First applications of self-driving labs in chemicals and pharmaceuticals

In self-driving laboratories, robots and artificial intelligence algorithms plan and carry out experiments, selecting each new experiment on the basis of the results of the previous one. The researcher intervenes only in exceptional cases. Originating in universities, these laboratories are now used in the chemical and pharmaceutical industries to optimise drug manufacturing processes and coating and adhesive formulations. According to researchers in the field and the European Commission's scientific advisors, their wider uptake across companies and sectors requires reliable transfer of results from the laboratory to production, common standards, rules governing the autonomy of these systems and new skills.

From trial and error to the self-driving laboratory

The development of a material, a chemical reaction or a manufacturing process currently proceeds by trial and error. The researcher plans an experiment, carries it out, analyses the results and decides which conditions to change in the next one, such as temperature, reaction time or the quantities of reagents. Advanced materials, such as those used in batteries, semiconductors or medical devices, are specifically designed to achieve improved or entirely new properties. With conventional methods their development can take between 10 and 30 years (European Commission, 2024). The European Commission's Group of Chief Scientific Advisors, seven independent scientists who provide scientific advice to the College of Commissioners, published an opinion on advanced materials in April 2026. In it, the Group describes trial-and-error experimentation as a costly method that slows both the discovery of materials and their transfer to production (Group of Chief Scientific Advisors, 2026).

Automation has long been present in some stages of this work. In the pharmaceutical industry, high-throughput experimentation, in which automated equipment runs many experiments in parallel, has been in use since the 1990s (Zhu et al., 2025). In these settings, liquid-handling robots dispense the reagents and software processes the measurements, while the choice of the next experiment remains with the researcher. This configuration corresponds to the first of the five levels of autonomy that Tobias and Wahab (2025) set out for laboratories, modelled on those defined by the Society of Automotive Engineers for self-driving vehicles. At the second level, the system also automates part of the planning, such as drafting the experimental protocol.

The self-driving laboratory begins at the third level, when robotics, automation and artificial intelligence plan and carry out the experiments (Nature Synthesis, 2026a). The system interprets the result of each experiment and designs the next one on that basis, over several consecutive cycles, while the researcher intervenes only in exceptional cases. At the fourth level, the system also revises the initial hypotheses provided by the researcher. The fifth level, a system that receives only general objectives and conducts research without further instructions, has not yet been achieved (Tobias and Wahab, 2025).

An example of the fourth level is Adam, the robot scientist developed by the universities of Aberystwyth and Cambridge. In 2009 it formulated hypotheses about the function of certain yeast genes, tested them with its own experiments and interpreted the results. The hypotheses, subsequently checked by manual experiments, proved to be both novel and correct (University of Cambridge, 2009).

Most current self-driving laboratories operate at the third level and carry out optimisation experiments, searching for the best combination of conditions within a set of variables defined by the researcher (Tobias and Wahab, 2025). The next experiment is selected using statistical methods such as Bayesian optimisation, which estimates how the experimental conditions affect the measured outcome and updates the estimate after each experiment. According to Pilon et al. (2026), this reduces the number of experiments needed to identify the optimal conditions.

Results in academic research

In 2020 Andrew Cooper's group at the University of Liverpool presented a mobile robot that moves around a laboratory designed for human researchers and uses the same instruments as they do. In eight days the robot carried out 688 experiments to improve a photocatalyst, a material that uses light to accelerate a chemical reaction, in this case the production of hydrogen. The photocatalyst it identified was six times more active than the initial formulations (Burger et al., 2020; University of Liverpool, 2020).

At the University of Amsterdam, Timothy Noël's group has developed RoboChem-Flex, a platform built from 3D-printed or commercially available components. The study presenting it provides the assembly instructions and the software is freely available. The system optimised six reactions of four types, including light-driven and enzyme-catalysed reactions, and the group then validated the identified conditions on a larger scale. In one of the experiments RoboChem-Flex achieved a yield of 71% of the theoretical maximum quantity of product, with a two-minute residence time in the reactor. Another research group had previously obtained the same yield in five hours (Pilon et al., 2026). Where samples are analysed with instruments already present in the laboratory, the platform costs approximately US$5,000, whereas commercial systems with robotic arms, liquid dispensers and pumps often exceed US$100,000 (Pilon et al., 2026).

First applications in the chemical and pharmaceutical industries

Nature Synthesis (2026a) devoted its October issue to self-driving laboratories in industry, with contributions from researchers at pharmaceutical, chemical and electronics companies. According to the editorial, investment in these laboratories is widespread across industry, among both start-ups and large corporations. Youn-Suk Choi, Vice President of Technology at the Samsung Advanced Institute of Technology, cites the need to accelerate innovation and to address staffing constraints among the reasons for adoption (Nature Synthesis, 2026c).

In the pharmaceutical sector, researchers from seven companies, including AbbVie, AstraZeneca and Eli Lilly, and from the University of Illinois Chicago describe industrial applications in drug development. These include measuring the solubility of active pharmaceutical ingredients and studying the different crystal structures, or solid forms, that an active ingredient can adopt. Others concern electrochemical synthesis and the preparation of lipid nanoparticles, which are used to deliver ribonucleic acid (RNA) medicines (Dvivedi et al., 2026).

The German chemicals group Evonik (2026) offers pharmaceutical companies a service based on self-driving laboratories for developing the manufacturing processes of active pharmaceutical ingredients. A Bayesian optimisation algorithm evaluates the results in real time and selects the next experiment. Scientists supervise the process through an interface that makes every decision traceable. According to the company, the system runs up to 100 experiments a day and continues the optimisation overnight. On 6 October 2026, at the CPHI pharmaceutical trade fair in Milan, the service received a CPHI Pharma Award in the digital transformation category.

In March 2025 the chemicals group Covestro (2025a) announced an automated laboratory for developing coating and adhesive formulations, partly in collaboration with customers. A formulation is a product's recipe. It typically combines between 7 and 15 components, including binders and crosslinkers, and properties such as hardness, adhesion, opacity, gloss and durability depend on their combination. The laboratory prepares the formulations, applies the products under different climatic conditions to simulate their use and measures their properties. It is designed to operate around the clock, with the aim of carrying out tens of thousands of tests a year.

Machine learning software correlates the compositions tested, both in new experiments and in earlier studies, with the measured properties. On this basis it estimates which new compositions are likely to achieve the properties required of the product. The laboratory prepares them and verifies the outcome (Covestro, 2025a). According to the company, the work remains under the direction of its experts, who take the final decision (Covestro, 2025b).

Scale-up, instrument integration and standards

Scale-up consists of reproducing a laboratory result in progressively larger quantities, up to industrial production, and is often the critical step. Zhihao Wang and Xiaonan Wang (2026) propose that self-driving laboratories should take scale into account from the outset, linking laboratory results to the subsequent stages of testing and process development. Each combination of conditions would thus be classified as ready for larger-scale testing, as unsuitable because incompatible with industrial constraints, or as requiring further study to reduce scale-related uncertainty.

Autonomous operation also requires the laboratory's instruments to exchange data and commands without manual intervention. According to the researchers from the pharmaceutical companies, many sensors are not compatible with other equipment and each instrument records its measurements in its own format (Dvivedi et al., 2026). Tobias and Wahab (2025) add that the application programming interfaces (APIs) through which software controls instruments vary from one manufacturer to another. Coordinating the procedures therefore requires complex orchestration software that links the instruments into a single workflow. For self-built instruments based on freely shared designs, the integration work rests with those who assemble the instruments. Moreover, there are no standard operating procedures or user communities to consult in the event of faults (Tobias and Wahab, 2025).

In the view of the Commission's scientific advisors, the lack of standards reduces the reliability of results. There are no common methods for quantifying measurement uncertainty and no shared interfaces or communication protocols between instruments. Instrument calibration remains unstable and reproducibility across sites is limited (Group of Chief Scientific Advisors, 2026).

In a Comment published in the same issue of Nature Synthesis, Brandon Sutherland, Varinia Bernales and Alán Aspuru-Guzik (2026) identify where research is currently being held back. According to the authors, the discovery of new materials is constrained above all by the time required to produce, characterise and validate promising materials in the quantities required by applications. Faster synthesis, better decisions under uncertainty and reliable transfer from the laboratory to production therefore depend on a coordinated system of standards, shared data and technology transfer infrastructure. In their view, no single laboratory, company or public body can build it alone.

Safety and skills

Chen, Li, Lin and Jiang (2026), of the University of Science and Technology of China and China Energy Investment Corporation, argue that an industrial self-driving laboratory requires not only safe robots and instruments but also rules governing the autonomy of the system. These rules establish how a proposal generated by artificial intelligence becomes an executable experiment, how actions are monitored and how results become a reliable basis for further development.

According to Tobias and Wahab (2025), coordinating these systems requires programming skills that are generally uncommon among chemistry and biology researchers, which makes self-driving laboratories less accessible, particularly for smaller institutions. Nessa Carson, interviewed in the same issue of Nature Synthesis (2026b), considers that industry needs teams combining chemistry, data science and engineering, together with rigorous management of experimental data.

In Europe these skills are scarce. According to the Commission's scientific advisors, there is a shortage of laboratory automation engineers and data stewards, and the lack of digital skills and of regulatory knowledge is slowing digitalisation and the adoption of artificial intelligence (Group of Chief Scientific Advisors, 2026).

On the implications for researchers, Ross King, a pioneer of the field and one of Adam's developers, suggests that over the next decade or so self-driving laboratories are more likely to increase the productivity of scientists than to replace them. According to Tobias and Wahab (2025), who interviewed him, the work most affected will be that of early-career researchers, who currently carry out most of the repetitive laboratory tasks.

Summary and outlook

Current self-driving laboratories mainly optimise conditions defined by the researcher (Tobias and Wahab, 2025). In industry, the first applications concern the development of pharmaceutical processes and of coating and adhesive formulations (Evonik, 2026; Covestro, 2025a).

For Europe, the Commission's scientific advisors recommend establishing testbeds for self-driving laboratories with open pilot lines, which are small-scale experimental plants shared among several users. Dedicated funding programmes would give small and medium-sized enterprises and research institutes access to equipment that they could not afford individually. The facilities would be designed from the outset to manage the uncertainty of results. Their effectiveness would be measured against indicators such as development cycle time and the energy consumed per new material identified (Group of Chief Scientific Advisors, 2026).

The advisors also observe that China has taken a leading position in self-driving laboratories (Group of Chief Scientific Advisors, 2026). In South Korea, the Ministry of Trade, Industry and Energy set a target in 2024 of establishing 500 self-driving laboratories by 2030 (Hwang et al., 2026).

Finally, the opinion refers to the Advanced Materials Act, the legislative proposal that the Commission expects to present by the end of 2026 (European Parliament, 2026). According to the European Commission (2025), the Act is intended to reduce the time needed to develop advanced materials, an objective to which, in the advisors' view, self-driving laboratories can contribute (Group of Chief Scientific Advisors, 2026).

Sources

European Commission (2024): Advanced Materials for Industrial Leadership, Communication COM(2024) 98, 27 February 2024

European Commission (2025): Commission seeks feedback on future Advanced Materials Act, news item, 21 October 2025

Group of Chief Scientific Advisors (2026): Advanced Materials, Scientific Opinion No. 18, April 2026

European Parliament (2026): Advanced Materials Act, Legislative Train Schedule, update of 20 September 2026

Nature Synthesis (2026a): The road to self-driving laboratories in industry, editorial, 8 October 2026

Nature Synthesis (2026b): Autonomous laboratories and digital chemistry in industry, interview with Nessa Carson by Peter Seavill, 20 July 2026

Nature Synthesis (2026c): Self-driving labs for research and development in industry, interview with Youn-Suk Choi by Jet-Sing Lee, 22 May 2026

Dvivedi, S., Minocha, N. et al. (2026): Towards self-driving laboratories in the biopharmaceutical industry, Nature Synthesis, 4 August 2026

Chen, L., Li, X., Lin, Q., Jiang, J. (2026): Self-driving laboratories need an autonomy safety harness, Nature Synthesis, 22 July 2026

Wang, Z., Wang, X. (2026): Scale-aware self-driving laboratories for multiscale process development, Nature Synthesis, 29 July 2026

Sutherland, B. R., Bernales, V., Aspuru-Guzik, A. (2026): Accelerating synthesis at scale requires a coordinated ecosystem, Nature Synthesis, 26 August 2026

Pilon, S., Savino, E., Bayley, O. M. et al. (2026): A flexible and affordable self-driving laboratory for automated reaction optimization, Nature Synthesis, 13 April 2026

Tobias, A. V., Wahab, A. (2025): Autonomous 'self-driving' laboratories: a review of technology and policy implications, Royal Society Open Science, 16 July 2025

Zhu, J., Huang, K., Pipe, J., Horbaczewsky, C., Tyrrell, A., Fairlamb, I. J. S. (2025): Rise of the Robochemist, arXiv preprint 2510.10337, October 2025

University of Cambridge (2009): Robot scientist becomes first machine to discover new scientific knowledge, press release, 2 April 2009

Burger, B., Maffettone, P. M. et al. (2020): A mobile robotic chemist, Nature, 8 July 2020

University of Liverpool (2020): Experiment too big? Hire a mobile robot scientist, press release, 8 July 2020

Hwang, J. et al. (2026): Self-driving laboratories in Korea: a new era of autonomous discovery, Digital Discovery, 18 March 2026

Evonik (2026): Evonik wins CPHI Pharma Award 2026 for Self Driving Labs, press release, 7 October 2026

Covestro (2025a): Covestro to open automated laboratory for developing coating and adhesive formulations, press release, 25 March 2025

Covestro (2025b): Thomas Büsgen Interview, Status Next, 7 May 2025