arXiv Robotics — research abstracts· Changxu Wei, Jingyi Zhang, Zhiyu Zheng, Tianyi Lin, Yixian Zhang, Yunpeng Qing, Jingyuan Liu, Zhi Hou, Jiaxing Huang·· 9 hours agoEditorial score63
StateSkill-Agent: An Effective and Efficient Embodied Agent via State-conditioned Skill
StateSkill-Agent: An Effective and Efficient Embodied Agent via State-conditioned Skill
Summary
This research proposes StateSkill, a new skill representation for embodied agents that models interactions as finite-state machines. The method integrates skill evolution and model optimization through StateGRPO, enabling co-evolution of skills and model capabilities. Extensive experiments validate the effectiveness of the approach across various embodied tasks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org