Closed-loop, autonomous experimentationenables accelerated and material efficientexploration of large reaction spaces without the need for user inter[1]vention.However, autonomous exploration of advanced materials with com[1]plex,multi-step processes and data sparse environments remains a challenge. In thiswork, we present AlphaFlow, a self-driven fluidic lab capable of auton[1]omousdiscovery of complex multi-step chemistries. AlphaFlow uses reinfor[1]cementlearning integrated with a modular microdroplet reactor capable of performingreaction steps with variable sequence, phase separation, washing, andcontinuous in-situ spectral monitoring. To demonstrate the power ofreinforcement learning toward high dimensionality multi-step chemistries, weuse AlphaFlow to discover and optimize synthetic routes for shell-growth ofcore-shell semiconductor nanoparticles, inspired by colloidal atomic layerdeposition (cALD). Without prior knowledge of conventional cALD para[1]meters,AlphaFlow successfully identified and optimized a novel multi-step reactionroute, with up to 40 parameters, that outperformed conventional sequences.Through this work, we demonstrate the capabilities of closed-loop,reinforcement learning-guided systems in exploring and solving challenges inmulti-step nanoparticle syntheses, while relying solely on in-house generateddata from a miniaturized microfluidic platform. Further application of Alpha Flowin multi-step chemistries beyond cALD can lead to accelerated funda[1]mentalknowledge generation as well as synthetic route discoveries and optimization.

图 1：多步骤化学中的维度问题。说明批量多步合成的复杂性和所需资源呈指数级增长，包括四个可能的步骤选择和多达 32 个连续步骤。

图 2：AlphaFlow 概述。

图 3：cALD 序列选择活动结果。

图 4：AlphaFlow 的体积和时间优化活动结果。

AlphaFlow: autonomous discovery and optimization of multi-step chemistry using a self-driven fluidic lab guided by reinforcement learning