Cross-Controller Transfer via In-Context Behavior Adaptation
Pools robot data across six controller types by showing the policy examples of commanded actions and the motions they caused. Reports 33 points better average success than naive pooling in real experiments.跨六类控制器共享机器人数据：把“指令动作与实际运动”的交互样例编码为上下文，让同一策略在线适应。真实实验报告平均成功率比直接混合数据高 33 个百分点。
BibTeX
@misc{yang2026cross,
title = {{Cross-Controller Transfer via In-Context Behavior Adaptation}},
author = {Jonathan Yang and Chelsea Finn and Dorsa Sadigh},
year = {2026},
eprint = {2610.39280},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.39280},
}