Reproduce “ActMem: From Latent Memory Reasoning to Grounded Robotic Action” 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

Linfeng Dong, Jianwei Niu, Jiuyi Zhao, Ruibing Feng, Xinghao Wu, Wu Yang, Shuai Tao, Wulong Liu and 1 more

ActMem uses interaction history to choose robot actions and the current view to ground motion. Its model passes contextual hidden states of the inferred subtask as a history signal, while a separate attention path supplies current visual features to the action expert. Across 26 RoboMemArena tasks, it reaches 71.1% task success and 80.8% stage completion, above SimpleMemVLA; very long histories remain a limitation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.