Reproduce “Hydra: A Navigation World Action Model with Discrete Latent Planning and Continuous Flow-Matching Execution” 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
Mohammad Nazeri, Alexandyr Card, Samira Huber, Anuj Pokhrel, Yujun Wang, Ruben Hammele, Daeun Song, Sören Pirk and 1 more
Hydra plans robot navigation by searching learned movement intents in its world model, scoring candidates in a shared latent representation, then decoding the selected intent into continuous commands. In physical trials, planning took about 0.9 seconds versus over 500 seconds for NWM, with success in 10/10 unobstructed and 8/10 occluded and blind-corner trials. The comparison does not isolate discretization alone or demonstrate onboard deployment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.