Reproduce “Harness Robotic OS: A Unified Embodied-Agent Runtime for Closed-Loop Quadruped Inspection” 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
Yaoyuan Yan, Zhiyou Heng, Haoxiang Jie, Gang Liu, Hongjie Yan, Wei Zhou
Harness Robotic OS (HROS) organizes quadruped navigation, multimodal hazard analysis, voice interaction, memory, and enterprise reporting into a governed closed-loop runtime. In a residential inspection deployment, its Argos implementation reached all eight waypoints, maintained localization error below 10 cm, completed missions within 60 minutes, and reported hazard-recognition rates of 85–95%, though the benefits of memory, voice control, and self-evolution were not yet quantitatively evaluated.
Key insights
- The central contribution is systems integration: deterministic autonomy skills are exposed to a higher-level agent without allowing the language model to issue unvalidated low-level actuator commands.
- Three-tier memory separates immediate mission context, time-indexed experiences, and stable site knowledge, providing a proposed basis for recurring inspections and site-specific adaptation.
- Safety-gated self-evolution preserves execution traces, human feedback, provenance, regression checks, and rollback, but remains an architectural proposal until its controlled trials are reported.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.