arXiv cs.AI· Apoorva Vashisth, Manav Kulshrestha, Damon Conover, Zachary Kingston, Aniket Bera·· 1 小时前AI 评分42
RePlan-VLA:面向长时程视觉-语言-动作模型的进度感知重规划框架
RePlan-VLA: Progress-Aware Replanning for Long-Horizon Vision-Language-Action Models
AI 导读
RePlan-VLA 是一个 VLA 无关的框架,通过监控机器人动作进度评估并自适应修订未来计划,把 VLA 模型的短时程操作能力转化为可靠的长时程执行,且无需对底层 VLA 策略做任何额外微调。在 RoboCasa365 和 CALVIN 基准上,成功率分别比次强基线提升 21.9% 和 10.4%。该方法还在 Franka Research 3 机器人上完成了真实环境部署验证。
正文 · 原文
Abstract:Vision-Language-Action (VLA) models have emerged as a promising paradigm for general-purpose robotic manipulation, yet their performance often degrades on long-horizon tasks that require multiple interdependent manipulations. Instead of burdening the VLA policies with long-horizon reasoning, memory, action generation, progress tracking, and error recovery, we argue for leveraging their short-horizon manipulation capabilities within an execution-aware framework. To this end, we introduce \textnormal{\textbf{RePlan-VLA}}, a VLA-agnostic framework that translates a VLA's short-horizon capabilities into reliable long-horizon execution. RePlan-VLA explicitly monitors the robot's motion for progress evaluation, and adaptively revises the future plan based on observed outcomes. We preserve the underlying VLA policy entirely, requiring no additional fine-tuning. Experiments on RoboCasa365 and CALVIN benchmarks show an improvement of 21.9% and 10.4% in success rate over the next strongest baseline. We also validate our approach for real-world deployment on a Franka Research 3 robot.
| Comments: | Apoorva Vashisth and Manav Kulshrestha contributed equally to this work. Prof. Zachary Kingston and Prof. Aniket Bera jointly advised this work |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2611.23843 [cs.RO] |
| (or arXiv:2611.23843v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2611.23843 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Apoorva Vashisth Ms. [view email]
[v1]
Wed, 18 Nov 2026 04:49:41 UTC (6,407 KB)
来源:arXiv cs.AI · arxiv.org