Reproduce “DreamFAR: Real-Time and Long-Horizon Interaction with Diverse Objects” 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
Yiming Dou, Botao He, Antonio Loquercio, Andrew Owens, Jitendra Malik, Sumedh Anand Sontakke, Ruoshi Liu
DreamFAR predicts robot action-conditioned futures and uses predicted contact transitions to select scene snapshots for retrieval alongside recent frames. In one held-out whiteboard rollout, its scene remains visually similar after 60 seconds of prediction. Across 1,000 held-out ABC episodes, it averages 16.4 dB PSNR versus 15.8 for Cosmos-Predict 2.5. A separate bottle-task ablation favors contact-selected memory; evidence is limited to the reported tests and one robot embodiment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.