TY  - UNPB
TI  - In-Context Learning for Robots: Methods and Applications
AU  - Huang, Haojian
AU  - Li, Zexi
AU  - Guo, Junhao
AU  - Zhang, Yehang
AU  - Peng, Wenxuan
AU  - Zhou, Bohan
AU  - Ruan, Weilin
AU  - Wu, Leyi
AU  - Wang, Chenxu
AU  - Su, Jianchong
AU  - Xie, Binghui
AU  - Chen, Wosong
AU  - Xu, Yingjie
AU  - Zhou, Tianhao
AU  - Chen, Suzeyu
AU  - Zhao, Pukun
AU  - He, Jiaqi
AU  - Li, Xinyi
AU  - Li, Runze
AU  - Dong, Peiran
AU  - Dang, Shaoxiang
AU  - Huang, Jing
AU  - Chen, Yingbing
AU  - Chang, Yifan
AU  - Zhang, Tianyi
AU  - Deng, Shiyuan
AU  - Wang, Haozhi
AU  - Wei, Yangkai
AU  - Li, Wenqian
AU  - Yang, Han
AU  - Zhou, Kaiwen
AU  - Liu, Huaping
AU  - Cheng, James
AU  - Shao, Rui
AU  - Wang, Donglin
AU  - Jin, Yaochu
AU  - Hao, Jianye
AU  - Chen, Ying-Cong
AU  - Li, Yinchuan
PY  - 2026
DA  - 2026/09/28
DB  - arXiv
M3  - Preprint
LA  - English
UR  - https://arxiv.org/abs/2609.36012
L1  - https://jethrojames.github.io/awesome-robots-icl/assets/robot-icl-survey.pdf
N1  - arXiv:2609.36012; 100 pages
AB  - General-purpose robots must infer what a new task requires and translate that understanding into appropriate physical action. In-context learning (ICL) for robots supports this process by using demonstrations and interaction to direct existing competence with neural parameters held fixed during deployment. We organize this literature review around the interfaces connecting contextual evidence to execution, distinguishing four families: context-conditioned policies, geometric demonstration transfer, world-model-based control, and skill- and agent-based execution. Comparing these interfaces clarifies their transfer assumptions and the roles of training, correspondence, and memory in making context useful. Across manipulation and navigation, we examine how these mechanisms preserve taught requirements as objects, environments, and execution conditions change. This analysis links method design to evaluation practices that distinguish responsiveness to teaching, physical transfer, and benefits from retained experience. The resulting agenda connects compositional task acquisition and faithful transfer with physical recursive self-improvement, in which experience improves the ability to learn subsequent tasks.
ER  -