arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.09017cs.ROcs.LG
收藏

接触锚定策略:接触条件创造强大的机器人效用模型

Contact-Anchored Policies: Contact Conditioning Creates Strong Robot Utility Models

New York University(纽约大学);University of California, Berkeley(加州大学伯克利分校);University of California, Los Angeles(加州大学洛杉矶分校);Hello Robot Inc.(Hello Robot公司);Ai2;University of Waterloo(滑铁卢大学)
Zichen Jeff Cui, Omar Rayyan, Haritheja Etukuru, Bowen Tan, Zavier Andrianarivo, Zicheng Teng, Yihang Zhou, Krish Mehta, Nicholas Wojno, Kevin Yuanbo Wu, Manan … 展开作者

Zichen Jeff Cui, Omar Rayyan, Haritheja Etukuru, Bowen Tan, Zavier Andrianarivo, Zicheng Teng, Yihang Zhou, Krish Mehta, Nicholas Wojno, Kevin Yuanbo Wu, Manan H Anjaria, Ziyuan Wu, Manrong Mao, Guangxun Zhang, Binit Shah, Yejin Kim, Soumith Chintala, Lerrel Pinto, Nur Muhammad Mahi Shafiullah

AI总结:

接触锚定策略通过接触条件实现强大的机器人效用模型,仅用23小时演示数据即在三种基本操作技能上实现跨环境泛化,并在零样本评估中超越现有最先进方法56%。

摘要:

机器人学习中普遍的范式试图在运行时通过语言提示跨环境、形态和任务进行泛化。一个根本性的张力限制了这种方法:语言通常过于抽象,无法引导所需的具身体验理解。在本工作中,我们引入了接触锚定策略(CAP),用空间中的接触点替代语言条件。同时,我们将CAP结构化为一个模块化效用模型的库,而不是一个单一的通用策略。这种分解使我们能够实现从现实到模拟的迭代循环:我们构建了EgoGym,一个轻量级的模拟基准,以快速识别故障模式并改进我们的模型和数据集,在现实世界部署之前。我们展示了通过接触条件和模拟迭代,CAP能够直接泛化到三种基本操作技能的新环境和形态,仅使用23小时的演示数据,并在零样本评估中比大型、最先进的VLAs高出56%。所有模型检查点、代码库、硬件、模拟和数据集都将开源。项目页面:https://cap-policy.github.io/

英文摘要:

The prevalent paradigm in robot learning attempts to generalize across environments, embodiments, and tasks with language prompts at runtime. A fundamental tension limits this approach: language is often too abstract to guide the concrete physical understanding required for robust manipulation. In this work, we introduce Contact-Anchored Policies (CAP), which replace language conditioning with points of physical contact in space. Simultaneously, we structure CAP as a library of modular utility models rather than a monolithic generalist policy. This factorization allows us to implement a real-to-sim iteration cycle: we build EgoGym, a lightweight simulation benchmark, to rapidly identify failure modes and refine our models and datasets prior to real-world deployment. We show that by conditioning on contact and iterating via simulation, CAP generalizes to novel environments and embodiments out of the box on three fundamental manipulation skills while using only 23 hours of demonstration data, and outperforms large, state-of-the-art VLAs in zero-shot evaluations by 56%. All model checkpoints, codebase, hardware, simulation, and datasets will be open-sourced. Project page: https://cap-policy.github.io/

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑