Reproduce “MotionCore: A Latent Neural Simulator for Video Generation with Rigid-Body Dynamics” 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

Zheng Wang, Zhengwei Gong, Yuanzhi Liang, Haibin Huang, Chi Zhang

MotionCore separates object motion prediction from video rendering. It uses a learned dynamics model to predict object-state trajectories, then guides a pretrained video generator with spatial control signals. On the strict81 Physics-IQ-Verified subset, it reports an ST-IoU of 0.373, compared with 0.297 for CogVideoX-5B-I2V. This result concerns a selected rigid-body benchmark and does not establish a causal simulator benefit.

Summary by alphaXiv.

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