Reproduce “NativeMEM: Native Memory Compression for Long-Horizon Robotic Manipulation” 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

Ziye Wang, Modi Shi, Chaojun Ni, Jiazhi Yang, Mengdi Li, Zhizhong Su, Tianwei Lin, Hongyang Li

NativeMEM adds visual history to a pretrained robot policy by compressing each past camera frame into one token with the policy’s vision encoder. A learned query gathers frame features, and the token joins the policy’s ordinary inputs. Across five simulated tasks, it averages 84.0% success versus 32.4% for Mem-0; results cover tested settings, not arbitrary long-horizon tasks.

Summary by alphaXiv.

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