Reproduce “Seeing Through the Illusion: Test-Time Scaling for World Action Models” 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
Cheikh Ahmed, Mahdi Mostajabdaveh, Milad Yazdani, Zirui Zhou
WISE selects among paired robot action–future proposals before execution, scoring action suitability, internal consistency, and task progress. With four candidates, RoboLab success was 24.0% across 50 episodes, versus 6.0% for single-rollout execution and 20.0% for selectors. On LIBERO, its mean was 78.3% versus 75.1%. These results cover two settings; internal consistency does not guarantee imagined futures match reality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.