Reproduce “Xiaomi Auto World Model 2.0: Causal World Models for Real-World Driving and Embodied Intelligence” 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

Zhenxin Zhu, Zhijian Huang, Cheng Chi, Yuru Wang, Jianhua Wu, Jiayi Guan, Yonggang Li, Mingfei Tu and 15 more

Xiaomi Auto World Model 2.0 predicts controlled future scenes across camera views, using a shared streaming video generator and reconstructed 3D context to anchor geometry. Replacing temporal softmax attention with gated linear attention reduces latency from 2.299 to 1.742 seconds per chunk, while FID shifts from 37.24 to 38.08. Simulated policy results support utility, not real-road safety.

Summary by alphaXiv.

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