Reproduce “Seeing What Execution Needs: Benchmarking Task-Conditioned Actionable Spatial Belief for 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
Yina Jian, Mingkun Xu
The authors evaluate command-conditioned view selection for robotic manipulation in a simulated coffee benchmark. The selector links functional evidence to the current command, ranks candidate views using a belief ledger and public metadata, and passes observations to a shared scripted skill. It reaches 98.8% success at H=5; without command binding, same-object functional-OOD success falls to 19.7%, under hidden postcondition checks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.