Scientific engineering of self-driving laboratories at scale

<jats:p>Scaling self-driving laboratories underpins the broader vision of intelligent scientific infrastructure. Scaling, however, creates a fundamental tension. It expands the space of discoverable phenomena and the opportunity for knowledge gain, but also introduces operational complexity, semantic heterogeneity and decision dependencies that can erode controllability, interpretability and functional coherence, a scale-induced disorder that we define as system entropy. To maintain functional coherence at scale, we progressively developed an engineering framework for suppressing system entropy and implemented it in AIchem. The framework combines hardware–software co-design, layered modularity, capability abstraction and skill encapsulation. Together, these principles organize heterogeneous research objects, instruments, computational tools and algorithmic models into programmable, composable units coordinated through closed-loop flows of tasks and data. To date, AIchem spans more than 2,600 m2 and encompasses 605 registered workstations. Across this infrastructure, AIchem supports research in various fileds, including battery, catalysis, biochemistry and functional materials, and has handled more than 10,000 research task dispatches. Together, these measures of deployment and use demonstrate a practical route to scientific engineering intelligent scientific infrastructure at scale.</jats:p>