arXiv Robotics — research abstracts· ZhiYuan Feng, Zhuoshi Pan, Zhaolu Kang, Zhongxia Chen, Shuo Yang, Sen Ye, Heng Wang, Han Li·· 11 hours agoEditorial score65
EnactOpt: Diagnosis-Guided Harness Evolution for Embodied Agents
EnactOpt: Diagnosis-Guided Harness Evolution for Embodied Agents
Summary
EnactOpt is a method that uses limited execution feedback to jointly evolve reusable textual skills and context-construction code for embodied agents, without updating model parameters. It improves task success by approximately 10-27 percentage points over initial harnesses and outperforms several baselines across benchmarks like VLABench, ALFWorld, and EAI.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org