Reproduce “Pixel: A Co-Designed Full-Stack Robot System for Learning General-Purpose Mobile Manipulation” 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
Tony Tao, Jason Jingzhou Liu, Ruslan Salakhutdinov, Deepak Pathak
Pixel co-designs a mobile-manipulation robot, demonstration interfaces, controller, and learning system so human and robot demonstrations can use shared task-space actions. A ground-anchored frame expresses head and hand targets together, which a whole-body controller translates into motion. Policies trained on human demonstrations averaged 82.0% task progress, versus 67.3% and 25.7% for alternatives; robot-data joint-space actions scored higher.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.