Reproduce “Joga: Learning Agile Humanoid Soccer Skills with Active Vision” 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
Avinash Narula, Khai Nguyen, Nolan Fey, Venki Parsuram, John Marangola, Martin Peticco, Pulkit Agrawal
Joga trains humanoid soccer policies in a simulator aligned with hardware perception and actuator data. It models state-dependent ball-detection errors and whole-body joint-command corrections, then finetunes dribbling in the aligned simulator. In a 135° turn test at a 1.2 m/s command, success increases from 3/5 to 5/5; each condition has five trials and success means controlling the ball without falling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.