Reproduce “How Much of Your VLA Is kNN” 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
Denis Grachev, Damien LaRocque, Omar Moured
The authors evaluate a training-free, multimodal k-nearest-neighbor policy for robot manipulation in LIBERO. It retrieves similar demonstration observations and blends their following action chunks, weighting closer matches more heavily. Across 40 tasks, it achieves 91.7% success on LIBERO and 82.4% on LIBERO-Plus, the smallest drop among evaluated policies, but this does not establish generalization beyond demonstrated behavior.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.