Reproduce “Learning Options for Compositional Motor Control with Adapter Banks” 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
Sreejan Kumar, Marcelo Mattar, Lea Duncker
CABRA learns reusable movement components from demonstrations for a recurrent controller. It uses ten residual adapters and a selector to recombine them. With the controller frozen, the authors optimize a selection policy for a novel two-petal arm movement. The student reaches about 0.005 meters L1 tracking error, versus about 0.04 for the teacher; this is target-specific, not zero-shot transfer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.