Reproduce “Physical Experience Builds Continual Spatial World Models” 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
Yuhang Sun
The authors introduce Spatial Change Memory (SCM), which pools evidence about relations between route-relevant changes across locations to predict transitions rather than relying on separate histories. In a controlled maze, shared SCM reaches 0.898 accuracy on a withheld transition, versus about 0.566 for comparison methods. Its advantage disappears when changes are unpredictable; physical-model refinement assumes a known equation family.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.