Reproduce “Knot So Simple: A Minimalistic Environment for Spatial Reasoning” 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
Zizhao Chen, Yoav Artzi
KnotGym tests whether agents can manipulate simulated ropes by linking images of a rope and a goal knot to continuous force actions. It defines success by matching the knot’s Gauss code, not its exact shape. After 1 million environment steps, DreamerV3 reaches 93.3% training success on two-crossing unknot, versus 11.1% for random actions; harder goal-conditioned tasks generalize less reliably.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.