Reproduce “PLaCA: Learning Patch-Level Future Latent Change for Action Generation” 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
Zhan Yi, Yuchen Wang, Chao Li, Lei Hong, Boyuan Yang
PLaCA predicts spatially located patch-level changes from current to future observations and feeds them into action generation rather than reconstructing future images. On LIBERO-Long, it achieves 79.0% task success versus 74.0% for Action-only; on CALVIN’s ABC-to-D setting, its average sequence length is 3.564 versus 2.784. Ablations support change prediction and explicit action conditioning in simulation tests, not gains beyond them.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.