SkillWeaver: Agentic Exploration over Neural Interaction Skills for Scalable Robot Data Generation
A VLM agent explores simulated tasks by calling reusable RL-trained interaction skills, verifying outcomes and keeping memory in a verifier-guided tree search, to generate robot data without scripts.VLM 智能体在仿真中调用可复用的强化学习交互技能来探索任务，借助验证、反思与记忆进行验证器引导的树搜索，无需脚本即可生成机器人数据。
BibTeX
@misc{zhu2026skillweaver,
title = {{SkillWeaver: Agentic Exploration over Neural Interaction Skills for Scalable Robot Data Generation}},
author = {He Zhu and Lusen Zhao and Kwan Man Cheng and Su Li and Katerina Fragkiadaki},
year = {2026},
eprint = {2609.36171},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.36171},
}