Reproduce “VLN-Flash: Parallel Action Decoding with Adaptive Execution for Vision-and-Language Navigation” 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
Ruizhe Xie, Honghua Chen, Binglin Wang, Tong Guo, Xu Tang, Puti Yan, Shaoxiong Cai
VLN-Flash uses a masked forward pass to score candidate action sequences together and select both a plan and when to observe again. Tested with Uni-NaVid and StreamVLN, it reduces complete-reply latency from 150.8 to 54.9 ms and from 182.8 to 71.1 ms, respectively, while improving navigation outcomes on four policy–dataset pairs. The real-robot comparison covered 60 trials in three environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.