[论文解读] Perspectives on Sim2Real Transfer for Robotics: A Summary of the R:SS 2020 Workshop

Sebastian Höfer, Kostas E. Bekris+13 more|arXiv (Cornell University)|Dec 7, 2020

Robot Manipulation and Learning参考文献 12被引用 34

一句话总结

本论文总结了 R:SS 2020 Sim2Real 工作坊的辩论、贡献摘要和小组讨论，概述了将机器人技能从仿真转移到现实的实践者指引和未决研究问题。

ABSTRACT

This report presents the debates, posters, and discussions of the Sim2Real workshop held in conjunction with the 2020 edition of the "Robotics: Science and System" conference. Twelve leaders of the field took competing debate positions on the definition, viability, and importance of transferring skills from simulation to the real world in the context of robotics problems. The debaters also joined a large panel discussion, answering audience questions and outlining the future of Sim2Real in robotics. Furthermore, we invited extended abstracts to this workshop which are summarized in this report. Based on the workshop, this report concludes with directions for practitioners exploiting this technology and for researchers further exploring open problems in this area.

研究动机与目标

• 总结关于 Sim2Real 转移在机器人领域的定义、可行性与重要性的辩论。
• 展示贡献摘要及其关于弥合现实差距的主题。
• 提供面向从业者的在真实机器人任务中应用 Sim2Real 技术的指南。
• 确定未解决的研究挑战以指导未来的 Sim2Real 工作。

提出的方法

• 组织了三场扮演对立角色的辩论，以探讨关于 Sim2Real 的核心问题。
• 概述了 18 篇同行评审摘要，展示用于 Sim2Real 的方法、表征和数据驱动的方法。
• 召开小组讨论以综合洞见与未来方向。
• 将洞见分类为从业者与研究者视角，提供可操作的指南。
• 突出提出将 Sim2Real 技术应用的具体建议及需要开展开放研究的领域。

实验结果

研究问题

• RQ1 What is Sim2Real, and is it a distinct field or a collection of methods?
• RQ2 To what extent can simulation-based training transfer to real robotic tasks, and what limits exist?
• RQ3 What strategies (e.g., domain randomization, intermediate representations, differentiable simulators) best bridge the reality gap?
• RQ4 What are practical guidelines and open questions to advance Sim2Real toward production-level robotics?

主要发现

• Sim2Real debates highlighted cost, democratization, safety, and the reality gap, with consensus that simulation alone is not sufficient for real-world success.
• Sim2Real can be viewed both as a field of study and as a suite of techniques, but the reality gap remains a central challenge.
• Bridging approaches include domain randomization, explicit transferable abstractions, intermediate representations, and leveraging real-world data to improve simulators.
• Panel discussion emphasized the value of task-agnostic simulators with cautions about applicability, as well as promising directions like differentiable simulators and meta-learning.
• The contributed abstracts span formalization, bridging with existing simulators, data-driven model improvements, and online parameter inference, illustrating diverse approaches to Sim2Real.
• Practitioners are advised to consider bootstrapping, data efficiency, hardware-in-the-loop optimization, and careful tuning when applying Sim2Real techniques.