Reproduce “DynaHarness: A Dynamic Physical Harness for Self-Evolving Robot Agents” 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

Haoyuan Deng, Jiebin Liu, Tengxiao Zhang, Langning Yan, Hongye Cao, Ziwei Wang

DynaHarness coordinates long-horizon manipulation by letting a slow Qwen3-VL-4B planner propose capabilities while a controller grounds, monitors, refuses, substitutes, or replans commands under one contract. On 800 development episodes, dynamic execution achieved 74.0% success versus 63.9% for replanning. After selection, achieved 75.2% versus 17.5% for frozen π0.5 on new states within the same 40 cells, not unseen tasks or domains.

Summary by alphaXiv.

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