Reproduce “Post-hoc Epistemic Uncertainty for Vision-Language-Action Policies” 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
Johannes Keller, Albrecht M. Bloße, Adrian Lindenmeyer, Thomas Neumuth, Daniel Schneider
The authors introduce post-hoc Gaussian-process variance heads to estimate demonstration coverage for a robot policy. The heads compare policy representations with those of training demonstrations, require no outcome labels or gradient updates, and leave actions unchanged. In tested LIBERO-10 acquisition, scores achieved performance similar to an ensemble-guided method. Referral rankings improved under tested perturbations, but the measure reflects embedding-space coverage, not whether unfamiliar inputs cause failure.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.