Reproduce “MORE: Learning Motion-Oriented Registers for Driving World Models and Planning” 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
Yingdong Gu, Haoyu Wang, Shaocheng Yan, Zhenjun Zhao, Hao Han, Pengcheng Shi, Xianfei Li, Qinghua Tan and 4 more
MORE learns compact, motion-conditioned registers during video pretraining and uses them as shared state for future prediction and trajectory planning. Its world model predicts future registers rather than relying on dense visual features alone. On the reported NAVSIM-v2 leaderboard, MORE scores 92.5 EPDMS versus NTR’s 90.9, a 1.6-point margin; this is leaderboard performance, not a matched-input comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.