Reproduce “ViSkill: Reinforcing VLM Agents with Evolving Visual-Native Skills” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hongxing Li, Dingming Li, Yixin Li, Yong Du, Wenqi Zhang, Weiming Lu, Jun Xiao, Yueting Zhuang and 1 more
ViSkill helps vision-language agents reuse experience on spatial tasks by retrieving visual skill cards for similar layouts. Each card combines annotated frames from an interaction with a distilled strategy, and eligible policy successes can be added to the library. In evaluations, ViSkill reaches 0.89 overall versus AtlasVA’s 0.87 under a shorter training horizon; AtlasVA ties its PrimitiveSkill average.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.