Reproduce “Map2Vid: Embodied Navigation Video Generation with Human-Friendly Motion Control” 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
Yulin Dai, Yuqian Fu, Tianwen Qian, Zhen Xing, Bojia Zi, Xiaoling Wang
Map2Vid introduces map-centered control for generating egocentric navigation video from bird’s-eye-view maps. Its model encodes the map alongside camera motion and trains with progressively masked auxiliary inputs. In a Wan2.2-Camera ablation, map-only conditioning achieved 51.3% SfM Success Rate, versus 33.3% for trajectory-only. The benchmark is simulation-based, and this result does not establish real-world robot navigation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.