Reproduce “VerifLM: Physically Verified Image-Edit-Guided Goal Pose Generation for 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
Xuanyou Chen, Jinyu Chen, Glen Chou
VerifLM turns edited goal images into robot-usable poses by treating image-derived poses as proposals, searching uncertain pose dimensions, and checking candidates with task-specific rules in simulation. With predefined meshes, it achieved 160/216 task-planning successes across nine tasks, exceeding Goal-VLA’s 102/216 and leading on eight tasks. In mesh-free mode, it achieved 146/216 planning successes, but depended on accurate reconstruction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.