This work outlines six defining characteristics of Self-Driving Laboratory (SDL) 2.0: interoperable, collaborative, generalizable, orchestrated, safe, and creative, forming a framework for next-generation autonomous laboratories.
29 April 2026 in Journal of the American Chemical Society
Autonomous laboratories hold great promise for accelerating materials discovery but often inherit hidden limits because experimental boundaries have been predefined by human intuition or literature precedents. Such a priori constraints risk excluding feasible regions, particularly since synthesizable conditions can shift with hardware or environmental factors. We present SPACESHIP, an AI framework integrated with automated experimental hardware for adaptive exploration of chemical spaces free from literature- or expert-derived feasibility constraints. Through an AI-based prediction, robotic synthesis, real-time characterization, and model update, SPACESHIP combines probabilistic models with an Autopilot acquisition strategy that dynamically switches between models to refine synthesizable regions using both successful and failed experiments. Applied to gold nanoparticle (NP) and nanorod (NR) synthesis, this AI-robotics system achieved 90% accuracy in only 23 experiments, compared with 512 required for the ground truth. It uncovered distinct growth regimes across optical property classes and expanded synthesizable regions by factors of 8 (NPs) and 4 (NRs) beyond literature maps, adapting to hardware-specific conditions rather than relying on fixed, external constraints. By merging machine learning with autonomous experimentation, SPACESHIP addresses the long-standing reproducibility gap in science by diagnosing and adapting to system-specific synthesizable boundaries that shift across laboratories and environments, rather than assuming one universal map.
Self-driving laboratories have recently emerged as a transformative research paradigm that integrates automated experimental systems with AI-based decision-making to significantly accelerate material discovery. While prior efforts have focused primarily on automating nanoparticle synthesis, achieving true experimental autonomy necessitates the inclusion of postsynthesis preprocessing steps such as washing, separation, and purification. Among these methods, the nanoparticle washing process presents a unique automation challenge. Despite its apparent simplicity, the process is inherently adaptive, with experimental conditions and outcomes that vary dynamically in response to subtle changes. Effective automation of nanoparticle washing requires both visual adaptivity—to detect changes in the appearance of dispersions or precipitates—and cognitive adaptivity—to handle failure cases, such as incomplete sedimentation or phase separation. This study introduces an integrated platform that automates this complex preprocessing task, significantly broadening the operational capabilities of current self-driving labs. The proposed platform offers a practical and scalable solution for realizing fully autonomous experimental workflows. By bridging the gap between synthesis and characterization through intelligent preprocessing, this work represents a critical step toward establishing an end-to-end autonomous pipeline for materials development.
We present NanoChef, a deep learning-based framework for the simultaneous optimization of synthesis sequences and reaction conditions in autonomous laboratories. Unlike traditional AI models focused solely on continuous variables, NanoChef incorporates positional encoding and MatBERT embedding to represent reagent sequences as vectorized inputs. This enables joint modeling of categorical and continuous variables in nanoparticle (NP) synthesis. In virtual experiments, NanoChef consistently identified global optima across synthesis-order-sensitive landscapes, requiring fewer than 40 cycles. For real-world Ag NP synthesis with a λmax of 513 nm by UV‒Vis absorption spectroscopy and high monodispersity, the framework outperformed fixed-order methods, achieving a 32% reduction in the full width at half maximum (FWHM) and reaching optimal recipes within 100 experiments. Extending to a three-reagent system, NanoChef newly discovered an oxidant‒last strategy that yielded the most uniform NPs. This work redefines synthesis order as a tunable design variable and demonstrates how lightweight AI architecture can accelerate autonomous chemistry.
Autonomous materials discovery platforms have accelerated high-throughput synthesis, and here we advance this paradigm by achieving real-time tracking of complex solid-state reaction pathways.