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 2602.20659cs.AI
收藏

递归信念视觉语言动作模型

Recursive Belief Vision Language Action Models

Department of Engineering Design, Indian Institute of Technology, Madras, Chennai, India(工程设计系,印度理工学院,马德拉斯,钦奈,印度)

Vaidehi Bagaria, Bijo Sebastian, Nirav Kumar Patel

AI总结:

RB-VLA通过信念与意图联合条件化扩散策略,实现长周期任务的高效执行,显著提升成功率并降低推理延迟。

摘要:

视觉-语言-动作模型必须使智能体能够在部分可观测条件下执行长周期任务。然而,大多数现有方法仍以观察驱动为主,依赖于短上下文窗口或重复查询视觉-语言模型(VLMs)。这导致任务进度丢失、感知混叠下的动作重复以及高推理延迟。虽然语义定位很重要,但长周期操作本质上需要持久、动作条件化的状态表示。当前的VLAs缺乏此类表示,并且在时间和物理推理方面有限,使其不适合多阶段控制。本文介绍了RB-VLA,一种以信念为中心的架构,通过自监督世界模型目标进行训练,能够维护一个紧凑的潜在状态编码,包含任务相关的历 史、动态和物体交互。每次任务查询一次,VLM提供高层意图,而信念跟踪任务进度,并在部分可观测条件下启用相位感知、因果基础的控制,而无需存储原始观察或随时间扩展内存。信念和意图共同条件化扩散策略以实现稳健的闭环执行。RB-VLA在长周期基准上优于先前的VLAs,在多阶段拾取和堆叠任务中分别达到比pi_0高52.5%和37.5%的成功率。它还相对于基线减少了五倍的推理延迟,并消除了现有VLAs中在时间步长上观察到的内存增长。消融实验显示,信念模块是性能的主要驱动力,使成功率从无信念的32.5%提升到有信念的77.5%。

英文摘要:

Vision-language-action models must enable agents to execute long-horizon tasks under partial observability. However, most existing approaches remain observation-driven, relying on short context windows or repeated queries to vision-language models (VLMs). This leads to loss of task progress, action repetition under perceptual aliasing, and high inference latency. While semantic grounding is important, long-horizon manipulation fundamentally requires persistent, action-conditioned state representations. Current VLAs lack such representations and exhibit limited temporal and physical reasoning, making them ill-suited for multi-stage control. This paper introduces RB-VLA, a belief-centric architecture trained with self-supervised world-model objectives that maintains a compact latent state encoding task-relevant history, dynamics, and object interactions. Queried once per task, the VLM provides high-level intent, while the belief tracks task progress and enables phase-aware, causally grounded control under partial observability without storing raw observations or scaling memory with time. The belief and intent jointly condition a diffusion policy for robust closed-loop execution. RB-VLA outperforms prior VLAs on long-horizon benchmarks, achieving 52.5 percent and 37.5 percent higher success rates on multi-stage pick-and-place and stacking tasks, respectively, compared to pi_0. It also reduces inference latency by up to five times relative to baselines and eliminates memory growth across timesteps observed in existing VLAs. Ablations show the belief module is the primary driver of performance, increasing success rates from 32.5 percent without belief to 77.5 percent with belief.

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。

↑