What makes a laboratory self-driving? Mapping the SDL terminology landscape and introducing the FLEx categorization framework
Self-driving laboratories (SDLs) are increasingly used to accelerate experimentation across chemistry, materials science, biotechnology, and related fields. Yet, the rapid development of autonomous experimentation has produced a fragmented terminology landscape in which overlapping terms, including autonomous laboratories, materials acceleration platforms, and robot scientists, are applied inconsistently. This complicates comparisons between systems and obscures what constitutes scientific autonomy. Here, we examine the definitions, historical development, disciplinary associations, and usage patterns of SDLs and related concepts. We further review existing approaches to categorizing autonomous laboratory systems and show that many rely on characteristics such as hardware complexity, system architecture, or human involvement that are difficult to apply consistently across disciplines and technologies. To provide a technology-independent alternative, we introduce the Feedback Loop Extent (FLEx) framework. FLEx classifies systems according to how far experimental feedback reaches into the scientific process, progressing from scripted execution without a scientific feedback loop to parameter optimization, autonomous experiment planning, hypothesis formulation, and scientific goal formulation. Applying FLEx to representative published systems shows that most contemporary SDLs autonomously plan experiments within human-defined objectives, whereas few demonstrably formulate hypotheses and none unambiguously reformulate broader scientific goals. FLEx provides a common language for distinguishing automation from different forms of autonomy and enables more transparent comparison and reporting of current and future SDLs.
Authors
- Frank Glorius (ORCID: https://orcid.org/0000-0002-0648-956X)
- Thorsten Teutenberg (ORCID: https://orcid.org/0000-0002-2477-6167)
- Florian Boser (ORCID: https://orcid.org/0000-0003-0550-3822)
- Stefan Braese (ORCID: https://orcid.org/0000-0003-4845-3191)
- Kjell Kochale
- Guangqi Wu (ORCID: https://orcid.org/0000-0002-9763-6982)
- Christian P. Haas (ORCID: https://orcid.org/0000-0002-9457-8019)
- Alexander Böser (ORCID: https://orcid.org/0009-0005-7618-7873)
- Shuyuan Zhang (ORCID: https://orcid.org/0000-0002-9297-4333)
- K. Cheung (ORCID: https://orcid.org/0000-0002-2431-3250)
- Nikhil Deshpande
- Alexei Lapkin
- Patrick Courtney
- Radhika Gudipati
Institutions
- Karlsruhe Institute of Technology (DE)
- University of Nottingham (GB)
- Shandong University (CN)
- University of Münster (DE)
- University of Cambridge (GB)
- Agilent Technologies (Germany) (DE)
- Institut für Energie- und Umwelttechnik (DE)
- Advanced Research + Invention Agency (GB)
Publication Details
- Journal
- ChemRxiv
- Published
- 2026-11-20
- DOI
- https://doi.org/10.26434/chemrxiv.15012533/v1
- Primary Topic
- Machine Learning in Materials Science
- Type
- preprint