Reproduce “JointGV: Structure-Aware Correction of Visual World Models with Parallel Graph Dynamics for Long-Horizon Modeling” 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

Tian Lyu, Yichen Jiang, Ziyuan Cheng, Yunda Yan, Xiangyu Pan, Junjie Jiang, Jialin Yu, Boyu Ning and 2 more

JointGV adds a task-defined scene graph alongside an existing visual world model. It predicts graph motion and positions, then uses them to make a bounded correction to visual predictions. In simulated pendulum control, an actor trained on corrected DINO-WM rollouts completed 46 of 50 six-second episodes, while one trained on original rollouts completed none; this does not demonstrate physical balancing.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.