Reproduce “MemTransfer: Benchmarking Memory Beyond Matched Experience in Embodied Decision-Making” 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
Haiming Tang, Xianjie Dai, Gujie Shao, Zuyi Guo, Jingguang Li, Kailang Ma, Yihong Tang, Heye Huang
MemTransfer tests whether warehouse agents can reuse demonstrations when starts, routes, or histories change. It compares six memory configurations under a frozen vision-language policy across 100 cases, varying starting pose, route availability, and history relevance. With one demonstration per task, Episodic memory achieves 100.0% success at its demonstration start but 51.7% at a new start; matched replay does not establish transfer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.