From Digital to Physical: Digital Agents as Autonomous Coaches for Physical Intelligence

📄 arXiv: 2601.21570 📥 PDF

作者: Zixing Lei, Genjia Liu, Yuanshuo Zhang, Qipeng Liu, Yuzhu Cai, Sixiang Chen, Jixian Wu, Yunhong Wang, Weixin Li, Chuan Wen, Bo Zhao, Shanghang Zhang, Wenzhao Lian, Siheng Chen

分类: cs.AI, cs.RO

发布日期: 2026-06-12

💡 一句话要点

提出EmboCoach-Bench以解决机器人自主工程问题

🎯 匹配领域: 支柱二：RL算法与架构 (RL & Architecture) 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 具身人工智能 自主工程 大型语言模型 强化学习 模仿学习 动态反馈机制 策略优化

📋 核心要点

  1. 现有的具身人工智能方法严重依赖人工监督，导致扩展能力受限，难以实现高效的策略开发。
  2. 本文提出EmboCoach-Bench基准，利用大型语言模型代理自主生成和优化具身策略，采用动态反馈机制提升策略开发效率。
  3. 实验结果显示，代理的平均成功率比人类设计的基线高出26.5%，并有效缩小了开源与专有模型之间的性能差距。

📝 摘要（中文）

具身人工智能领域正在快速发展，然而现有方法依赖于劳动密集型的手动监督，限制了其扩展能力。为此，本文提出了EmboCoach-Bench，一个评估大型语言模型（LLM）代理自主工程具身策略的基准框架。该框架涵盖32个专家策划的强化学习和模仿学习任务，采用可执行代码作为通用接口。通过动态闭环工作流程，代理利用环境反馈迭代草拟、调试和优化解决方案。实验结果表明，代理的成功率比人类工程基线高出26.5%，并且在自我纠正能力方面表现出色，能够从几乎完全失败的情况下恢复任务性能。此研究为具身智能的自我演化奠定了基础，推动了从手动调优到可扩展自主工程的转变。

🔬 方法详解

问题定义：本文旨在解决具身人工智能领域中，现有方法对人工监督的高度依赖，导致策略开发效率低下的问题。

核心思路：提出EmboCoach-Bench基准，利用大型语言模型（LLM）代理自主生成和优化具身策略，通过动态闭环反馈机制提升策略开发的自主性和效率。

技术框架：整体架构包括环境反馈模块、策略生成模块和自我调试模块。代理通过与环境的互动获取反馈，迭代优化策略。

关键创新：最重要的创新在于引入动态闭环工作流程，使代理能够在执行过程中实时调整和优化策略，与传统静态生成方法形成鲜明对比。

关键设计：在设计中，采用了物理信息驱动的奖励设计和扩散策略架构，确保代理在复杂任务中的有效性和适应性。

📊 实验亮点

实验结果表明，代理的平均成功率比人类设计的基线高出26.5%。此外，代理在环境反馈的支持下，能够有效缩小开源与专有模型之间的性能差距，并在自我纠正能力上表现出色，成功恢复任务性能。

🎯 应用场景

该研究的潜在应用领域包括智能机器人、自动化控制系统和虚拟教练等。通过自主工程的能力，能够显著提高机器人在复杂环境中的适应性和决策能力，推动智能系统的广泛应用。

📄 摘要（原文）

The field of Embodied AI is witnessing a rapid evolution toward general-purpose robotic systems, fueled by high-fidelity simulation and large-scale data collection. However, this scaling capability remains severely bottlenecked by a reliance on labor-intensive manual oversight from intricate reward shaping to hyperparameter tuning across heterogeneous backends. Inspired by LLMs' success in software automation and science discovery, we introduce \textsc{EmboCoach-Bench}, a benchmark evaluating the capacity of LLM agents to autonomously engineer embodied policies. Spanning 32 expert-curated RL and IL tasks, our framework posits executable code as the universal interface. We move beyond static generation to assess a dynamic closed-loop workflow, where agents leverage environment feedback to iteratively draft, debug, and optimize solutions, spanning improvements from physics-informed reward design to policy architectures such as diffusion policies. Extensive evaluations yield three critical insights: (1) autonomous agents can qualitatively surpass human-engineered baselines by 26.5\% in average success rate; (2) agentic workflow with environment feedback effectively strengthens policy development and substantially narrows the performance gap between open-source and proprietary models; and (3) agents exhibit self-correction capabilities for pathological engineering cases, successfully resurrecting task performance from near-total failures through iterative simulation-in-the-loop debugging. Ultimately, this work establishes a foundation for self-evolving embodied intelligence, accelerating the paradigm shift from labor-intensive manual tuning to scalable, autonomous engineering in embodied AI field.