Reproduce “Pros4D: Modeling Evolving 4D Worlds in Latent Space for Dynamic 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
Rongxin Gao, Dongxuan Liu, Chu Li, Chengyang Ji, Jialing Huang, Runyu Chen, Kairun Wen, Chenxin Li and 7 more
Pros4D represents observed motion in ordered continuous latent states and autoregressively generates future states, trained with language, visual-semantic, and geometric supervision. At inference, it uses only the observed video prefix and question, without external geometry calls. On Pros4D-Bench, Pros4D-8B reaches 59.5% future accuracy versus 40.7% for Qwen3-VL-8B, but this benchmark result does not establish arbitrary future prediction or embodied success.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.