Reproduce “CondenseVLA: Learnable History Condensation for Efficient Multi-Frame VLA” 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
Feiyang Hong, Yujie Wei, Bo Zhao, Xiu Su, Hongxun Yao, Shuo Yang
CondenseVLA compresses recent visual history into learned cues for a robot action decoder, rather than conditioning it on every cached frame. In a controlled RMBench comparison using the same 20-feature FIFO, it averages 21.6% success across nine tasks, versus 14.3% for dense-history CronusVLA. Gains are larger on M(1) than M(n) tasks, while persistent memory remains unresolved.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.