Reproduce “Dexterous In-Hand Object Pose Reaching with Learned Safety Filters” 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
Ian Chuang, Hiya Shah, Vamsi Chilamkur, Iman Soltani, Claire Tomlin
DexSafe trains policies for dexterous in-hand pose reaching, moving held objects to successive positions and orientations. It transfers simulation-trained policies to a real hand, then adds a separately trained, adjustable residual safety filter. In real-robot state-policy tests, filtering reduced drops from 16 to 4 per 1,200 goals, while coarse success fell from 95.2% to 91.5%; policies and filters are trained per object.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.