Reproduce “LTM: Latent Transition Based In-Context Memory for Vision-Language-Action Models” 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
Qixin Sun, WenLong Gao, Xiansheng Chen, Zhongyuan Wang, Si Liu, Pengwei Wang
LTM-π gives a robot an ordered memory of compact codes for changes between consecutive observations, alongside its current view, to support history-dependent control. On RoboMME, DreamDojo LTM-π averages 58.54% success, 1.71 percentage points above T²Mem, while fixed random memory scores 44.50% versus 57.67% for the full model. The comparison uses prior reported baselines and is not a matched-budget test.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.