Reproduce “SWIFT: Slow World, Instant Freshness, and Timely Action for Asynchronous Autonomous Driving” 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

Jiazhe Wang, Tianyi Yang, Ruihang Li, Fan Lu, Xianming Zeng, Jianyun Xu, Xingliang Liu, Lei Yang

SWIFT runs a world model asynchronously, combining its latest completed context with fresh camera images through a faster action model at each decision. On NAVSIM, it scores 94.2 PDMS with 1.0-second-old context and 94.0 with 1.5-second-old context, at 198.4 ms, versus 93.9 PDMS at 927.6 ms synchronously. This deployment comparison uses two GPUs and does not establish on-vehicle performance.

Summary by alphaXiv.

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