Reproduce “OpenGo: An OpenClaw-Based Robotic Dog with Real-Time Skill Switching” 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

Hanbing Li, Xuewei Cao, Zhiwen Zeng, Yuhan Wu, Yanyong Zhang, Yan Xia

OpenGo, from the University of Science and Technology of China, presents an OpenClaw-based robotic dog framework that uses large language models to enable real-time skill switching. The system integrates a validated skill library, an LLM-based dispatcher, and a self-learning framework to perform tasks on a Unitree Go2 robot, mitigating LLM hallucination and supporting natural language interaction.

Key insights

  • Constraining the LLM's role to skill selection and bounded parameter tuning, rather than direct low-level action generation, effectively mitigates 'hallucination' and enhances safety and reliability in physical robot deployment.
  • A modular, pre-validated skill library provides a robust and interpretable abstraction layer, simplifying the integration of complex capabilities and ensuring robot-compatible execution.
  • Integrating a closed-loop self-learning framework with human feedback enables continuous improvement in skill selection and parameterization, adapting robot behavior over time without altering core control code.

Summary by alphaXiv.

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