Reproduce “Adaptive Mean Flow for Responsive Closed-Loop Robot Control” 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
Aksel Vaaler, Marco Job, Christian Holden, Olav Egeland
AMF enables policies to update action sequences in one Mean Flow step, anchoring predictions to a shifted, corrupted prior sequence with more corruption farther into the future. Across 12 simulated tasks, its mixed-task average was 0.82 versus 0.66 for MF-WarmStart. On real-world Push-T, it scored 0.60 versus 0.35 for FM-8 and FM-RTC, with 14 ms inference latency; discrete gripper commands remain a limitation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.