Reproduce “CST-WM: A Causally Structured World Model for Embodied Visual Tracking” 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
Junyi Hu, Shuaihang Yuan, Jiazhao Liang, Yi Fang
CST-WM plans robot actions for following and re-acquiring a person by imagining future views. Its action mask prevents actions from directly changing predicted target evidence, while allowing them to affect evidence through robot motion and camera views. In Habitat 3.0, it achieves 0.68 agreement in candidate-action rankings, versus 0.47 for adapted NWM; this is evidence from the tested setup, not proof of hidden-state recovery.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.