Reproduce “SpatialHarness: Test-Time Spatial Scaffolding for Fine Robotic 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
Jiayu Wang, Yue Yu, Bin Zhu, Zhiyao Yang, Jingjing Chen
SpatialHarness gives a frozen manipulation policy additional spatial context through two rendered views and object poses from a simulated scene synchronized with the robot. In real-robot tests, mean success across four tasks rose from 18.33% to 83.33%, with the policy and physical cameras fixed. Plug insertion rose from 26.67% to 66.67%. The comparison evaluates the full harness, not virtual views alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.