自驱动实验室（SDL）通过连接自动化实验、结构化数据、预测模型与在线决策，使科学研究由预设流程执行转向基于实验反馈的迭代发现。梳理SDL的通用技术架构与成熟度判据，重点比较药物发现与开发中的代表性应用及工业实践，并从理化性质、吸收-分布-代谢-排泄-毒性（ADMET）、靶点成药性和制剂成药性4个子领域总结现有平台、技术路线、应用成熟度及主要瓶颈。证据显示，SDL成熟度主要取决于实验可标准化程度、反馈周期、读出稳定性和异常的机器可识别性：制剂、递送材料和限定偶联工艺已出现L2～L3级湿实验闭环，经典药物体外ADMET仍以预测平台和人在回路系统为主，体内药动学距在线闭环更远。面向成药性评价的SDL不能照搬材料发现模式，其核心在于在生物变异、异步多目标实验和质量合规条件下形成可验证反馈。

Self-driving laboratories (SDL) connect automated experimentation, structured data, predictive models, and online decisionmaking, shifting scientific research from predefined workflow execution to iterative, feedback-driven discovery. This review summarizes the general technical architecture and maturity criteria of SDLs, with emphasis on representative applications and industrial practices in drug discovery and development. It further examines four subdomains of drug developability—physicochemical properties, ADMET assessment, target druggability evaluation, and formulation developability—in terms of available platforms, technical routes, application maturity, and major bottlenecks. Current evidence indicates that SDL maturity is primarily governed by assay standardizability, feedback time, readout stability, and machine-recognizable exceptions. L2—L3 wet-laboratory closed loops have emerged in formulation, delivery materials, and well-defined conjugation processes, whereas classical in vitro ADMET remains dominated by predictive platforms and human-in-the-loop systems, and in vivo pharmacokinetics remains further from online closed-loop operation. SDLs for drug developability assessment cannot simply replicate materials-discovery workflows; their central challenge is to establish verifiable feedback under biological variability, asynchronous multi-objective experiments, and quality and compliance constraints.

R9