Science acceleration and accessibility with self-driving labs
Richard B. Canty, Jeffrey A. Bennett, Keith A. Brown, Tonio Buonassisi, Sergei V. Kalinin, John R. Kitchin, Benji Maruyama, Robert G. Moore, Joshua Schrier, Martin Seifrid, Shijing Sun, Tejs Vegge and Milad Abolhasani, “Science acceleration and accessibility with self-driving labs”, Nature Communications 16, 3856 (2025). https://doi.org/10.1038/s41467-025-59231-1
This summary was written by a language model from the paper and has not been checked line by line. Rely on the published paper for details.
This is a Perspective, not a research article: it reports no new experiments or data, and the summary below describes its arguments and recommendations.
Question
The authors ask how self-driving laboratories (SDLs) in chemistry and materials science can be made accessible to a broad research community rather than to a few well-funded groups, and what technical, workforce and institutional obstacles stand in the way (Canty et al. 2025). They frame SDLs not only as tools that accelerate experiments but as potential hubs for human-machine and human-human collaboration, which could let a larger and more diverse community tackle problems too large or multidisciplinary for individual labs.
Approach
The perspective first defines an SDL as a combination of digital tools that predict, propose experiments and update beliefs, plus automated hardware that executes experiments, working toward a human-set goal. It distinguishes SDLs from conventional high-throughput or cloud laboratories by their judicious choice of experiments, adaptation of methods, and integration of multiple tools into workflows. It then contrasts two deployment paradigms, centralized facilities that users access by application or fee and distributed networks of group-level SDLs that share jobs, information and materials, and compares them qualitatively on ownership, staffing, setup and operating costs, throughput, and breadth of research areas. The rest of the piece reviews roadblocks at three levels (the workforce, the lab, and the community) and closes with recommendations.
Key findings
On deployment, the authors argue that centralized facilities pool resources into more capable SDLs, appeal to industry and national funders, and can more easily set common protocols and obtain regulatory certification, but may be slow to adapt and may find specialized equipment hard to justify. Distributed SDLs are cheaper per unit and more flexible, but demand much more coordination of data, digital twins and standards. They favor hybrid schemes, for example groups developing workflows on low-cost “frugal twin” systems before submitting them to a facility, or building instruments to facility guidelines so they can plug in.
On the workforce, they describe three roles (developers, technicians and users) with different training needs, and note that team science is mostly learned on the job; they call for curricula, micro-credentials and pedagogical low-cost platforms. In the lab, they point to poorly documented or restricted vendor APIs, opaque data formats, the tension between general and instrument-specific software, limited adoption of middleware such as ROS, SiLA 2 and BlueSky outside the most advanced groups, and early use of LLMs to write integration code, which needs safeguards. Machine learning challenges include encoding constraints, translating long-term objectives such as sustainability into measurable ones, the limits of proxy measurements in extrapolative campaigns, and the need to measure experimental uncertainties directly.
At the community level, they identify a self-reinforcing cycle in which immature, black-box systems yield few industrially convincing demonstrations, which limits funding. They argue that trust in SDL data requires recording environmental and state metadata, calibration and benchmarking studies, and honest reporting of shortcomings, operational performance, degree of human involvement and waste, and that current funding rewards scientific results rather than shared infrastructure and long-term software maintenance.
Significance
The perspective sets out a community agenda for SDLs that emphasizes access and reproducibility alongside speed. Its concrete proposals include identifiers for SDLs (analogous to ORCID) for data provenance and credit, a consortium to maintain living standards for automation, software and data interfaces, an institute for automated laboratory infrastructure that could fund long-term software and training, and SDL “installation wizards” that help groups choose equipment and adapt workflows. These ideas are developed further in the concept article democratization of research.