文献类型：期刊文献

英文题名：BioProVLA-Agent: An Affordable, Protocol-Driven, Vision-Enhanced VLA-Enabled Embodied Multi-Agent System with Closed-Loop-Capable Reasoning for Biological Laboratory Manipulation

作者：Du, Zhaohui[1,2]; Wang, Zhe[1,2]; Fei, Hongmei[4]; Cao, Xiwen[1,2]; Xiao, Ting[1,2]; Wang, Qi[3]; Jin, Huanbo[1,2]; Gu, Jiaming[1,2]; Lu, Quan[1,2]; Liu, Zhe[1,2]

机构：[1] Key Laboratory of Smart Manufacturing in Energy Chemical Process Ministry of Education, East China University of Science and Technology, Shanghai, China; [2] Department of Computer Science and Engineering, East China University of Science and Technology, Shanghai, China; [3] Department of Laboratory Medicine, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China; [4] School of Information Science and Technology, Shihezi University, Shihezi, China

年份：2026

外文期刊名：arXiv

收录：EI(收录号：20260287286)

语种：英文

外文关键词：Intelligent agents - Interface states - Laboratories - Molecular biology - Robots - User interfaces - Vision - Visual languages - Waste disposal

摘要：Biological laboratory automation holds promise for reducing repetitive manual work, improving workflow reproducibility, and broadening access to experimental execution. Yet building embodied agents that operate reliably in real wet-lab environments remains difficult. Biological protocols are typically expressed as unstructured natural language, common labware such as tubes and bottles are often transparent or reflective, and multi-step procedures require state-aware execution rather than one-shot instruction following. In addition, existing robotic laboratory platforms often depend on expensive hardware, dedicated instruments, fixed workflows, or robotics-oriented interfaces, which can limit their accessibility for resource-constrained laboratories and make them difficult to use or adapt for domain users without robotics expertise. Here, we introduce BioProVLA-Agent, an affordable, protocol-driven, and vision-enhanced embodied multi-agent system enabled by Vision-Language-Action (VLA) models for biological laboratory manipulation. BioProVLA-Agent uses experimental protocols as the task interface and couples protocol understanding with closed-loop-capable reasoning and embodied execution. A Tailored LLM Protocol Agent transforms unstructured protocols into executable and verifiable subtask units, while a VLM-RAG Verification Agent reasons over real-time visual observations, robot states, retrieved operation knowledge, and reference success/failure examples to assess task readiness and completion. The verified subtasks are then executed by a VLA Embodied Agent based on a lightweight VLA policy. To address wet-lab visual perturbations, we further develop AugSmolVLA, an online augmentation strategy tailored to transparent labware, specular reflections, illumination shifts, and overexposure. We evaluate BioProVLA-Agent on a hierarchical biological manipulation benchmark spanning 15 atomic tasks, 6 composite workflows, and 3 representative bimanual tasks, including centrifuge-tube loading, tube sorting, waste disposal, cap twisting, and liquid pouring. Across normal and high-exposure settings, AugSmolVLA improves execution stability over ACT, X-VLA, and the original SmolVLA, with pronounced gains in precise placement, transparent-object manipulation, composite workflows, and visually degraded scenes. These results demonstrate a practical route toward affordable, protocol-centered, and verification-capable embodied AI systems for biological laboratory manipulation. Copyright ? 2026, The Authors. All rights reserved.