VLA-Precision: Asymmetric Co-Bootstrapping for Efficient Real-World Online RL of Vision-Language-Action Models

• 类型：arxiv
• 标识：2609.04355
• 链接：https://arxiv.org/abs/2609.04355
• 主分类：multimodal
• 形态：method
• TLDR：Pretrained vision-language-action (VLA) models enable broad manipulation but remain unreliable in tasks demanding precision and repeatability. Applying real-world online reinforcement learning (RL) to VLA post-training enables autonomous trial-and-error improvement beyond demonstrations alone, but exposes two bottlenecks: 1) unreliable value signals can induce policy drift; 2) large-VLA overhead constrains throughput and sample efficiency. To address these challenges, we present VLA-Precision, an efficient real-world online RL framework featuring the Asymmetric Co-Bootstrapping (ACoB) algorith
• 副分类：engineering
• 待LLM分类：否
• 标题中文：[标题中文] VLA-Precision：面向视觉-语言-动作模型高效真实世界在线强化学习的非对称协同自举
• TLDR中文：预训练的 vision-language-action (VLA) 模型支持广泛的操控任务，但在需要精确性与可重复性的任务中仍不可靠。将真实世界在线强化学习 (RL) 应用于 VLA 后训练，可实现超越单纯示教的自主试错改进，但也暴露出两个瓶颈：1) 不可靠的价值信号会导致策略漂移；2) 大型 VLA 的开销限制了吞吐量和样本效率。为应对这些挑战，我们提出了 VLA-Precision，一个高效的真实世界在线 RL 框架，其核心是 Asymmetric Co-Bootstrapping (ACoB) 算法……
• 来源文件：
• /inbox/tom/_candidates/2026-09-28-agent-rag-longcontext-candidates.json