Reproduce “HeMeR: Heterogeneous Memory Reconciliation for Embodied Agents via Structured KV Reuse” 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

Saehun Chun, Wonje Choi, Jinwoo Jang, Taeyoon Kwack, Honguk Woo

HeMeR helps embodied agents update changing task memories while preserving compatible cached computation. It tracks memory chunks and recomputes affected content, including dependent chunks when procedural adapters change. In two single-agent real-world tasks, using Qwen3-VL-4B-Instruct, it reached 83.20% goal-conditioned success versus MemRL’s 64.28%, with inference latency of 1.87 seconds per decision step versus 2.94 seconds; concurrent multi-agent updates remain unaddressed.

Summary by alphaXiv.

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