Reproduce “Kinematic Interface for the Wild: Modular Bimanual Loco-Manipulation Capture from 360$^{\circ}$ Cameras Alone” 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
Benjamin Yang, Weiying Wang, Shenggao Li, Keming Yan, Sasha Wilkinson, Zelin Wang, Yip Fun Yeung, Lingfeng Sun
KIWI uses two lenses on each wrist-worn 360-degree camera: one records manipulation, while the other maps the room. Offline fusion of visual observations and IMU data produces shared-frame bimanual poses. Across six recordings, it supported more than 99.9% of 246,116 query frames, versus 75.2% for a front-only baseline. This measures visual support, not pose error, robot replay success, or performance beyond these recordings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.