Siming Lan and Mingfei Han and Xing Hu and Dusan Malic and Christian Fruhwirth-Reisinger and Alexander Prutsch and Wei Lin and Samuel Schulter and Horst Possegger and Linfeng Li and Jian Zhao and Zepeng Yang and Yuhang Song and Bojun Lin and Tianle Zhang and Yuchen Yuan and Chi Zhang and Xuelong Li and Youngseok Kim and Sihwan Hwang and Hyeonjun Jeong and Aodi Wu and Xubo Luo and Erjia Xiao and Lingfeng Zhang and Yingbo Tang and Hao Cheng and Renjing Xu and Wenbo Ding and Lei Zhou and Long Chen and Hangjun Ye and Xiaoshuai Hao and Shuangzhi Li and Junlong Shen and Xingyu Li and Hao Ruan and Jinliang Lin and Zhiming Luo and Yu Zang and Cheng Wang and Hanshi Wang and Xijie Gong and Yixiang Yang and Qianli Ma and Zhipeng Zhang and Wenxiang Shi and Jingmeng Zhou and Weijun Zeng and Kexin Xu and Yuchen Zhang and Haoxiang Fu and Ruibin Hu and Yanbiao Ma and Xiyan Feng and Wenbo Zhang and Lu Zhang and Yunzhi Zhuge and Huchuan Lu and You He and Seungjun Yu and Junsung Park and Youngsun Lim and Hyunjung Shim and Faduo Liang and Zihang Wang and Yiming Peng and Guanyu Zong and Xu Li and Binghao Wang and Hao Wei and Yongxin Ma and Yunke Shi and Shuaipeng Liu and Dong Kong and Yongchun Lin and Huitong Yang and Liang Lei and Haoang Li and Xinliang Zhang and Zhiyong Wang and Xiaofeng Wang and Yuxia Fu and Yadan Luo and Djamahl Etchegaray and Yang Li and Congfei Li and Yuxiang Sun and Wenkai Zhu and Wang Xu and Linru Li and Longjie Liao and Jun Yan and Benwu Wang and Xueliang Ren and Xiaoyu Yue and Jixian Zheng and Jinfeng Wu and Shurui Qin and Wei Cong and Yao He},
journal= {arXiv preprint arXiv:2601.05014},
year = {2026}
}
备注
Official IROS 2025 RoboSense Challenge Report; 51 pages, 37 figures, 5 tables; Competition Website at https://robosense2025.github.io/