# Fileset

[d4dd00387j.pdf](https://mdr.nims.go.jp/filesets/609d138c-8c98-4712-b6b0-28ce006c3c79/download)

## Creator

[Naruki Yoshikawa](https://orcid.org/0000-0003-1546-8709), [Yuki Asano](https://orcid.org/0000-0003-2115-1417), [Don N. Futaba](https://orcid.org/0000-0002-7083-2772), [Kanako Harada](https://orcid.org/0000-0002-0221-7890), [Taro Hitosugi](https://orcid.org/0000-0002-7795-0683), [Genki N. Kanda](https://orcid.org/0000-0002-6372-241X), [Shoichi Matsuda](https://orcid.org/0000-0002-0640-3404), [Yuuya Nagata](https://orcid.org/0000-0001-5926-5845), [Keisuke Nagato](https://orcid.org/0000-0003-2399-3087), [Masanobu Naito](https://orcid.org/0000-0001-7198-819X), [Tohru Natsume](https://orcid.org/0000-0002-1510-2582), [Kazunori Nishio](https://orcid.org/0000-0002-8201-358X), [Kanta Ono](https://orcid.org/0000-0002-3285-9093), [Haruka Ozaki](https://orcid.org/0000-0002-1606-2762), [Woosuck Shin](https://orcid.org/0000-0002-8394-962X), [Junichiro Shiomi](https://orcid.org/0000-0002-3552-4555), [Kunihiko Shizume](https://orcid.org/0000-0001-8655-5207), [Koichi Takahashi](https://orcid.org/0000-0002-4235-9914), Seiji Takeda, [Ichiro Takeuchi](https://orcid.org/0009-0005-1905-2366), [Ryo Tamura](https://orcid.org/0000-0002-0349-358X), [Koji Tsuda](https://orcid.org/0000-0002-4288-1606), [Yoshitaka Ushiku](https://orcid.org/0000-0002-9014-1389)

## Rights



## Other metadata

[Self-driving laboratories in Japan](https://mdr.nims.go.jp/datasets/23bf1018-936d-4f16-9837-ef7ad52c6e9a)

## Fulltext

Self-driving laboratories in Japanrsc.li/digitaldiscoveryAs featured in:See Naruki Yoshikawa et al., Digital Discovery, 2025, 4, 1384.Showcasing Japan’s advancements in self-driving laboratory development. Left: Maholo LabDroid from RIKEN Center for Biosystems Dynamics Research, Hyogo, Japan. Right: Robotic powder grinding system from The University of Osaka, Osaka, Japan.Self-driving laboratories in JapanThis perspective highlights Japan’s initiatives in self-driving laboratory (SDL) development, showcasing diverse applications across materials science, biology, chemistry, and software. In addition, it covers national funding programs, research communities and ecosystems, and industries supporting progress in this fi eld. Additionally, the perspective emphasizes the critical roles of education, standardization, and benchmarking in fostering the continued growth of SDL research. Image reproduced by permission of Yumiko Miyahara from Digital Discovery, 2025, 4, 1384–1403.The artwork by Yumiko Miyahara includes an image by Kanda et al., licensed under CC BY 4.0.Registered charity number: 207890DigitalDiscoveryPERSPECTIVEOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article OnlineView Journal  | View IssueSelf-driving laboaMedical Research Laboratory, Institute of ITokyo, Tokyo, Japan. E-mail: yoshikawa.narbDepartment of Mechanical Engineering, ThcInstitute of Engineering Innovation, The UndNational Institute of Advanced Industrial SJapaneGraduate School of Medicine, The UniversifGraduate School of Engineering, The UnivegDepartment of Chemistry, School of SciencehSchool of Materials and Chemical TechnolJapaniRIKEN Center for Biosystems Dynamics ResjLaboratory Automation Suppliers' AssociatikCenter for Green Research on Energy anInstitute for Materials Science (NIMS), IbaralCenter for Advanced Battery Collaborationand Environmental Materials, National InIbaraki, JapanmInstitute for Chemical Reaction Design anUniversity, Hokkaido, JapanCite this: Digital Discovery, 2025, 4,1384Received 9th December 2024Accepted 14th May 2025DOI: 10.1039/d4dd00387jrsc.li/digitaldiscovery1384 | Digital Discovery, 2025, 4, 138ratories in JapanNaruki Yoshikawa, *a Yuki Asano, bc Don N. Futaba, d Kanako Harada, efTaro Hitosugi, gh Genki N. Kanda, aij Shoichi Matsuda, kl Yuuya Nagata, mnKeisuke Nagato, bf Masanobu Naito, o Tohru Natsume, pq Kazunori Nishio, hKanta Ono, r Haruka Ozaki, ijst Woosuck Shin, d Junichiro Shiomi, bcuKunihiko Shizume, c Koichi Takahashi, i Seiji Takeda,v Ichiro Takeuchi, uwRyo Tamura, xy Koji Tsuda uxy and Yoshitaka Ushiku zaaSelf-driving laboratories (SDLs) are transforming the scientific discovery process worldwide by integratingautomated experimentation with data-driven decision-making. Japan, known for its automation industry,is actively contributing to this field. This perspective introduces Japan's efforts in SDL development,including diverse applications across materials science, biology, chemistry, and software. In addition, itcovers national funding programs, research communities, and Japanese industries supporting progressin this field. It also highlights the importance of education, standardization, and benchmarking for thefuture growth of SDL research.1 IntroductionSelf-driving laboratories (SDLs)1 are transforming the process ofscientic discovery. It involves the automation of experimentsfor large-scale data generation and data-driven decision-makingfor efficient exploration of the candidate space. As globalinterest in SDLs continues to grow, Japanese researchers areactively contributing to the eld. This perspective offers anoverview of research efforts related to SDLs in Japan.Japan has been renowned for its advanced automationtechnology, holding a 46% share of the global industrial robotmarket as of 2023.2 This background has fostered an affinity forlaboratory automation research, where robots oen playntegrated Research, Institute of Scienceuki@tmd.ac.jpe University of Tokyo, Tokyo, Japaniversity of Tokyo, Tokyo, Japancience and Technology (AIST), Ibaraki,ty of Tokyo, Tokyo, Japanrsity of Tokyo, Japan, The University of Tokyo, Tokyo, Japanogy, Institute of Science Tokyo, Tokyo,earch, Hyogo, Japanon, Hyogo, Japand Environmental Materials, Nationalki, Japan, Center for Green Research on Energystitute for Materials Science (NIMS),d Discovery (WPI-ICReDD), Hokkaido4–1403a central role. In 1988, Matsuda et al. demonstrated the opti-mization of reaction conditions using an automated system.3This system can be considered as one of the earliest SDLs inJapan, as it incorporates a laboratory robot with decision-making by the simplex method. A fully automated laboratorysystem for testing blood samples built by a Japanese hospital inthe 1980s is reported by Sasaki et al.4 Their approach gainedglobal attention as a promising method to reduce laboratorytesting costs.5The development of SDLs could offer valuable solutions toaddress Japan's social challenges, particularly its decliningbirth rate and shrinking workforce. This demographic shi hascreated a need for innovative solutions to maintain productivitynJST, ERATO Maeda Articial Intelligence in Chemical Reaction Design andDiscovery Project, Hokkaido, JapanoResearch Center for Macromolecules and Biomaterials, National Institute forMaterials Science (NIMS), Ibaraki, JapanpNational Institute of Advanced Industrial Science and Technology (AIST), Tokyo,JapanqRobotic Biology Institute Inc., Tokyo, JapanrDepartment of Applied Physics, The University of Osaka, Osaka, JapansInstitute of Medicine, University of Tsukuba, Ibaraki, JapantCenter for Articial Intelligence Research, University of Tsukuba, Ibaraki, JapanuRIKEN Center for Advanced Intelligence Project, Tokyo, JapanvIBM Research, Tokyo, JapanwGraduate School of Engineering, Nagoya University, Aichi, JapanxCenter for Basic Research on Materials, National Institute for Materials Science,Ibaraki, JapanyGraduate School of Frontier Sciences, The University of Tokyo, Chiba, JapanzNexaScience, Inc., Tokyo, JapanaaOMRON SINIC X Corp., Tokyo, Japan© 2025 The Author(s). Published by the Royal Society of Chemistryhttp://crossmark.crossref.org/dialog/?doi=10.1039/d4dd00387j&domain=pdf&date_stamp=2025-06-07http://orcid.org/0000-0003-1546-8709http://orcid.org/0000-0003-2115-1417http://orcid.org/0000-0002-7083-2772http://orcid.org/0000-0002-0221-7890http://orcid.org/0000-0002-7795-0683http://orcid.org/0000-0002-6372-241Xhttp://orcid.org/0000-0002-0640-3404http://orcid.org/0000-0001-5926-5845http://orcid.org/0000-0003-2399-3087http://orcid.org/0000-0001-7198-819Xhttp://orcid.org/0000-0002-1510-2582http://orcid.org/0000-0002-8201-358Xhttp://orcid.org/0000-0002-3285-9093http://orcid.org/0000-0002-1606-2762http://orcid.org/0000-0002-8394-962Xhttp://orcid.org/0000-0002-3552-4555http://orcid.org/0000-0001-8655-5207http://orcid.org/0000-0002-4235-9914http://orcid.org/0009-0005-1905-2366http://orcid.org/0000-0002-0349-358Xhttp://orcid.org/0000-0002-4288-1606http://orcid.org/0000-0002-9014-1389http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jhttps://pubs.rsc.org/en/journals/journal/DDhttps://pubs.rsc.org/en/journals/journal/DD?issueid=DD004006Perspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Onlinewith fewer people. SDLs can reduce the burden of labor-intensive experimental work in laboratories, enabling researchto continue with fewer staff members. Additionally, SDLsimprove researchers' work-life balance, an increasingly impor-tant consideration in Japan's work culture. Moreover, Japan'sdemographic shi threatens the transmission of technicalexpertise to future generations. SDLs can help preserve thespecialized knowledge of experienced professionals by auto-mating tasks and replicating their skills. While these advan-tages are particularly relevant to Japan, other countries facingsimilar demographic trends may also benet from SDLs in thenear future.The remainder of this paper is structured as follows. Tointroduce the current state of automation in Japan, we providea review of Japanese SDLs across three application areas—materials science, biology, and organic chemistry in Section 2 to4. Section 5 explores the soware aspects of SDLs by intro-ducing research efforts on AI for scientic discovery. Section 6highlights national funding programs to promote advance-ments in automation-focused studies. Section 7 addresses theactivities to form research communities and ecosystems. Anoverview of the Japanese industries supporting SDL develop-ment is provided in Section 8. Finally, Section 9 discusses thefuture directions of SDLs, and Section 10 concludes the paper.The geographical locations of SDLs introduced in this article areshown in Fig. 1. This perspective is based on a workshop held atthe Institute of Science Tokyo in October 2024.2 Materials scienceThe materials industry is a key sector of Japan's economy, andactive research in materials science is being conducted there. Aspart of the government's strategy to strengthen materialsinnovation, data-driven researchmethods are actively promotedthrough projects like the DxMT.6 This section reviews efforts inJapan to advance SDLs in materials science. Section 2.1 outlinesan automated system for synthesizing and evaluating thin-lmmaterials. Section 2.2 provides an overview of the MaiMLformat, a standardized data format for measurement analysisinstruments. Section 2.3 details a robotic experiment setup fordiscovering electrochemical materials. Section 2.4 and 2.5highlight autonomous polymer synthesis achieved by tworesearch groups. Section 2.6 explores the development ofFig. 1 Geographical locations of SDLs introduced in this article.© 2025 The Author(s). Published by the Royal Society of Chemistryradiative cooling materials. Applications of robot arms formechanochemical synthesis and autonomous X-ray diffractionanalysis are covered in Section 2.7. Finally, Section 2.8 intro-duces Process informatics.2.1 Autonomous experiments for thin-lm materialsThin-lm research is important for a wide range of elds,including semiconductor devices, sensors, catalysts, optics, andvarious coatings. Shimizu, Hitosugi, and colleagues reporteda closed-loop system for inorganic thin-lm materials in 2020.7The system combines Bayesian optimization, automatedsynthesis, and automated physical property evaluation(Fig. 2(a)). The system consists of a robot arm positioned at thecenter of a hexagonal chamber, which is connected to sixsatellite chambers with an automated sputter thin-lmsynthesis equipment and an automated electrical resistanceevaluation system (Fig. 2(b)). The robot arm handles all sampletransfers between the satellite chambers. Autonomous experi-ments aimed at minimizing the electrical resistance of Nb-doped TiO2 thin lms achieved a throughput 10 times higherthan manual methods.7 Furthermore, the system discovereda novel electrolyte material for all-solid-state Li batteries.Specically, by mixing Li3PO4 and Li1.5Al0.5Ge1.5(PO4)3(Fig. 2(c)), an amorphous thin lm (Li1.8Al0.03Ge0.05PO3.3) showshigher Li-ion conductivity than either of the original materials[Li3PO4 and Li1.5Al0.5Ge1.5(PO4)3] (Fig. 2(d)).8To reduce the number of experiments in the Bayesian opti-mization process, the hyperparameters of the kernel andacquisition functions were tuned.9,10 Leveraging the knowledgeand expertise of materials researchers is essential for tuning.Fig. 2 (a) Photograph and (b) schematic of autonomous experimentalsystem for exploration of inorganic thin-filmmaterials. Copyright 2020Shimizu et al.7 and reprinted with permission under CC BY 4.0. (c)Autonomous experimental cycle for exploration of ionic conductors.(d) Ionic transport properties of fabricated amorphous ionic conduc-tors thin films. Reprinted with permission from Kobayashi et al.8Copyright 2023 American Chemical Society.Digital Discovery, 2025, 4, 1384–1403 | 1385http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jFig. 3 Schematic of the utilization of the standardized data format ofMeasurement, Analysis, Instrument Markup Language (MaiML).Fig. 4 Schematic illustration of the data-driven high-throughputautomated robotic experiments for searching multi-componentselectrolyte for rechargeable batteries. Figure adapted with permissionfrom Matsuda et al.18 Copyright 2022, Elsevier.Digital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article OnlineFor example, materials researchers can anticipate the processwindow of synthesis parameters and the scale of changes inphysical properties, which is the key to tuning the values ofhyperparameters.Since this system can connect multiple measurement andanalysis instruments, it can acquire various physical propertiesfrom multiple aspects to generate big data. For this purpose,a standardized measurement and analysis data format(Measurement Analysis Instrument Markup Language: MaiML)was applied to the system.11 The format is described in the nextsection. For interested readers, a review on autonomousexperimental systems in materials science is available fromIshizuki et al.122.2 Standardization of data format with MaiMLRepresenting data in a standardized, structured format iscrucial for facilitating automated analysis by computers.Various data formats have been developed to store differenttypes of information. For example, Chemical DescriptionLanguage (cDL) has been introduced to describe experimentalprocedures in organic chemistry.13 Currently, measurementinstruments from different manufacturers oen provide data indifferent formats. This lack of a standard format requires usersto convert the formats manually or to prepare data conversionsoware. Therefore, there is a strong need for a standardformat. In response, the Japan Analytical Instruments Manu-facturers Association (JAIMA), in collaboration with its membercompanies and the Ministry of Economy, Trade, and Industry(METI), established a data format called the MeasurementAnalysis Instrument Markup Language (MaiML). In May 2024,MaiML was registered as a Japanese Industrial Standard (JIS K0200).The MaiML format was developed as a standardized dataformat with independent availability to achieve an instrument-agnostic data structure. The format follows the ndable,accessible, interoperable, and reusable (FAIR) data principles.14An XML format describes the processes of measurement, pre-processing, and postprocessing steps. Detailed descriptions ofsample fabrication processes and measurement conditionsensure the reproducibility of experiments. Additionally, logs foreach measurement operation provide traceability. The formatalso includes tamper-prevention features and data encryptioncapabilities. These features allow MaiML to encompass essen-tial information for the reproducibility of sample fabrication tomeasurement and analysis, thus contributing to databaseconstruction (Fig. 3). Guidelines for MaiML are available onhttps://www.maiml.org/.2.3 High throughput robotic experiments for rechargeablebatteriesData-driven automated robotic experiments are an effectivemethod to accelerate the development of new-materials even inthe eld of rechargeable batteries.15 Specically, optimizing thecomposition of electrolytes and identifying effective additivecombinations involves evaluating an enormous number ofpotential candidate materials.16 Historically, this process has1386 | Digital Discovery, 2025, 4, 1384–1403relied heavily on trial-and-error approaches, leading to signi-cant bottlenecks in the development of new electrolytematerials.To overcome these challenges, Matsuda et al. developed therobotic experimental setup for searching electrochemicalmaterials discovery using high-throughput combinatorialtechniques by use of miniaturized microplate type electro-chemical cells.17 The system consisted of a liquid handlingdispenser and a 96-channel electrochemical analyzer equippedwith a robotic microplate handling arm, with a searchthroughput of over 1000 samples per day. By integrating withBayesian optimization techniques, they discovered the speciccomposition of electrolyte that enhances the cycle life oflithium-oxygen batteries,18 demonstrating its effectiveness insignicantly accelerating the identication of optimal electro-lyte compositions (Fig. 4). A key feature of their system is its© 2025 The Author(s). Published by the Royal Society of Chemistryhttps://www.maiml.org/http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jFig. 5 (A) Press process automation system for polymer materialsdevelopment, adapted from Asano et al.28 (B) Dielectric propertymeasurement system. Reprinted with permission from Asano et al.29Copyright 2024, IEEE. (C) Automation system for heat transfer mate-rials development.Perspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Onlineautonomous exploration capability, which is achieved with theuse of NIMO, an automation soware for establishing a closed-loop workow between articial intelligence and roboticexperimentation.19 The details of NIMO are described in Section5.2. Recently, the group extended their interest for practicaldesign of battery cells, and reported the development of fullyautomated sequential robotic experimental setup for the cellfabrication of stacked-type battery cells with fabricationthroughput over 80 cells per day, which is 10 times higher thanconventional human-based experiments.202.4 Advances in autonomous synthesis for polymersAutonomous synthesis in polymer science has progressedsignicantly since the early 2000s, with high-throughput (HTP)techniques driving advances in polymerization. Early develop-ments focused on automating polymer synthesis processes,particularly for combinatorial studies, to enhance speed andreproducibility.21 A major milestone was the application ofautomation to precision-controlled polymerization, such asReversible Addition–Fragmentation Chain Transfer (RAFT) andAtom Transfer Radical Polymerization (ATRP). These tech-niques allow precise control over polymer architecture, molec-ular weight, and functionality, enabling the creation of complexand customized polymers.22,23 HTP methodologies have facili-tated the rapid creation of polymer libraries and optimization ofsynthesis parameters, though dataset sizes remain a limitation.Recent innovations in platforms like Chemspeed have repli-cated most manual processes, integrating Python-based toolslike Chemspyd24 for real-time adaptive control and processoptimization.Despite these advancements, challenges persist in charac-terizing critical physical properties, such as mechanical andthermal performance. For example, tensile testing for adhesivematerials requires specialized setups and skilled sample prep-aration. Addressing this, Naito and Sato developed a high-throughput testing system for adhesives, which uses miniatur-ized specimens to provide more realistic performance metricswhile reducing material usage.25 Moreover, integrating machinelearning techniques, such as Bayesian optimization, with owsynthesis has enabled autonomous experiments for optimizingradical polymerization.26,27 In summary, while autonomoussynthesis has revolutionized polymer research, the integrationof automation and intelligent algorithms promises furtheradvancements, addressing current challenges and unlockingnew possibilities for material innovation.2.5 Flexible lab-automation using robot arms for polymermaterials developmentPolymer materials are widely used in academia and industry.Most polymer materials used for actual products are in the formof composites to reinforce multi-functionality, for instance bybeing mixed with dielectric llers. However, automation in thedevelopment processes of polymer composite materials is stilllimited because of the challenges in handling materials in theform of powders and granules and molding processes requiredfor property characterization and application.© 2025 The Author(s). Published by the Royal Society of ChemistryA Japanese team has been developing an automated systemfor polymer materials development. One of the sub-processes isthe press process (Fig. 5(A)).28 A robot arm was adopted toconstruct a system that handles the tools for the operations,such as press plates and forks, and to increase the exibility ofthe system for future adaptability rather than built-in automa-tion of a dedicated system. The control soware operates boththe robots and the press machines. An experimental closed loopwas formed to obtain effective press parameters and evaluatesthe thickness of the polymer lm by image processing and pressparameters. Another automated sub-process is the propertymeasurement systems (Fig. 5(B)), such as the one for a dielectricproperty utilizing the force-sensing capability of a robot arm forstabilization of polymer placement.29 The automated systemsuccessfully measured the dielectric properties with the sameaccuracy as trained humans.The exibility of the system was enhanced by the use ofa gripper interface that enables the grasping of multiple toolsand thus the completion of complex pressing processes witha single robot arm. The same type of robot arms was used inboth the press and measurement sub-processes. The use ofa single type of robot arm reduces the development and main-tenance costs for an automated workow than the developmentof multiple dedicated automation machines for each process.2.6 Automated lm development system generating massivedata for radiative coolingAmid growing concerns over urban heat islands, sky radiatortechnology, which selectively emits thermal radiation in theDigital Discovery, 2025, 4, 1384–1403 | 1387http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jFig. 6 Automated system for powder material experiments. (A)Robotic powder grinding system using a Soft Jig, where the jig'ssoftness ensures safe grinding without the need for force sensing. (B)Autonomous powder X-ray diffraction system for preparing XRDsamples and performing automated Rietveld analysis. (C) Enhancedrobotic powder grinding using visual and audio feedback, with grindingsounds providing particle size information for more efficient grinding.(D) Robotic mechanochemical synthesis controlling reaction path-ways through force conditions applied by a pestle.Digital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Onlineatmospheric window region (8–13 mm) to directly radiate heatinto outer space, has received signicant attention. Shiomi et al.have been developing thermal radiative metamaterials bycombining electromagnetic eld analysis andmachine learningto optimize metamaterials, achieving high-performancethermal emitters.30,31 While metamaterials are advantageousin the wavelength selectivity of radiation, the difficulty infabrication poses a signicant barrier to widespreadapplication.As a scalable and cost-effective approach to developingradiative coolers, they are working on organic–inorganic hybridcoatings consisting of a polymer matrix and inorganic llers.The thermal radiation properties in the atmospheric windowregion and the solar reection properties in the visible range forheat shielding depend on the optical absorption of the polymersand inorganic llers themselves, along with the resonances andscattering phenomena occurring at their interfaces. Conse-quently, the parameter space becomes extremely large becauseit incorporates not only material parameters, such as the typeand mixing ratio of polymers and inorganic llers, but alsoprocess parameters, such as casting and drying conditions. Forexample, a coating involving ve materials and three processparameters, such as mixing time, casting speed, and dryingtemperature, yields an 8-dimensional search space. Exploringjust 10 choices per parameter results in 108 combinations,leading to a combinatorial explosion. To address this challengewith high throughput, they are developing an automatedcoating system capable of producing 1000 coatings per dayunder various conditions and automatically acquiring infraredand visible reection spectra (Fig. 5(C)). Additionally, they havedeveloped a spectral prediction model, useful for controllingradiative properties, using a dataset of over 10 000 data.2.7 Robotic mechanochemical synthesis and autonomousXRD analysisMechanochemical synthesis, which induces chemical reactionsthrough mechanical force, offers an energy-efficient, solvent-free method for producing materials such as metal–organicframeworks (MOFs) and energy-related compounds. However,traditional methods like manual grinding or ball milling oenstruggle with reproducibility and control. To overcome thesechallenges, Nakajima et al. developed a force-controlled roboticmechanochemical synthesis system,32 combined with anautonomous X-ray diffraction (XRD) analysis workow33 (Fig. 6).This system not only provides precise control over grindingforce and speed but also enables automated, high-throughputstructural analysis through autonomous XRD.In their experiments with perovskite materials, the roboticsynthesis system demonstrated superior reproducibilitycompared to manual and ball milling methods, especially forforce-sensitive reactions. By adjusting the grinding force andspeed, they could signicantly inuence the reaction pathways,allowing for precise control of reaction outcomes. For instance,increased grinding force produced higher yields of Cs4PbBr6,while variations in speed shied the reaction toward otherphases like CsPbBr3.1388 | Digital Discovery, 2025, 4, 1384–1403Once the synthesis was complete, the fully autonomous XRDsystem seamlessly handled sample preparation, measurement,and data analysis. This integration allowed for high-throughputanalysis and minimized human error, particularly in thereproducibility of low-angle diffraction patterns, which arecrucial for characterizing materials like lead halide perovskites.This combined robotic synthesis and autonomous XRDapproach offers a powerful tool for both advancing the under-standing of reaction mechanisms and accelerating thediscovery of novel materials. Future work will explore theapplication of this system to a wider range of materials.2.8 Process informatics – robotic objective processexploration system (ROPES)In order for new materials to be incorporated into nal prod-ucts, process development and production technology devel-opment are necessary, and AI robot-driven development is alsoeffective (Fig. 7(a)). Process Informatics is located downstreamof Materials Informatics. Experimental-based Bayesian optimi-zation was demonstrated in the powder-lm-dying process ofa catalyst layer in polymer electrolyte fuel cells (PEFCs).34The catalyst layer of a solid polymer electrolyte fuel cell iscomposed of carbon as an electronic conductor, uoropolymeras a proton conductor, pores in the gas diffusion space, andplatinum nanoparticles as a reaction catalyst. The arrangementof the three-dimensional microstructure changes signicantlydepending on how it is applied and dried, and optimization isrequired. The autonomous experiment system shown inFig. 7(b) discovered new drying-process parameters among 85candidates with 40 trials minimizing the defect ratio (Fig. 7(c)).Not only the high-throughput exploration but also ve process© 2025 The Author(s). Published by the Royal Society of Chemistryhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jFig. 7 ROPES of powder-film-formation-process, (a) schematic of anautonomous system, (b) experimental setup with two robot hands,36(c) results of exploration of drying process parameters minimizingdefect ratio,34 (d) a scalable autonomous system, (e) high-throughputprototyped fuel cell catalyst layer samples. A video about ROPES isavailable online.35Fig. 8 The LabDroid Maholo including peripheral equipment.Reprinted from Kanda et al.38 under CC BY 4.0.Perspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Onlineroutes were found. Furthermore, we also prototyped an elevator-sized autonomous system as shown in Fig. 7(d), which canautomatically demonstrate factory processes, i.e., die-coatingand zone-heating, and evaluate defect ratio, but also micron-sized surface roughness and electronic impedance. The small-sized samples are prototyped with various process parameters(Fig. 7(e)). A video demonstration of this system is availableonline.35Process informatics is a methodology for process explorationusing small amounts of prototyped material samples, whichwill lead to the accelerated social implementation of incorpo-ration into nal products.3 BiologyAutomation in biological experiments has been pursued acrossvarious elds and processes. For example, PCR experiments—widely recognized due to COVID-19 testing—had already beguntransitioning from manual water bath operations to automatedthermal cyclers by the late 1980s.37 Numerous specializedinstruments like automated pipetting systems have beendeveloped, signicantly improving the efficiency of xed oper-ations. However, more versatile robotic systems are needed,especially for automation in basic research. In Japan, theversatile humanoid robot Maholo LabDroid, developed byYaskawa Electric Corporation and the Robotic Biology InstituteInc., is widely used from fundamental to clinical research(Fig. 8).© 2025 The Author(s). Published by the Royal Society of Chemistry3.1 Maholo LabDroidMaholo LabDroid has been employed in molecular and cellularbiology, drug screening, culturing and immunostaining ofinduced pluripotent stem cells (iPS cells), and in closed-loopcell culture using AI and robotics.39–43 By combining this robotwith optimization AI, Kanda et al., have successfully conductedautonomous experiments targeting cell cultures for regenera-tive medicine.38 Specically, they used a batch Bayesian opti-mization algorithm with LabDroid Maholo to autonomouslyexplore combinations of seven parameters—such as reagentconcentrations, processing times, and cell handling intensi-ties—involved in differentiating iPS cells into retinal cells,achieving efficient induction without human intervention.Beyond basic research, Maholo LabDroid is also utilized inclinical studies. A research team at Kobe City Eye Hospitalcreated a sterile environment by integrating the robot witha clean booth and successfully transplanted cells cultured bythe robot into patients during clinical research on retinal cells.44As of October 2024, Maholo LabDroid is not sold outside Japan.3.2 Robotic crowd biologyA Japan-led team has proposed the concept of Robotic crowdbiology in 2017, which involves aggregating hundreds of robotsinto a robotic experimentation center for cloud-based experi-ments.45 Researchers submit their desired experimental proto-cols to the center via a network, where AI and robots efficientlyexecute the experiments and return the results. This approachoffers various benets, such as improving the reproducibilityand traceability of scientic experiments, fundamentallyresolving research misconduct, increasing the utilization ratesof expensive advanced equipment, and efficiently conductinghigh-biosafety-level experiments. Ultimately, it envisionsmaking biological research widely accessible to all humanity.In the Robotic Crowd Biology concept, managing numerousrobots and devices makes it impractical for humans to instructeach one individually. To address this, the development of AIDigital Discovery, 2025, 4, 1384–1403 | 1389http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jDigital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Onlineand soware for directing robots is actively underway. Aresearch team at the University of Tsukuba has developedalgorithms to achieve parallel scheduling necessary for effi-ciently operating multiple robots.46,47 While some schedulingalgorithms exist in the eld of factory automation (FA), theauthors' formulation considers time constraints specic to lifescience experiments—such as reaction times and the degrada-tion of living cells and reagents—that are set among processes.To automate the monitoring of multiple devices that wouldotherwise require human oversight, the same team developedimage recognition soware, ne-tuned for identifying labwareused in life science experiments, to monitor the status of lab-ware placed inside automated dispensing robots.48 Additionally,a research team at RIKEN has demonstrated that it is possible toautomatically convert experimental procedures written innatural language into robot-operating code using largelanguage models.49 Furthermore, the development of AI-basedsystems to manage robots and experimental equipment fromhigher levels is currently being extensively pursued. Supportingthese efforts, a prototyping lab has been established at RIKENBDR in Kobe, and a demonstration facility is planned at theInstitute of Science Tokyo. In recent years, researchers outsideJapan have also begun exploring ways to take advantage ofa cloud-based experimental platform, spurring discussions andinitiatives worldwide toward its realization.504 Organic chemistryIn organic chemistry research, experimental procedures stilllargely depend on researchers' expertise and manual opera-tions. However, there is a continuous demand for more efficientalternatives to these traditional methods, resulting in thedevelopment of various innovative approaches. Recently, auto-mated synthesis robots have drawn signicant attention fortheir potential to automate and even autonomously conductorganic chemistry research. Utilizing these robots can achievehigh reproducibility and experimental precision, offeringsubstantial improvements in efficiency compared to conven-tional manual processes. This technological advancementsimplies labor-intensive synthetic experiments and consider-ably reduces the workload of researchers.In the eld of organic synthesis, two primary types of robotsare commonly employed: articulated robots and Cartesiancoordinate robots. Here, articulated robots are highly exible,with multiple joints that enable complex, multi-directionalmovements. This makes them ideal for intricate tasks, suchas transferring reaction vessels, adjusting equipment, andperforming precise reagent additions in conned spaces. Incontrast, Cartesian coordinate robots operate along xed linearaxes, making them well-suited for high-precision, repeatabletasks like liquid handling, reagent dispensing, and automatedsample preparation with minimal positioning errors. Manyprocesses in organic synthesis can oen be segmented intosimple operations that are well-suited for execution by Carte-sian coordinate robots. For example, the polymerization ofpoly(quinoxaline-2,3-diyl)s via living polymerization of diiso-cyanobenzene derivatives has been successfully automated1390 | Digital Discovery, 2025, 4, 1384–1403using a Cartesian coordinate robot.51,52 In this process, theresulting polymer thin lms were reported to exhibit uniqueselective reection behavior. It was found that even slightinaccuracies in the monomer composition and variations in thedegree of polymerization had a signicant impact on theselective reection wavelength. Therefore, precise control ofthese parameters was critically important. In particular, it wasnecessary to dispense volumes with an accuracy of less than10 mL. However, in the early 2010s, among the commerciallyavailable automated systems investigated in the study, noarticulated robot was known to achieve this level of dispensingprecision with organic solvents. Consequently, a Cartesianrobot (Chemspeed SWING) was employed for this purpose.Additionally, direct integration of Cartesian coordinaterobots with various analytical instruments is under activeinvestigation. For instance, integration with a UV-visible-NIRspectrophotometer has enabled the development of a solu-bility prediction model for porphyrins,53 while integration withchromatography systems has facilitated the automatic evalua-tion of asymmetric catalysts, advancing the development ofa high-performance catalyst.54 More recently, combining robots,chromatography systems, and the PHYSBO package introducedin Section 5.2 has been explored for autonomous optimizationof chemical reaction conditions.55Beyond commercial laboratory automation systems, low-costhardware is crucial for reducing the entry barrier to SDLs.56Kuwahara et al. developed a 3D-printed robot named FLUID todemocratize automation in materials synthesis.57 They showedits utility by demonstrating the coprecipitation of cobalt andnickel to form binary materials. All design les and controlsoware are released under an open-source license, allowingusers to modify and adapt the system to their own researchenvironments.5 AI for scienceSince SDLs involve the automation of data-driven decision-making,1 SDL research requires developing intelligent sowareas well as automation hardware. In this section, we review thesoware side of SDLs in Japan, focusing on the eld of AI forscience. The foundation models for material discovery devel-oped in IBM Research-Tokyo are described in Section 5.1. Blackbox optimization soware packages developed by Japaneseteams are introduced in Section 5.2. Research by OMRON SINICX is covered in Section 5.3 and several applications of largelanguage models for scientic research are outlined inSection 5.4.5.1 Foundation models for material discoveryThe integration of AI models into self-driving laboratoriesenhances their capabilities, enabling the preselection ofpromising materials before chemical synthesis and guidingexperiments to achieve desirable properties. Property predictionand structural generation are particularly promising AI appli-cations. However, traditional AI models developed by individualresearch groups within specic material domains are oen© 2025 The Author(s). Published by the Royal Society of Chemistryhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jFig. 9 Black box optimization methods depending on the aim ofexploration. Reprinted from ref. 71–73 under CC BY 4.0.Perspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Onlinelimited in size and small training datasets (order of 10 to 100samples), resulting in insufficient modeling accuracy.To address these limitations, recent advancements inspiredby large language models (LLMs) have been adopted in mate-rials informatics. These approaches involve pre-training largemodels in a self-supervised manner using massive datasets.Such pre-trained models effectively capture general materialrepresentations and can be ne-tuned with small domain-specic datasets for various downstream applications. Exam-ples of foundation models include MolCLR,58 a GNN-basedarchitecture pre-trained with 10 million molecular samples,and ChemBERTa,59 which uses the RoBERTa architecture pre-trained with 77 million SMILES samples. Although thesemodels perform well in regression and classication tasks, theirsingle-modal nature leaves room for improvement. Recentstudies explore multi-modal representations of materials toenhance modeling capabilities. Shirasuna et al.60 introducedmodels that fuse SMILES andmolecular graphs using aMixture-of-Experts approach, achieving superior performance oversingle-modal models. As noted in Takeda et al.,61 multi-modalmodeling is an emerging eld with vast opportunities,including efficient fusion methods and strategic modalityselection. Other modalities, such as SELFIES, 3D atom posi-tions, electron density, optical spectra, and text descriptions,are also under consideration. These foundation models cansignicantly enhance SDL capabilities by utilizing theirpredictive and generative functions, enabling the design ofmore promising candidate materials. For instance, ref. 62demonstrates how a SMILES-based foundation model wasintegrated into a human-in-the-loop workow connected to anSDL. Similarly, as demonstrated by RoboRXN,63 syntheticpathways predicted by a foundation model can be executed inautomated robotic laboratories. Some of these models areopenly accessible on GitHub64 and Hugging Face, encouragingopen development within the materials informatics commu-nity. Specically, Foundation Model for Materials (FM4M) hasachieved widespread adoption through active community-building efforts through the AI Alliance,65 bridging academiaand industry toward shared innovation goals.5.2 Black box optimization methods and NIMOBlack box optimization techniques are useful as an AI to suggestexperimental conditions to be tested, which can be the brain ofself-driving labs (Fig. 9). Bayesian optimization is probably thebest-known technique for achieving desired material proper-ties. The Python packages COMBO (COMmon Bayesian opti-mization)66 and PHYSBO (optimization tool for PHYSics basedon Bayesian optimization)67 can quickly perform the Bayesianoptimization calculations. On the other hand, there are variousneeds in materials research, but also the improvement ofmaterial properties, and other techniques are required. Forexample, to construct phase diagrams with a small number ofexperiments, the Python package PDC (Phase DiagramConstruction) has been developed by adopting the uncertaintysampling strategy.68 To visually explore phase diagrams usingPDC, a web application called AIPHAD (Articial Intelligence© 2025 The Author(s). Published by the Royal Society of Chemistrytechniques for PHAse diagram) is freely available.69,70 In addi-tion, algorithms such as BLOX (BoundLess Objective-freeeXploration)71 for overlooking the material property spacesand SLEPA (Self-Learning Entropic Population Annealing)72 forobtaining the material property distributions with the smallnumber of experiments have been developed as open sourcesoware. Furthermore, the black box optimization techniqueusing quantum annealer and Ising machines called FMQA(Factorization Machine with Quantum Annealing) has beendeveloped to explore vast material space.73 Although thesemethods have been mainly developed in materials science, webelieve that they can be used not only in materials science butalso in any self-driving labs for biology, organic chemistry, etc.To achieve a self-driving lab by combining the roboticexperimental devices introduced in Section 2.3 and three black-box optimization techniques (PHYSBO, PDC, and BLOX),a generic soware NIMO (NIMS Orchestration System) has beendeveloped.19,74 In NIMO, a robotic experiment and a black-boxoptimization method are treated as modules, and the systemis designed to enable various autonomous automated materialexplorations by selecting these modules. As a demonstrationexperiment, autonomous automated experiments on electro-lytes for lithium metal electrodes were carried out using therobotic experimental setup for searching electrochemicalmaterials discovery (see Section 2.3) controlled by NIMO. Asa result, a total of 384 electrolytes were successfully developedDigital Discovery, 2025, 4, 1384–1403 | 1391http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jDigital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Onlineas autonomous automated experiments without human inter-vention using NIMO. Globally, various types of orchestrationsoware (OS) have been developed.75–77 NIMO has an advantageover other OSs due to its focus on AI algorithms, particularly itsuse of diverse black-box optimization methods, beyondBayesian optimization. These methods are designed to be easilyintegrated into other OSs, allowing us to enhance the capabil-ities of self-driving laboratories by combining NIMO with othersystems. Additionally, we are developing new AI algorithms toaddress a range of exploration needs. By incorporating theseinto NIMO, we aim to establish it as an evolving open-sourcesoware (OSS). Furthermore, the three algorithms alreadyimplemented in NIMO are widely used in materials science,78chemistry,79 and drug discovery,80 leveraging both experimentaland simulation data. Thus, we believe that NIMO can facilitateinteractions between these different domains in Japan.5.3 From foundation model to AI scientistAs discussed in Section 5.1, foundation models are beingdeveloped across various elds, with comprehensive applica-tions being considered for SDLs. These applications extendbeyond providing scientic knowledge in respective elds;related to Section 5.2, LLMs can also encompass optimizationtasks,81 and as will be discussed in Section 5.4, they includea series of research activities such as hypothesis generation,experimental design and implementation, paper writing, andreective review as AI scientist.82OMRON SINIC X Corporation, a research subsidiary ofOMRON that focuses on healthcare and factory automation, isadvancing various projects utilizing foundation models in AIand robotics to realize real-world AI scientists. Following thephilosophy of OMRON's founder, “To the machine, the work ofthe machine, to man the thrill of further creation”, the companyhas been researching how to understand and support humancreation through AI and robots. A particular focus has beenunderstanding human research and experimental work, con-ducting research and development through open innovationwith universities and public research institutions whilereceiving competitive research funding.The company's research into understanding research dataencompasses several key streams. The rst focuses on lawdiscovery, which involves tackling the symbolic regressionproblem to discover scientic laws between variables frommeasured data.83 This machine learning challenge takes tabulardata containing variable values as input and outputs mathe-matical formulas showing relationships between variables.84The second stream involves map-based visualization, primarilytargeting materials science, where representation learningtechniques are applied to various data types including materialstructures, measurement data, and property-describing text.85The goal is to create visualizations where materials with similarproperties are mapped close to each other. The third streamconcentrates on novel material design86 and property predic-tion,87 including research on generating new crystal structuresand developing Transformer architectures to predict propertiesof crystal structures with unknown characteristics.1392 | Digital Discovery, 2025, 4, 1384–1403Regarding the reproduction of experimental work, much ofthe robotics research has focused on powder manipulation.Powders present more challenging handling requirements thanliquids in terms of weighing, grinding, and mixing, makingthem an engaging research topic in robotics. For instance,research on powder-weighing robots has achieved sub-milligram precision in liing and dropping powder usingspoons through simulation-based learning.88 In powdergrinding research, efficient powder processing has been ach-ieved by utilizing multiple modalities of information, includingvisual and vibration data.89In parallel, research is being conducted on AI robots that canexecute various tasks while understanding environmental dataand linguistic instructions.90 Future plans involve connecting theabove-mentioned AI for research understanding with experi-mental automation robots, advancing this research to realize AIrobot scientists capable of conducting real-world experiments.5.4 Large language models for automated scientic researchLLMs have been adopted by multiple SDLs91,92 because of theirpowerful ability in natural language processing. A number ofapplications on LLM for automated scientic research havebeen released from Japan. Sakana AI, a Japan-based startup,proposed the AI Scientist,82,93 which aims for fully automaticscientic discovery by harnessing the power of LLMs.Hatakeyama-Sato et al. explored the ability of GPT-4 in variouschemical tasks and elucidated their current limitations.94 Theyalso applied GPT-4 for parameter selection of a polymer prop-erty prediction95 and a semiautomated system for synthesizingpolyamic acid particles.96 Jiang et al. developed ProtoCode,97a tool that can extract protocol information from naturallanguage text and convert it to intermediate representationformats. They demonstrated its ability by generating thermalcycler operation les from polymerase chain reaction (PCR)protocols written in natural language. Machi et al. developeda framework for reviewing the results of automated conversionsof structured organic synthesis procedures extracted from theliterature.98 In this framework, organic synthesis procedures inthe literature are transformed into a structured chemicaldescription language (cDL)13 using both a proposed rule-basedmethod and a generative large language model-based method.91The results from both methods are presented simultaneously tousers, facilitating efficient transformation and renement.As the potential of LLMs has become widely recognized, therace for their development has intensied. Both commercialmodels, such as GPT-4,99 and open-weight models like LLaMA100and DeepSeek,101 are now widely available. To support thedevelopment of LLMs in Japan, the Ministry of Economy, Tradeand Industry (METI) and the New Energy and Industrial Tech-nology Development Organization (NEDO) started the Genera-tive AI Accelerator Challenge (GENIAC)102 in February 2024. Thegovernment-funded project subsidizes the computational costsof LLM training for the selected players. The rst term(February–August 2024) supported 10 projects including SakanaAI's, and the second term (October 2024–April 2025) selected 20projects including 3 projects about LLMs for medicine.© 2025 The Author(s). Published by the Royal Society of Chemistryhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jFig. 10 Schematic of MEEP policy based on OODA loop. Observephase: materials doc, orient phase: materials big data, decide phase:human, act phase: autonomous prototyping.Perspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Online6 AI-robot-driven scienceJapan Science and Technology Agency (JST), one of the majornational funding agencies for science, supports multipleprojects related to SDL research. Two JST-Mirai projects,Accelerating Life Sciences by Robotic Biology,103 MaterialsExploration space Extension Platform (MEEP),104 and two JSTMoonshot projects, Co-evolution of Human and AI-Robots toExpand Science Frontiers105 and AI & Robots that Harmonizewith Humans to Create Knowledge and Cross Its Borders106collaboratively founded the AI-Robot-Driven Science Initiative107in 2023 to promote the new scientic methodology realized byAI and robotics. In 2024, “Research innovation throughautonomous-driven research systems” has been designated asone of the strategic objectives authorized by the Ministry ofEducation, Culture, Sports, Science and Technology (MEXT),108and a new funding program “R&D Process Innovation by AI andRobotics”109 has been started. In this section, we introducethese government-supported projects on AI-robot-drivenscience.6.1 Accelerating life sciences by robotic biologyThis project, part of JST-Mirai Program in the “Common Plat-form Technology, Facilities, and Equipment” area, addressescritical issues in life sciences—such as low reproducibility,inefficient use of costly equipment, research misconduct, andthe labor-intensive nature of laboratory work—throughadvanced laboratory automation. While laboratory automationtools are increasingly available, most are limited to specictasks, still relying on human operators to manage samples,reagents, and data interpretation. Consequently, human errorand labor remain constraints on the effectiveness of automa-tion. This project aims to overcome these limitations by devel-oping a comprehensive suite of technologies. These includea standardized experimental protocol description language andIoT-based systems architectures designed for the coordinatedoperation of diverse robotic and automated equipment. Theproject's application areas broadly span the biological sciences,including proteomics, genome editing, and stem cell culture.This interdisciplinary effort involves leading institutions likeRIKEN, AIST, University of Tsukuba, and major industry part-ners including YASKAWA Electric and TECAN Japan. Theproject began with a feasibility study (2018–2020) and moved tofull-scale development in 2021, with completion anticipated inMarch 2025. Funded at approximately 1.1 billion yen (∼7.3million USD), this initiative strives to redene experimentalworkows, minimizing human involvement and enhancingreproducibility and operational efficiency across the lifesciences. Many of the project's outcomes are presented inSection 3.1 (Maholo LabDroid) and 3.2 (Robotic crowd biology).6.2 Materials exploration space extension platformMEEP was launched in the JST-Mirai project in 2021.110–112Researchers' experiment and intuition are important in mate-rials research and development (R&D), and the researchers'inspiration should be more effectively utilized by AI and robot© 2025 The Author(s). Published by the Royal Society of Chemistrysystems. MEEP's proof of concept is 1000 times throughput ofexploration of ion-conductive materials for solid-state batteries.The materials exploration space is overwhelmingly expanded inR&D sites with the following three methods;(1) High-throughput autonomous exploration systems;� “make”: Autonomous prototyping system with vacuumcoating7–9� “measure”: “Materials doc” including autonomous crystalanalysis system113� “save”: materials property prediction system114(2) Data-driven/hypothesis driven hybrid system (Fig. 10);OODA loop with “make”–“measure”–“save”–“understand”induces inspiration85(3) Knowledge sharing;The knowledge obtained from the data is shared among R&Dinstitutes, R&D companies, and measuring instrumentmanufacturers.6.3 Co-evolution of human and AI-robots to expand sciencefrontiersIn the current movement toward scientic automation throughAI and robotics, the focus is largely on conducting reproducible,high-throughput experiments to accelerate scientic discov-eries. However, such automation proves effective primarilywhen hypotheses can be tested at low cost and within a shorttime frame, and when experiments involve repetitive actions onrigid objects. As experimental models become increasinglycomplex, hypothesis testing generally requires greater invest-ment of both time and resources. For instance, in life sciences,experiments with high-delity model organisms—far morecomplex than cell-based tests—pose signicant challenges forautomation. These organisms are typically small, exible, andvariable, limiting the ability of current robotics to performprecise experimental procedures based solely on pre-programmed instructions.Digital Discovery, 2025, 4, 1384–1403 | 1393http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jDigital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article OnlineIn scenarios that demand experimentation in extreme envi-ronments where each task cannot be pre-specied, autonomousrobots capable of self-directed learning and action becomeessential. Such autonomy allows robots to leverage their uniqueability to recongure physical capabilities, presenting a novel,robotics-specic approach to scientic exploration. This projectunder JST Moonshot Goal 3 aims to realize autonomous AIrobotic scientists by 2050. Through transdisciplinary research,it integrates mathematical foundations, scientic AI, robotic AI,and robotic hardware.6.4 AI & robots that harmonize with humans to createknowledge and cross its bordersThis project aims to develop AI robots that can harmonize withhumans to create knowledge and transcend its boundaries by2050. The initiative represents a signicant advancement in theintegration of articial intelligence with scientic research andinnovation processes.The project has established clear milestones, with the rsttarget set for 2025: developing AI robots capable of under-standing, reproducing, and explaining research conducted byhuman scientists, while also generating novel hypotheses. By2030, the project envisions these AI robots collaborating withresearchers across various elds to drive innovation, resultingin the publication of peer-reviewed papers. The ultimate goalfor 2050 is to create an environment where researchers and AIsystems can work together to produce Nobel prize-level researchachievements.The research framework encompasses three interconnectedcomponents. The experiment automation AI robot system isdesigned to conceptualize experiments based on researchhypotheses, estimate specic procedures in cyberspace, andexecute them in physical space. This includes developingautomated synthesis capabilities and understanding experi-mental papers. The Claim and analysis AI focuses on compre-hending multimodal scientic data and providing language-based evidence, utilizing a foundation model that can under-stand relationships between research papers and generatecomprehensive analyses. The Description and dialogue AIsystem aims to summarize experimental results and updatehypotheses through interactive discussions with researchers,incorporating researcher feedback to improve performancewithout requiring large datasets.To achieve these objectives, the project employs variousadvanced technologies including large language models,multimodal AI systems, and automated synthesis devices. Theproject particularly emphasizes the importance of combiningdeductive thinking for continuous performance improvementwith inductive thinking and abduction for paradigm disrup-tion, ultimately aiming to create a new approach to scienticdiscovery that leverages both human expertise and articialintelligence capabilities.6.5 Research process innovation with AI and robotAs a public funding program related to self-driving laboratoriesin Japan, the “R&D Process Innovation by AI and Robotics:1394 | Digital Discovery, 2025, 4, 1384–1403Technical Foundations and Practical Applications” eld waslaunched by the JST in 2024. This funding primarily targetsyoung researchers and conducts three phases of three-and-a-half-year research projects, with about 30 research proposals expectedto be selected in total. The purpose of this program is to revo-lutionize the R&D process through the use of AI and robotics. Byintroducing AI and robotics, the program aims not only to freeresearchers and engineers from simple tasks but also to enablethem to tackle complex challenges beyond conventional cogni-tive and physical capabilities. It is anticipated that by advancingR&D through collaboration between researchers, engineers, andAI and robotics, unprecedented scientic discoveries and tech-nological innovations will be realized, transforming the nature ofR&D. This program seeks proposals from researchers in AI,robotics, and applied elds such as life science and materialsscience, with the goal of creating foundational technologies thatcontribute to innovating the R&D process using AI and robotics.By fostering close collaboration among researchers in theseelds, the program promotes the construction of methodologiesand their practical applications for R&D powered by AI androbotics. By linking foundational technology development withpractical applications in scientic and technological research,the program aims to build a general-purpose framework forautonomously driven R&D, creating new scientic discoveriesand technological innovations.7 EcosystemCollaboration among researchers or industry partners is indis-pensable for developing SDLs that require experts from variouselds. This section reviews the efforts to promote collaborationfor SDLs. We introduce Japanese communities for lab auto-mation users and developers in Section 7.1. An initiative ina national research institute for sharing modules is described inSection 7.2.7.1 Community—LASA, LADEC and Digital LaboratoryConsortiumIn Japan, the Laboratory Automation Suppliers' Association(LASA) was established in May 2019 to accelerate the develop-ment of complex laboratory automation systems by fosteringa regional community of developers.115 Recognizing thatmodern laboratory automation demands expertise acrosshardware, soware, operational management, and applicationdomains, LASA provides a platform where various experts withdifferent backgrounds collaborate closely from the planningstage. Through regular events like the monthly workshop andthe annual Laboratory Automation Developers Conference(LADEC), LASA offers opportunities for members to stay upda-ted on the latest developments, share insights, and engage inface-to-face collaborations. These events feature talks, discus-sions, and activities that cover a broad spectrum of topics, fromhardware customization to soware and AI development, aswell as operational best practices. Several outcomes inuencedby collaborations and discussions within that community havebeen published to date.43,44,46,47,49,116–120© 2025 The Author(s). Published by the Royal Society of Chemistryhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jPerspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article OnlineAs of October 2024, LASA has grown to over 3200 membersfrom academia, industry, government agencies, and media,facilitating cross-disciplinary interactions among researchers,engineers, management, and students. By bridging the gapbetween diverse elds of expertise, LASA plays a critical role inadvancing laboratory automation development in Japan andserves as a model for regional developer communities.As an organization dedicated tomaterials science, the DigitalLaboratory Consortium121 was established in September 2023,bringing together more than 40 companies to exchange infor-mation and develop technologies for automatic and autono-mous experiments.7.2 AUTOkobo at AISTAt the National Institute of Advanced Industrial Science andTechnology (AIST) Japan, efforts are underway to increaseresearch efficiency and streamline experimental processes byimproving access to laboratory automation. This initiative,called “AUTOkobo,” was developed as part of the in-house“Multi-Modal AI” (MMAI) project,122 which aims to improvedigital transformation (DX) literacy among researchers. TheMMAI project enables even beginners to develop advanced AItechnologies through education and shared programmingresources. Recognizing the need for more efficient data acqui-sition, the AUTOkobo was launched in 2023 and has sinceexpanded to all seven research departments at AIST.The AUTOkobo focuses on automating traditionally batch-based experimental processes, particularly in areas such as,polymers, inorganic materials, thin lms, and ceramics. CentralFig. 11 Conceptual image of AUTOkobo infrastructure acceleratingSDL development in AIST. Automation modules, such as liquiddispensing or powder dispensing, are developed and shared allowingeach SDL to integrate them into their systems to perform complextasks.© 2025 The Author(s). Published by the Royal Society of Chemistryto its approach is the modularization of laboratory processes(Fig. 11, below). Automation modules, such as robots and liquidor powder dispensing devices, are provided free of charge,allowing researchers to integrate them into their workows toperform complex tasks. In Self-Driving Lab (SDL) systems,robotics and peripherals manage the ow of materials betweeninstruments, while the AUTOkobo team supports system inte-gration to reduce the burden on the researchers. In addition, theAUTOkobo is developing a workspace that serves as botha showroom and a workshop for automation technologies,allowing researchers to engage in hands-on examination of theavailable modules and instruments. This modular approachaddresses key challenges in laboratory automation, includinglow costs, system exibility, and integration of new systems intoconventional experimental setups. By providing exible auto-mation modules, researchers can automate experiments withoutnancial risk, and once completed, modules can be reused byothers. In this way, the AUTOkobo approach not only saves timeand resources but also promotes the widespread adoption ofautomation across diverse research domains, including poly-mers, inorganic materials, thin lms, and ceramics.8 Industry support for SDLdevelopmentJapan has active manufacturing industries, and universities andresearch institutes oen collaborate closely with companies todevelop SDLs. This section outlines the contributions of Japa-nese industries to SDL development and highlights partner-ships between academia and industry. Section 8.1 introducesrobotic arms developed in Japan, Section 8.2 highlights custom-made automation systems created through collaborations, andSection 8.3 showcases soware development efforts.8.1 Robotic armsRobotic arms play a central role in some SDLs for their dexterityin object handling. Japan has competitive robot manufacturers,including FANUC and YASKAWA, which are two of the “Big 4”,the four largest industrial robot manufacturers in the world. Theyprovide different kinds of robot arms for SDLs. For example,Maholo LabDroid (Section 3.1) was developed by YASKAWA andthe Robotic Biology Institute. A collaborative robot COBOTTA byDENSO WAVE is adopted in the autonomous X-ray diffractionanalysis system33 (Section 2.7) and a robotic pipetting system forplant pots.123 Industrial robotic arms from DENSO WAVE areintegrated into a multiarm robotic platform for scientic explo-ration.124 An industrial robot MELFA from MITSUBISHI ELEC-TRIC has been incorporated into an automatic gamma-rayactivation analysis system at Japan Atomic Energy Agency.125 Adual-arm robot NEXTAGE from Kawada Robotics has beenutilized in the cell culture system in a pharmaceutical companyEisai126 and powder dispensing system by ExaWizards.1278.2 Custom-made lab automation systemSDLs in Japan are oen developed through close collaborationbetween academia and industry. For example, ShimadzuDigital Discovery, 2025, 4, 1384–1403 | 1395http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jFig. 12 Schematic of the digital laboratory (dLab) for thin filmmaterials.131Digital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article OnlineCorporation and Kobe University jointly developed an autono-mous laboratory system for biotechnology research anddemonstrated the optimization of medium conditions forbacteria to improve glutamic acid production.128 KiQ Roboticsdeveloped Lab Auto, an integrated system that automates RNAevolution experiments at The University of Tokyo.129 Here, wepresent three case studies from SDLs developed by the authorsto highlight the real-world challenges and their solutions.The rst case study is the Lab Full Automation system fromForDx.130 The collaboration between ForDX and FurukawaElectric Advanced Engineering Co., Ltd brought expertise indeveloping automation technology that integrates dispensingand measurement devices. However, they lacked the capabilityto create a self-driving laboratory by incorporating AI algo-rithms. To address this, NIMO (Section 5.2) was employed toestablish a closed-loop system between experiments and AI,enabling the development and commercialization of self-driving laboratory devices. In Japan, the development of hard-ware and soware for machine learning has traditionally beenconducted independently. We recognize that fostering closeintegration between these two elements is a key challenge forthe widespread adoption of self-driving laboratories in Japan.Another case study is from HORIBA, Ltd that is collaboratingwith ROPES (Section 2.8). The FC(fuel cell)-ROPES, shown inFig. 7(d) and (e), was developed by a nationally-funded projectby the New Energy and Industrial Technology DevelopmentOrganization (NEDO) with the initial users from R&D sites atJapanese fuel cell stack OEMs, including Toyota, Honda,Panasonic, and Toshiba. These companies required a higher-throughput system for exploring process parameters thatcould be applied in real factories. While they were interested inautomation, traditional R&D methods based on manual laborwere estimated to be more cost-effective compared to automa-tion systems developed independently by individual companies.To overcome this barrier and promote system adoption, it wasnecessary to reduce the unit price and increase the operatingefficiency. To achieve this, the FC-ROPES was designed withthree design-philosophies: (i) scalability, including a wide rangeof process parameters and customizable evaluation units (e.g.,customer-selectable objective functions) (ii) desktop size forquick delivery and fast returns and adaptability to projectchanges, and (iii) the ability to function as a pilot line for realfactories. As demand for fuel cells grows and prices decrease,the system can be expanded horizontally to other applicationsusing powder-lm-formation processes, such as batteries orceramic lms. Furthermore, when the number of users of theproduction process R&D increases, commoditization is likely tospread among academic researchers as well.Finally, Nishio, Hitosugi et al. have recently constructeda digital laboratory, called dLab, which interconnects instru-ments using robots to collect experimental data (includingsynthesis processes, measured physical properties, andmeasurement conditions) for solid materials research in thin-lm form.131 Several modular experimental instruments areinterconnected (Fig. 12), allowing automated materialsynthesis, measurement, and analysis. Data from the instru-ments are output in the MaiML format and stored in a cloud-1396 | Digital Discovery, 2025, 4, 1384–1403based database. JEOL Ltd has developed an automated scan-ning electron microscope for thin lm samples that can beconnected to the dLab system (Fig. 12). Rigaku Corp. hasdeveloped a thin-lm X-ray diffractometer that works witha robot for experiments, both of which are commercially avail-able. Additionally, Shimadzu Corp. and HORIBA, Ltd provideoptical properties measurement systems. All these instrumentsfollow established standards for physical connections andcommunication protocols, which are publicly available.132 ThedLab system autonomously synthesized high-quality LiCoO2(001) thin lms, optimizing the X-ray diffraction peak-intensityratio using Bayesian methods. The diffraction pattern les inMaiML format on the cloud were automatically analyzed, andBayesian optimization autonomously proposed the next thin-lm deposition condition to obtain better-quality thin lms. Itshowcases advanced autonomous material synthesis driven bydata and robotics for materials science.In addition, companies oen collaborate to develop labora-tory automation systems. A robotic system for mouse tail veininjection developed by Preferred Networks and Chugai Pharma-ceutical133 is a notable example, which is now commercialized asAUTiv.134 TORCH, Inc. provides laboratory automation solutionsto companies, such as automated pouring and closing of samplecontainers with collaborative robots at Lion Corporation.1358.3 Soware developmentTo realize SDLs, it is important to develop programs to controleach device. These programs are implemented in Python, Lab-VIEW, and various other languages. In Japan, there arecompanies that can help with the development of the controlprograms of devices. For example, CJS Inc. has created a Lab-VIEW program to control a syringe pump, contributing todeveloping an automated autonomous odor blending system.136Furukawa Electric Advanced Engineering has developeda program for automated dispensing equipment, which hasbeen implemented in the robotic experimental setup forsearching electrochemical materials discovery (see Section 2.3).9 For the future of SDLsWe have reviewed the current state of SDLs in Japan. Whilesignicant research has been conducted, their adoption© 2025 The Author(s). Published by the Royal Society of Chemistryhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jPerspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Onlineremains limited due to various challenges. Further democrati-zation is necessary for wider adaptation of SDL technologies.56This section identies the key barriers and explores potentialstrategies to alleviate them, facilitating broader implementa-tion of SDLs.Adopting SDLs may face several challenges, one of which isa reluctance to change. Many researchers oen prefer to rely ontraditional methods and hesitate to adopt next-generationstrategies, perhaps due to familiarity with traditionalmethods, technological barriers, lack of trust in the newmethodologies, etc. This issue can be mitigated through theeducation of young researchers, particularly during the earlystages of their training. Educational efforts in Japan aimed atnurturing future automation developers are introduced inSection 9.1. Another challenge lies in the high cost of hardwareand soware development. Many current SDLs are monolithic,meaning they are entirely custom-built with non-replaceablecomponents. Standardization plays an important role inreducing costs and enhancing exibility (modularity andexpandability). Standardized interfaces enable modularsystems, allowing users to customize setups, select affordablehardware, and optimize their systems based on their require-ments. Standardization also allows soware reuse. This topic isfurther explored in Section 9.2. Finally, benchmarks are vital inhelping users select appropriate methods and guiding devel-opers in taking their rst steps into SDL development. Severalbenchmarks for SDL research are proposed in Section 9.3 tosupport this objective.9.1 EducationTo enhance the development of SDLs in the future, educationprograms that familiarize students with the concept of SDLs arenecessary. Several lecture courses have already been provided inJapanese universities for this purpose.At Keio University, a practical training course on AI-roboticscience has been offered to undergraduates. As part of theJST-Mirai Program described in Section 6.1, from 2021 to 2024,a 5–6 days intensive course titled “Automation of ScienticExperiments” was conducted using actual robots.137 The coursewas designed to allow students to experience how AI connectedto experimental robots can discover new knowledge throughautomated experiments. Specically, students learned toprogram experimental protocols for liquid-handling experi-ments, which are fundamental techniques for PCR tests andchemical experiments—using Python to fully control the OT-2liquid-handling robot (Opentrons Labworks Inc.). They alsobuilt an automated experimental planning AI that interpretsresults, plans subsequent experiments, and instructs the robot,creating an AI-robot system that autonomously performsscientic experiments. By rst manually performing the exper-imental processes and then implementing them into the robotand AI, students gained a deeper understanding of how thecollaboration between humans, automation, and AI canenhance problem-solving capabilities, foster creativity, andimprove learning outcomes in addition to their respective roles.Participants ranged from students engaged in life science© 2025 The Author(s). Published by the Royal Society of Chemistryresearch to those using a micropipette for the rst time, but allwere new to programming robots for experiments. Throughcollaborative learning and tackling assignments, they gainedvarious stimulating experiences. Plans are underway to transferthe course to other universities.The Department of Chemistry at The University of Tokyooffers Information Chemistry as a regular lecture course,teaching the basics of materials informatics and SDL to third-year undergraduates and above.138 The lecture also includeshands-on experience with machine learning and teaching robotsto perform specic motions. This is a unique lecture in Japan asthere are still only a few institutions that can teach robotics inchemistry.139 At the Institute of Tokyo Science (Science Tokyo),materials informatics can be studied systematically in an orga-nization called TAC-MI; basic education on SDL has also beeninitiated.140 These educational efforts can provide young studentsand potential future researchers with a foundational under-standing of new technologies as well as offer a hands-on expe-rience that helps reduce the mental barriers to their adoption.141In addition to lectures at universities, textbooks written inlocal languages help disseminate knowledge about SDLs toa broader audience. A book entitled ‘Intersection of Materials,Machine Learning, and Robotics’142 has been published tosupport the development of the community. LASA (see Section7.1) is also planning to create a textbook on laboratory auto-mation. These resources will be valuable to the eld byproviding researchers with foundational knowledge andmaking SDL development more accessible.9.2 StandardizationSDLs typically connect devices such as robot arms andmeasurement instruments. As automation system require-ments change, hardware and soware may need replacement.Standardization is crucial for creating exible systems thatallow hardware to be replaced while keeping soware changesminimal. Key areas of standardization include hardwaredimensions, physical connection interfaces, and communica-tion protocols. Modularity achieved through standardizationcan lower development costs and encourage broader adoptionof SDL technologies.Several initiatives have been undertaken by various groups toaddress this need for standardization. For instance, the IVIFoundation provides open industry standard soware archi-tectures, including the Virtual Instrument Soware Architecture(VISA),143 which aims to improve the interchangeability of testand measurement instruments that communicate througha variety of I/O buses. Similarly, the SiLA Consortium144develops open standards to support the integration of labora-tory automation systems. Recently, the Laboratory and Analyt-ical Device Standard (LADS)145 has been introduced asa manufacturer-independent, open standard for analytical andlaboratory equipment. It is built upon Open PlatformCommunications Unied Architecture (OPC UA) and aims toimprove the plug and play interoperability of analytical devices.These efforts involve manufacturers and impact the direction ofhardware development.Digital Discovery, 2025, 4, 1384–1403 | 1397http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jDigital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article OnlineJapan Analytical Instruments Manufacturers Association(JAIMA) advances the standardization of laboratory andanalytical instruments in Japan. JAIMA contributed to thedevelopment of LADS OPC UA and the standardization of theMaiML format (Section 2.2) as Japanese Industrial Standards(JIS K 0200).Although standards play a crucial role in enhancing inter-operability, the real challenge lies in ensuring their widespreadadoption and compliance. Close collaboration between stake-holders, including academia and industry, is essential todeveloping practical and effective standards. Additionally, theexistence of multiple competing standards may undermine theintended benets of standardization. To facilitate globaladoption, international collaboration is also critical topromoting the broad implementation of these standards.9.3 Benchmark for SDLsBenchmarks have been helpful for various research areas byfacilitating the comparison of different methods, and the needfor standard benchmarks for SDLs was highlighted during theauthors' meeting. Here we propose potential benchmarkingtasks to evaluate the practical utility of lab automation systemsto enhance the international discussion on establishing SDLbenchmarks.9.3.1 Powder dispensing. In comparison with liquiddispensers, powder dispensing is still tricky because of thecomplex mechanics of small particles. The use of robotic armshas been investigated recently88,146 to overcome the limitationsof commercial devices such as Chemspeed or Quantos (MettlerToledo). Due to the widespread nature of the task and thediversity of approaches available, powder dispensing is suitablefor a benchmarking task. Benchmark criteria can include speed,accuracy, and hardware cost, allowing researchers to developrobotic systems optimized for specic performance aspects.9.3.2 Viscous liquid handling. Working with viscous liquidis required in various experiments. For instance, this may be thecase when using compounds with melting points near roomtemperature, such as di-tert-butyl dicarbonate or tri-tert-butyl-phosphine, as reagents, or when using highly viscous polymersor oligomers as reaction substrates. Furthermore, in industrialproduction, reducing the amount of solvent oen enhancesproductivity, which frequently necessitates the consideration ofhigh-viscosity, highly concentrated solutions. However,currently available liquid handlers sometimes have difficultiesin treating viscous liquids, and optimization of the liquidhandling parameters is necessary for desirable performance.147Benchmarks on viscous liquid handling would help developerschoose an appropriate device for their use case. Polyethyleneglycol (PEG) can be used as a viscosity standard because of thediversity in viscosity.9.3.3 Robot performance in realistic tasks. Precise objecthandling is oen required to complete regular tasks in SDLs,such as microplate placing.148 Although the manufacturersusually provide the repeatability of their robotic arm, thismetric does not always reect their precision in practicalapplications since various factors, such as payload, can affect1398 | Digital Discovery, 2025, 4, 1384–1403the performance of the robot. Harazono, et al.116 developeda platform to evaluate the microplate handling accuracy ofrobot arms. Practical benchmarks like this are useful forassessing the real-world performance of robotic systems.10 ConclusionsThis perspective has highlighted the advancements in SDLresearch in Japan, spanning material sciences, biology, andorganic chemistry. Furthermore, we have explored the roles ofcommunity collaboration, funding initiatives, and the industrythat supports the growth of SDLs.Compared to SDLs developed in other countries, a distinc-tive characteristic of Japan's SDL development might be thestrong collaboration between academia and industry. As dis-cussed in Section 8, universities and national research insti-tutes in Japan frequently partner with industrial collaborators,including small companies, to develop custom automationhardware tailored to specic research needs. In addition toJapan's strong automation industry, active measurement andanalysis equipment manufacturers, holding an 8% of the globalmarket share as of 2021 and ranking third aer the UnitedStates and Germany,149 are supporting the development ofJapanese SDLs. The integral nature of SDLs also aligns with thestrengths of Japanese manufacturers, as Japan tends to havea comparative advantage in products with a more integralarchitecture.150 However, it has been pointed out that Japaneserobotics research is lagging, reecting the relatively weakperformance in the eld of AI.151 Advancements in AI may alsobe crucial for the further development of SDLs in Japan.In addition to research institutes, companies also showstrong interest in SDLs. Japan has a large concentration ofmaterials, automation, and scientic equipment industries.These companies are making efforts to introduce SDLs toenhance the creativity of researchers. Although the number ofSDLs is still small, their use is steadily spreading. By buildingsynergy among diverse players, SDLs in Japan will continue todevelop and drive innovation in the eld.Data availabilityAs this is a Perspective article, no primary research results, data,soware or code have been included.Author contributionsConceptualization (organizers)—D. N. F., T. H., K. Nis., W. S., S.T., K. Tsu., N. Y.; writing (original dra)—Section 1: N. Y.;Section 2.1, 2.2: T. H., K. Nis.; Section 2.3: S. M.; Section2.4: M. N.; Section 2.5, 2.6: Y. A., J. S., K. S.; Section 2.7: K. O.;Section 2.8: K. Nag.; Section 3: G. N. K., T. N., H. O., K. Tak.;Section 4: Y. N.; Section 5.1: S. T.; Section 5.2: R. T.; Section 5.3:Y. U.; Section 5.4: N. Y.; Section 6.1: K. Tak.; Section 6.2: K. Nag.;Section 6.3: K. H.; Section 6.4: Y. U.; Section 6.5: I. T.; Section7.1: T. H., G. N. K.; Section 7.2: D. N. F., W. S.; Section 8: T. H., K.Nag., K. Nis., R. T., N. Y.; Section 9.1: T. H., G. N. K., H. O.;© 2025 The Author(s). Published by the Royal Society of Chemistryhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jPerspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article OnlineSection 9.2, 9.3, 10: N. Y.; Writing (review & editing)—D. N. F., N.Y.; project administration—N. Y.Conflicts of interestTohru Natsume is an executive at Robotic Biology Institute Inc.,which may benet nancially from the increased scientic useof Maholo LabDroid.AcknowledgementsY. A., K. S., and J. S. acknowledge the support of JSPS KAKENHI(23K03771), JST CREST (JPMJCR21O2), and Daikin Industries,Ltd. K. H. acknowledges the support of JST Moonshot R&DJPMJMS2033. T. H. and K. Nishio acknowledge the support ofthe JST-Mirai Program JPMJMI21G2, JST CREST (JPMJCR22O4),JSPS KAKENHI (24K01599), NEDO Project (P20003), and MEXTProgram: Data Creation and Utilization-Type Material Researchand Development Project Grant Number JPMXP1122712807.G. N. K. acknowledges the support of Grant-in-Aid for ScienticResearch (C) (JP23K11820). G. N. K. and N. Y. are supported byMedical Research Center Initiative for High Depth Omics,Nanken-Kyoten, and Multilayered Stress Diseases(JPMXP1323015483), Science Tokyo. Y. N. acknowledges thesupport of JST-ERATO (JPMJER1903), JSPS KAKENHI GrantNumbers JP21H01924 and JP23H03810, and the Institute forChemical Reaction Design and Discovery (ICReDD), which wasestablished by the World Premier International ResearchInitiative (WPI), MEXT, Japan. K. Nagato acknowledges thesupport of JST-Mirai Program JPMJMI21G2, JST-Mirai ProgramJPMJMI19G3, and NEDO Project 23200629-0. M. N. acknowl-edges the support of JST CREST (JPMJCR19J3), MEXT Program:Data Creation and Utilization-Type Material Research andDevelopment Project Grant Number JPMXP1122714694, andCOI-NEXT (JPMJPF2102). H. O. and K. Takahashi acknowledgethe support of JST-Mirai Program (JPMJMI18G4 andJPMJMI20G7). H. O. also acknowledges the support of Grant-in-Aid for Early-Career Scientists (JP22K17992) and Grant-in-Aidfor Transformative Research Areas (A) (JP23H04149). K. O.acknowledges the support of JST-Mirai Program JPMJMI19G1,MEXT Program: Data Creation and Utilization-Type MaterialResearch and Development Project (Digital TransformationInitiative Center for Magnetic Materials) (Grant NumberJPMXP1122715503), and MEXT Program: Developinga Research Data Ecosystem for the Promotion of Data-DrivenScience. I. T. acknowledges the support of MEXT KAKENHI(20H00601), JST CREST (JPMJCR21D3), JST Moonshot R&D(JPMJMS2033-05), and RIKEN Center for Advanced IntelligenceProject. R. T. acknowledges the support of JST-PRESTO(JPMJPR24T8). Y. U. and K. O. acknowledge the support of JSTMoonshot R&D JPMJMS2236. K. Takahashi and Y. U. are sup-ported by Advanced General Intelligence for Science Program(AGIS), the RIKEN TRIP initiative. We thank Yasuhiro Naito ofKeio University for assistance with the writing of Section 9.1. Weare grateful to Yumiko Miyahara for preparing the graphicalabstract and Fig. 1. We also thank Yusaku Nakajima (The© 2025 The Author(s). Published by the Royal Society of ChemistryUniversity of Osaka) for providing Fig. 6. We appreciate ToruIshikuma (JAIMA) for reviewing the content of Section 9.2.References1 G. Tom, S. P. Schmid, S. G. Baird, Y. Cao, K. Darvish,H. Hao, S. Lo, S. Pablo-Garćıa, E. M. Rajaonson,M. Skreta, N. Yoshikawa, S. Corapi, G. D. Akkoc,F. Strieth-Kalthoff, M. Seifrid and A. Aspuru-Guzik, Chem.Rev., 2024, 124, 9633–9732.2 World Robotics 2023 Report: Asia ahead of Europe and theAmericas, https://ifr.org/ifr-press-releases/news/world-robotics-2023-report-asia-ahead-of-europe-and-the-americas, accessed May 2025.3 R. Matsuda, M. Ishibashi and Y. Takeda, Chem. Pharm.Bull., 1988, 36, 3512–3518.4 M. Sasaki, T. Kageoka, K. Ogura, H. Kataoka, T. Ueta andS. Sugihara, Clin. Chim. Acta, 1998, 278, 217–227.5 J. Boyd, Science, 2002, 295, 517–518.6 About DxMT, https://dxmt.mext.go.jp/en/about, (accessedApril 2025).7 R. Shimizu, S. Kobayashi, Y. Watanabe, Y. Ando andT. Hitosugi, APL Mater., 2020, 8, 111110.8 S. Kobayashi, R. Shimizu, Y. Ando and T. Hitosugi, ACSMater. Lett., 2023, 5, 2711–2717.9 R. Nakayama, R. Shimizu, T. Haga, T. Kimura, Y. Ando,S. Kobayashi, N. Yasuo, M. Sekijima and T. Hitosugi, Sci.Technol. Adv. Mater.:Methods, 2022, 2, 119–128.10 H. Xu, R. Nakayama, T. Kimura, R. Shimizu, Y. Ando,S. Kobayashi, N. Yasuo, M. Sekijima and T. Hitosugi, Sci.Technol. Adv. Mater.:Methods, 2023, 3, 2210251.11 K. Nishio et al., Talk at the Accelerate Conference, 2024(https://www.accelerate24.ca/program-3).12 N. Ishizuki, R. Shimizu and T. Hitosugi, Sci. Technol. Adv.Mater.:Methods, 2023, 3, 2197519.13 S. H. M. Mehr, M. Craven, A. I. Leonov, G. Keenan andL. Cronin, Science, 2020, 370, 101–108.14 S. Ichimura, T. Sigehuzi, T. Yasunaga and S. Inoue, OyoButuri, 2023, 92, 142–146.15 M. Aykol, P. Herring and A. Anapolsky, Nat. Rev. Mater.,2020, 5, 725–727.16 K. Xu, Chem. Rev., 2014, 114, 11503–11618.17 S. Matsuda, K. Nishioka and S. Nakanishi, Sci. Rep., 2019, 9,6211.18 S. Matsuda, G. Lambard and K. Sodeyama, Cell Rep. Phys.Sci., 2022, 3, 100832.19 R. Tamura, K. Tsuda and S. Matsuda, Sci. Technol. Adv.Mater.:Methods, 2023, 3, 2232297.20 S. Matsuda, S. Kimura and M. Takahashi, BatteriesSupercaps, 2024, 7, e202400509.21 S. Schmatloch, M. A. R. Meier and U. S. Schubert,Macromol.Rapid Commun., 2003, 24, 33–46.22 S. Oliver, L. Zhao, A. J. Gormley, R. Chapman and C. Boyer,Macromolecules, 2019, 52, 3–23.23 G. K. K. Clothier, T. R. Guimarães, S. W. Thompson,S. C. Howard, B. W. Muir, G. Moad and P. B. Zetterlund,Angew. Chem., Int. Ed., 2024, 63, e202320154.Digital Discovery, 2025, 4, 1384–1403 | 1399https://ifr.org/ifr-press-releases/news/world-robotics-2023-report-asia-ahead-of-europe-and-the-americashttps://ifr.org/ifr-press-releases/news/world-robotics-2023-report-asia-ahead-of-europe-and-the-americashttps://ifr.org/ifr-press-releases/news/world-robotics-2023-report-asia-ahead-of-europe-and-the-americashttps://dxmt.mext.go.jp/en/abouthttps://www.accelerate24.ca/program-3http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jDigital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Online24 M. Seifrid, F. Strieth-Kalthoff, M. Haddadnia, T. C. Wu,E. Alca, L. Bodo, S. Arellano-Rubach, N. Yoshikawa,M. Skreta, R. Keunen and A. Aspuru-Guzik, DigitalDiscovery, 2024, 3, 1319–1326.25 C. Kang, J. J. M. Machado, Y. Sekiguchi, M. Ji, C. Sato andM. Naito, J. Adhes., 2023, 99, 2080–2096.26 S. Takasuka, S. Ito, S. Oikawa, Y. Harashima, T. Takayama,A. Nag, A. Wakiuchi, T. Ando, T. Sugawara, M. Hatanaka,T. Miyao, T. Matsubara, Y.-Y. Ohnishi, H. Ajiro andM. Fujii, Sci. Technol. Adv. Mater.:Methods, 2024, 4, 2425178.27 C. Kang, M. Ji, Y. Sekiguchi, M. Naito and C. Sato, J. Adhes.,2023, 101, 18–40.28 Y. Asano, K. Okada, S. Nakagawa, N. Yoshie and J. Shiomi,Robot. Auton. Syst., 2025, 185, 104868.29 Y. Asano, K. Okada and J. Shiomi, IEEE/RSJ InternationalConference on Intelligent Robots and Systems (IROS), 2024,DOI: 10.1109/IROS58592.2024.10802186.30 A. Sakurai, K. Yada, T. Simomura, S. Ju, M. Kashiwagi,H. Okada, T. Nagao, K. Tsuda and J. Shiomi, ACS Cent.Sci., 2019, 5, 319–326.31 J. Guo and J. Shiomi, Next Energy, 2024, 2, 100078.32 Y. Nakajima, K. Kawasaki, Y. Takeichi, M. Hamaya,Y. Ushiku and K. Ono, Digital Discovery, 2024, 3, 2130–2136.33 Y. Yotsumoto, Y. Nakajima, R. Takamoto, Y. Takeichi andK. Ono, Digital Discovery, 2024, 3, 2523–2532.34 K. Nagai, T. Osa, G. Inoue, T. Tsujiguchi, T. Araki,Y. Kuroda, M. Tomizawa and K. Nagato, Sci. Rep., 2022,12, 1615.35 Experience the ”feeling of powder”!! ROPES, https://www.youtube.com/watch?v=QySre6yKSSM, (accessed April2025).36 Autonomous Experiment with AI&Robot, https://youtu.be/_JM4BZHsSKQ, (accessed April 2025).37 H. Zhu, H. Zhang, Y. Xu, S. Laššáková, M. Korabečná andP. Neužil, BioTechniques, 2020, 69, 317–325.38 G. N. Kanda, T. Tsuzuki, M. Terada, N. Sakai, N. Motozawa,T. Masuda, M. Nishida, C. T. Watanabe, T. Higashi,S. A. Horiguchi, T. Kudo, M. Kamei, G. A. Sunagawa,K. Matsukuma, T. Sakurada, Y. Ozawa, M. Takahashi,K. Takahashi and T. Natsume, eLife, 2022, 11, e77007.39 W. A. Kamel, E. Sugihara, H. Nobusue, S. Yamaguchi-Iwai,N. Onishi, K. Maki, Y. Fukuchi, K. Matsuo, A. Muto, H. Sayaand T. Shimizu, Mol. Cancer Ther., 2017, 16, 182–192.40 M. Sasamata, D. Shimojo, H. Fuse, Y. Nishi, H. Sakurai,T. Nakahata, Y. Yamagishi and H. Sasaki-Iwaoka, SLASTechnol., 2021, 26, 441–453.41 A. W. Liu, A. Villar-Briones, N. M. Luscombe and C. Plessy,F1000Research, 2022, 11, DOI: 10.12688/f1000research.109251.1.42 S. Kaneko, T. Mitsuyama, K. Shiraishi, N. Ikawa, K. Shozu,A. Dozen, H. Machino, K. Asada, M. Komatsu, A. Kukita,K. Sone, H. Yoshida, N. Motoi, S. Hayami, Y. Yoneoka,T. Kato, T. Kohno, T. Natsume, G. v. Keudell, V. Saloura,H. Yamaue and R. Hamamoto, Cancers, 2021, 13, 2126.43 K. Ochiai, N. Motozawa, M. Terada, T. Horinouchi,T. Masuda, T. Kudo, M. Kamei, A. Tsujikawa,1400 | Digital Discovery, 2025, 4, 1384–1403K. Matsukuma, T. Natsume, G. N. Kanda, M. Takahashiand K. Takahashi, SLAS Technol., 2021, 26, 209–217.44 M. Terada, Y. Kogawa, Y. Shibata, M. Kitagawa, S. Kato,T. Iida, T. Yorimitsu, A. Kato, K. Matsukuma, T. Maeda,M. Takahashi and G. N. Kanda, SLAS Technol., 2023, 28,449–459.45 N. Yachie, R. B. Consortium and T. Natsume, Nat.Biotechnol., 2017, 35, 310–312.46 T. D. Itoh, T. Horinouchi, H. Uchida, K. Takahashi andH. Ozaki, SLAS Technol., 2021, 26, 650–659.47 Y. Arai, K. Takahashi, T. Horinouchi, K. Takahashi andH. Ozaki, SLAS Technol., 2023, 28, 264–277.48 K. Tachibana and H. Ozaki, OT2Eye (1.0.1), 2024, DOI:10.5281/zenodo.13923719.49 T. Inagaki, A. Kato, K. Takahashi, H. Ozaki and G. N. Kanda,arXiv, 2023, preprint, arXiv:2304.10267, DOI: 10.48550/arXiv.2304.10267.50 C. Armer, F. Letronne and E. DeBenedictis, PLoS Biol., 2023,21, e3001919.51 Y. Nagata, Y.-Z. Ke and M. Suginome, Chem. Lett., 2015, 44,53–55.52 Y. Nagata, K. Takagi and M. Suginome, J. Am. Chem. Soc.,2014, 136, 9858–9861.53 R. Shirasawa, I. Takemura, S. Hattori and Y. Nagata,Commun. Chem., 2022, 5, 158.54 N. Tsuji, P. Sidorov, C. Zhu, Y. Nagata, T. Gimadiev,A. Varnek and B. List, Angew. Chem., Int. Ed., 2023, 62,e202218659.55 S. Akiyama, H. Akitsu, R. Tamura, W. Matsuoka, S. Maeda,K. Tsuda and Y. Nagata, ChemRxiv, 2024, preprint, DOI:10.26434/chemrxiv-2024-bnj6p-v2.56 S. Lo, S. G. Baird, J. Schrier, B. Blaiszik, N. Carson, I. Foster,A. Aguilar-Granda, S. V. Kalinin, B. Maruyama, M. Politi,H. Tran, T. D. Sparks and A. Aspuru-Guzik, DigitalDiscovery, 2024, 3, 842–868.57 M. Kuwahara, Y. Hasukawa, F. Garcia-Escobar, S. Maeda,L. Takahashi and K. Takahashi, ACS Appl. Eng. Mater.,2025, 3, 978–987.58 Y. Wang, J. Wang, Z. Cao and A. Barati Farimani, Nat. Mach.Intell., 2022, 4, 279–287.59 S. Chithrananda, G. Grand and B. Ramsundar, arXiv, 2020,preprint, arXiv:2010.09885, DOI: 10.48550/arXiv.2010.09885.60 V. Y. Shirasuna, E. Soares, E. V. Brazil, K. F. A. Gutierrez,R. Cerqueira, S. Takeda and A. Kishimoto, ICML 2024 AIfor Science Workshop, 2024, https://openreview.net/forum?id=PkDHmg2JLI.61 S. Takeda, A. Kishimoto, L. Hamada, D. Nakano andJ. R. Smith, Proc. AAAI Conf. Artif. Intell., 2024, 37, 15376–15383.62 Example 6 – Human-in-the-loop design of more solublemolecules with IBM's embedding model (smi-TED),https://mehradans92.github.io/dZiner/solubility_inference_IBM_representation_learning.html,(accessed April 2025).© 2025 The Author(s). Published by the Royal Society of Chemistryhttps://doi.org/10.1109/IROS58592.2024.10802186https://www.youtube.com/watch?v=QySre6yKSSMhttps://www.youtube.com/watch?v=QySre6yKSSMhttps://youtu.be/_JM4BZHsSKQhttps://youtu.be/_JM4BZHsSKQhttps://doi.org/10.12688/f1000research.109251.1https://doi.org/10.12688/f1000research.109251.1https://doi.org/10.5281/zenodo.13923719https://doi.org/10.48550/arXiv.2304.10267https://doi.org/10.48550/arXiv.2304.10267https://doi.org/10.26434/chemrxiv-2024-bnj6p-v2https://doi.org/10.48550/arXiv.2010.09885https://doi.org/10.48550/arXiv.2010.09885https://openreview.net/forum?id=PkDHmg2JLIhttps://openreview.net/forum?id=PkDHmg2JLIhttps://mehradans92.github.io/dZiner/solubility_inference_IBM_representation_learning.htmlhttps://mehradans92.github.io/dZiner/solubility_inference_IBM_representation_learning.htmlhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jPerspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Online63 RoboRXN: automating chemical synthesis, https://research.ibm.com/blog/roborxn-automating-chemical-synthesis, (accessed April 2025).64 Foundation Model for Materials – FM4M, https://github.com/IBM/materials, accessed April 2025.65 Materials and Chemistry Working Group j AI Alliance,https://thealliance.ai/working-groups/materials-and-chemistry, (accessed April 2025).66 T. Ueno, T. D. Rhone, Z. Hou, T. Mizoguchi and K. Tsuda,Materials Discovery, 2016, 4, 18–21.67 Y. Motoyama, R. Tamura, K. Yoshimi, K. Terayama, T. Uenoand K. Tsuda, Comput. Phys. Commun., 2022, 278, 108405.68 K. Terayama, R. Tamura, Y. Nose, H. Hiramatsu,H. Hosono, Y. Okuno and K. Tsuda, Phys. Rev. Mater.,2019, 3, 033802.69 R. Tamura, H. Morito, G. Deffrennes, M. Naito, Y. Nose,T. Abe and K. Terayama, Commun. Mater., 2024, 5, 139.70 AIPHAD Investigate Phase Diagrams using AI, https://aiphad.org/, (accessed April 2025).71 K. Terayama, M. Sumita, R. Tamura, D. T. Payne,M. K. Chahal, S. Ishihara and K. Tsuda, Chem. Sci., 2020,11, 5959–5968.72 J. Li, J. Zhang, R. Tamura and K. Tsuda, Digital Discovery,2022, 1, 295–302.73 K. Kitai, J. Guo, S. Ju, S. Tanaka, K. Tsuda, J. Shiomi andR. Tamura, Phys. Rev. Res., 2020, 2, 013319.74 nimo, https://github.com/NIMS-DA/nimo, (accessed April2025).75 L. M. Roch, F. Häse, C. Kreisbeck, T. Tamayo-Mendoza,L. P. E. Yunker, J. E. Hein and A. Aspuru-Guzik, PLoS One,2020, 15, e0229862.76 J. Li, Y. Tu, R. Liu, Y. Lu and X. Zhu, Advanced Science, 2020,7, 1901957.77 M. Sim, M. G. Vakili, F. Strieth-Kalthoff, H. Hao,R. J. Hickman, S. Miret, S. Pablo-Garćıa and A. Aspuru-Guzik, Matter, 2024, 7, 2959–2977.78 R. Katsube, K. Terayama, R. Tamura and Y. Nose, ACSMater. Lett., 2020, 2, 571–575.79 K. Terayama, Y. Osaki, T. Fujita, R. Tamura, M. Naito,K. Tsuda, T. Matsui and M. Sumita, J. Chem. TheoryComput., 2023, 19, 6770–6781.80 Y. Murakami, S. Ishida, Y. Demizu and K. Terayama, DigitalDiscovery, 2023, 2, 1347–1353.81 T. Liu, N. Astorga, N. Seedat and M. van der Schaar, TheTwelh International Conference on LearningRepresentations, 2024, https://openreview.net/forum?id=OOxotBmGol.82 C. Lu, C. Lu, R. T. Lange, J. Foerster, J. Clune and D. Ha,arXiv, 2024, preprint, arXiv:2408.06292, DOI: 10.48550/arXiv.2408.06292.83 Y. Matsubara, N. Chiba, R. Igarashi and Y. Ushiku, J. Data-centric Mach. Learn. Res., 2024, 1(3), 1–38, https://openreview.net/forum?id=qrUdrXsiXX.84 F. Lalande, Y. Matsubara, N. Chiba, T. Taniai, R. Igarashiand Y. Ushiku, NeurIPS 2023 AI for Science Workshop,2023, https://openreview.net/forum?id=AIfqWNHKjo.© 2025 The Author(s). Published by the Royal Society of Chemistry85 Y. Suzuki, T. Taniai, K. Saito, Y. Ushiku and K. Ono, Mach.Learn.: Sci. Technol., 2022, 3, 045034.86 N. Chiba, Y. Suzuki, T. Taniai, R. Igarashi, Y. Ushiku,K. Saito and K. Ono, Commun. Mater., 2023, 4, 106.87 T. Taniai, R. Igarashi, Y. Suzuki, N. Chiba, K. Saito,Y. Ushiku and K. Ono, The Twelh InternationalConference on Learning Representations, 2024, https://openreview.net/forum?id=fxQiecl9HB.88 Y. Kadokawa, M. Hamaya and K. Tanaka, IEEE/RSJInternational Conference on Intelligent Robots and Systems(IROS), 2023, DOI: 10.1109/IROS55552.2023.10342463.89 Y. Nakajima, M. Hamaya, Y. Suzuki, T. Hawai,F. v. Drigalski, K. Tanaka, Y. Ushiku and K. Ono, IEEE/RSJInternational Conference on Intelligent Robots and Systems(IROS), 2022, DOI: 10.1109/IROS47612.2022.9981081.90 K. Shirai, C. C. Beltran-Hernandez, M. Hamaya,A. Hashimoto, S. Tanaka, K. Kawaharazuka, K. Tanaka,Y. Ushiku and S. Mori, IEEE International Conference onRobotics and Automation (ICRA), 2024, DOI: 10.1109/ICRA57147.2024.10611112.91 N. Yoshikawa, M. Skreta, K. Darvish, S. Arellano-Rubach,Z. Ji, L. Bjørn Kristensen, A. Z. Li, Y. Zhao, H. Xu,A. Kuramshin, A. Aspuru-Guzik, F. Shkurti and A. Garg,Auton. Robots, 2023, 47, 1057–1086.92 D. A. Boiko, R. MacKnight, B. Kline and G. Gomes, Nature,2023, 624, 570–578.93 Y. Yamada, R. T. Lange, C. Lu, S. Hu, C. Lu, J. Foerster,J. Clune and D. Ha, arXiv, 2025, preprint,arXiv:2504.08066, DOI: 10.48550/arXiv.2504.08066.94 K. Hatakeyama-Sato, N. Yamane, Y. Igarashi, Y. Nabae andT. Hayakawa, Sci. Technol. Adv. Mater.:Methods, 2023, 3,2260300.95 K. Hatakeyama-Sato, S. Watanabe, N. Yamane, Y. Igarashiand K. Oyaizu, Digital Discovery, 2023, 2, 1548–1557.96 K. Hatakeyama-Sato, H. Ishikawa, S. Takaishi, Y. Igarashi,Y. Nabae and T. Hayakawa, Polym. J., 2024, 56, 977–986.97 S. Jiang, D. Evans-Yamamoto, D. Bersenev,S. K. Palaniappan and A. Yachie-Kinoshita, SLAS Technol.,2024, 29, 100134.98 K. Machi, S. Akiyama, Y. Nagata and M. Yoshioka, DigitalDiscovery, 2025, 4, 172–180.99 J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya,F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman,S. Anadkat et al., arXiv, 2023, preprint, arXiv:2303.08774,DOI: 10.48550/arXiv.2303.08774.100 H. Touvron, T. Lavril, G. Izacard, X. Martinet,M.-A. Lachaux, T. Lacroix, B. Rozière, N. Goyal,E. Hambro, F. Azhar et al., arXiv, 2023, preprint,arXiv:2302.13971, DOI: 10.48550/arXiv.2302.13971.101 D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu,S. Ma, P. Wang, X. Bi et al., arXiv, 2025, preprint.102 GENIAC, https://www.meti.go.jp/english/policy/mono_info_service/geniac/index.html, (accessed April2025).103 Accelerating Life Sciences by Robotic Biology, https://www.jst.go.jp/mirai/en/program/core/JPMJMI20G7.html,(accessed April 2025).Digital Discovery, 2025, 4, 1384–1403 | 1401https://research.ibm.com/blog/roborxn-automating-chemical-synthesishttps://research.ibm.com/blog/roborxn-automating-chemical-synthesishttps://research.ibm.com/blog/roborxn-automating-chemical-synthesishttps://github.com/IBM/materialshttps://github.com/IBM/materialshttps://thealliance.ai/working-groups/materials-and-chemistryhttps://thealliance.ai/working-groups/materials-and-chemistryhttps://aiphad.org/https://aiphad.org/https://github.com/NIMS-DA/nimohttps://openreview.net/forum?id=OOxotBmGolhttps://openreview.net/forum?id=OOxotBmGolhttps://doi.org/10.48550/arXiv.2408.06292https://doi.org/10.48550/arXiv.2408.06292https://openreview.net/forum?id=qrUdrXsiXXhttps://openreview.net/forum?id=qrUdrXsiXXhttps://openreview.net/forum?id=AIfqWNHKjohttps://openreview.net/forum?id=fxQiecl9HBhttps://openreview.net/forum?id=fxQiecl9HBhttps://doi.org/10.1109/IROS55552.2023.10342463https://doi.org/10.1109/IROS47612.2022.9981081https://doi.org/10.1109/ICRA57147.2024.10611112https://doi.org/10.1109/ICRA57147.2024.10611112https://doi.org/10.48550/arXiv.2504.08066https://doi.org/10.48550/arXiv.2303.08774https://doi.org/10.48550/arXiv.2302.13971https://www.meti.go.jp/english/policy/mono_info_service/geniac/index.htmlhttps://www.meti.go.jp/english/policy/mono_info_service/geniac/index.htmlhttps://www.jst.go.jp/mirai/en/program/core/JPMJMI20G7.htmlhttps://www.jst.go.jp/mirai/en/program/core/JPMJMI20G7.htmlhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jDigital Discovery PerspectiveOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Online104 Materials Exploration Platform; Expanding Search Space byhigh-throughput technology, https://www.jst.go.jp/mirai/en/program/core/JPMJMI21G2.html, (accessed April 2025).105 Goal 3: HARADA Kanako Project j Moonshot R&D, https://www.jst.go.jp/moonshot/en/program/goal3/33_harada.html, (accessed April 2025).106 Goal 3: USHIKU Yoshitaka Project jMoonshot R&D, https://www.jst.go.jp/moonshot/en/program/goal3/36_ushiku.html, (accessed April 2025).107 AI Robot-Driven Science Initiative, https://ai-robot-science.com/en/, (accessed April 2025).108 Research innovation through autonomous-driven researchsystems, https://www.mext.go.jp/b_menu/houdou/2023/mext_00006.html, (accessed April 2025).109 R&D Process Innovation by AI and Robotics: TechnicalFoundations and Practical Applications, https://www.jst.go.jp/kisoken/presto/en/research_area/area2024-1.html, (accessed April 2025).110 Common Platform Technology, Facilities, and Equipment”mission area j JST-Mirai Program, https://www.jst.go.jp/mirai/en/program/core/index.html, (accessed April 2025).111 MEEP introduction video, short version (3 min. Englishsubtitle), https://youtu.be/Wg7E_rqMB5A, (accessed April2025).112 MEEP introduction video, full version (10 min. Englishsubtitle), https://youtu.be/_P6DomhaF8k, (accessed April2025).113 Y. Suzuki, H. Hino, T. Hawai, K. Saito, M. Kotsugi andK. Ono, Sci. Rep., 2020, 10, 21790.114 T. Fukushima, H. Akai, T. Chikyow and H. Kino, Phys. Rev.Mater., 2022, 6, 023802.115 A. Kato, T. Horinouchi, H. Ozaki and G. N. Kanda, SLASTechnol., 2024, 29, 100211.116 Y. Harazono, H. Shimono, K. Hata, T. Mitsuyama andT. Horinouchi, SLAS Technol., 2024, 29, 100200.117 S. Taguchi, Y. Suda, K. Irie and H. Ozaki, SLAS Technol.,2023, 28, 55–62.118 Y. T. Fukai and K. Kawaguchi, Bioinformatics, 2022, 39,btac799.119 M. Jinno and R. Nonoyama, Robomech J., 2025, 12, 14.120 M. Jinno, R. Nonoyama, Y. Sakurai, R. Yoshikawa,T. Kinoshita and J. Yasuda, ROBOMECH J., 2024, 11, 13.121 Digital Laboratory Consortium, https://digital-laboratory.jp/en/index.html, (accessed April 2025).122 Developing Technology to Predict Various Functions fromComplex Material Data Using Multimodal-AI, https://www.aist.go.jp/aist_j/press_release/pr2022/pr20220630_2/pr20220630_2.html, in Japanese, (accessed April 2025).123 J. Zhang, W. Wan, N. Tanaka, M. Fujita, K. Takahashi andK. Harada, IEEE Trans. Autom. Sci. Eng., 2024, 21, 5503–5522.124 M. M. Marinho, J. J. Quiroz-Omaña and K. Harada, IEEERobot. Autom. Mag., 2024, 31, 10–20.125 T. Osawa, J. Radioanal. Nucl. Chem., 2015, 303, 1141–1146.126 The use of robots in research, https://www.eisai.co.jp/innovation/research/digital/robots/index.html, inJapanese, (accessed April 2025).1402 | Digital Discovery, 2025, 4, 1384–1403127 ExaWizards co-developed “exaBase robotics powderdispensing for NEXTAGE” with Kawada Robotics thatautomates powder dispensing for manufacturing witha humanoid dual-arm robot, https://exawizards.com/archives/21909/, in Japanese, (accessed April 2025).128 K. Fushimi, Y. Nakai, A. Nishi, R. Suzuki, M. Ikegami,R. Nimura, T. Tomono, R. Hidese, H. Yasueda, Y. Tagawaand T. Hasunuma, Sci. Rep., 2025, 15, 6648.129 Development of “Lab Auto” that automates dispensationand agitation, https://prtimes.jp/main/html/rd/p/000000004.000073464.html, in Japanese, (accessed April2025).130 Autonomous lab automation system LFA series, https://www.fordx.co.jp/product/labautomation.html, inJapanese, (Accessed April 2025).131 K. Nishio, A. Aiba, K. Takihara, Y. Suzuki, R. Nakayama,S. Kobayashi, A. Abe, H. Baba, S. Katagiri, K. Omoto,K. Ito, R. Shimizu and T. Hitosugi, Digital Discovery, 2025,DOI: 10.1039/D4DD00326H.132 Standardization, https://digital-laboratory.jp/en/standardization.html, (Accessed April 2025).133 T. Ko, K. Nishiwaki, K. Terada, Y. Tanaka, S. Mitsumata,R. Katagiri, J. Taketo, N. Horiba, H. Igata and K. Mizuno,IEEE International Conference on Robotics and Automation(ICRA), 2022, DOI: 10.1109/ICRA46639.2022.9811804.134 Automatic mouse tail vein injection system AUTiv, https://www.summitpharma.co.jp/japanese/service/products/autiv/, in Japanese (accessed April 2025).135 Using collaborative robots for pouring and closing samplecontainers, https://www.lion.co.jp/en/rd/new-activity/digital/case02.php, (accessed April 2025).136 Y. Fukui, K. Minami, K. Shiba, G. Yoshikawa, K. Tsuda andR. Tamura, Digital Discovery, 2024, 3, 969–976.137 Automation Of Scientic Experiments, https://syllabus.sfc.keio.ac.jp/courses/2024_46846?locale=en,(accessed April 2025).138 Information Chemistry, https://www.chem.s.u-tokyo.ac.jp/chem_research/information_en.html, (accessed April2025).139 Information Chemistry (Lecture Catalog), https://catalog.he.u-tokyo.ac.jp/detail?code=0530085&year=2025,(accessed April 2025).140 Academy for Convergence of Materials and Informatics(TAC-MI), https://www.tac-mi.titech.ac.jp/en/, (accessedApril 2025).141 TAC-MI Curriculum, https://www.tac-mi.titech.ac.jp/en/curriculum/, (accessed April 2025).142 Intersection of Materials, Machine Learning, and Robotics, inJapanese, ed. T. Hitosugi, Tokyo Kagaku Dozin, Tokyo,2024, ISBN: 978-4807913480.143 D. Cheij, IEEE Autotestcon Proceedings. IEEE SystemsReadiness Technology Conference, 2001, DOI: 10.1109/AUTEST.2001.948916.144 SiLA Rapid Integration, https://sila-standard.com/,(accessed April 2025).© 2025 The Author(s). Published by the Royal Society of Chemistryhttps://www.jst.go.jp/mirai/en/program/core/JPMJMI21G2.htmlhttps://www.jst.go.jp/mirai/en/program/core/JPMJMI21G2.htmlhttps://www.jst.go.jp/moonshot/en/program/goal3/33_harada.htmlhttps://www.jst.go.jp/moonshot/en/program/goal3/33_harada.htmlhttps://www.jst.go.jp/moonshot/en/program/goal3/33_harada.htmlhttps://www.jst.go.jp/moonshot/en/program/goal3/36_ushiku.htmlhttps://www.jst.go.jp/moonshot/en/program/goal3/36_ushiku.htmlhttps://www.jst.go.jp/moonshot/en/program/goal3/36_ushiku.htmlhttps://ai-robot-science.com/en/https://ai-robot-science.com/en/https://www.mext.go.jp/b_menu/houdou/2023/mext_00006.htmlhttps://www.mext.go.jp/b_menu/houdou/2023/mext_00006.htmlhttps://www.jst.go.jp/kisoken/presto/en/research_area/area2024-1.htmlhttps://www.jst.go.jp/kisoken/presto/en/research_area/area2024-1.htmlhttps://www.jst.go.jp/kisoken/presto/en/research_area/area2024-1.htmlhttps://www.jst.go.jp/mirai/en/program/core/index.htmlhttps://www.jst.go.jp/mirai/en/program/core/index.htmlhttps://youtu.be/Wg7E_rqMB5Ahttps://youtu.be/_P6DomhaF8khttps://digital-laboratory.jp/en/index.htmlhttps://digital-laboratory.jp/en/index.htmlhttps://www.aist.go.jp/aist_j/press_release/pr2022/pr20220630_2/pr20220630_2.htmlhttps://www.aist.go.jp/aist_j/press_release/pr2022/pr20220630_2/pr20220630_2.htmlhttps://www.aist.go.jp/aist_j/press_release/pr2022/pr20220630_2/pr20220630_2.htmlhttps://www.eisai.co.jp/innovation/research/digital/robots/index.htmlhttps://www.eisai.co.jp/innovation/research/digital/robots/index.htmlhttps://exawizards.com/archives/21909/https://exawizards.com/archives/21909/https://prtimes.jp/main/html/rd/p/000000004.000073464.htmlhttps://prtimes.jp/main/html/rd/p/000000004.000073464.htmlhttps://www.fordx.co.jp/product/labautomation.htmlhttps://www.fordx.co.jp/product/labautomation.htmlhttps://doi.org/10.1039/D4DD00326Hhttps://digital-laboratory.jp/en/standardization.htmlhttps://digital-laboratory.jp/en/standardization.htmlhttps://doi.org/10.1109/ICRA46639.2022.9811804https://www.summitpharma.co.jp/japanese/service/products/autiv/https://www.summitpharma.co.jp/japanese/service/products/autiv/https://www.summitpharma.co.jp/japanese/service/products/autiv/https://www.lion.co.jp/en/rd/new-activity/digital/case02.phphttps://www.lion.co.jp/en/rd/new-activity/digital/case02.phphttps://syllabus.sfc.keio.ac.jp/courses/2024_46846?locale=enhttps://syllabus.sfc.keio.ac.jp/courses/2024_46846?locale=enhttps://www.chem.s.u-tokyo.ac.jp/chem_research/information_en.htmlhttps://www.chem.s.u-tokyo.ac.jp/chem_research/information_en.htmlhttps://catalog.he.u-tokyo.ac.jp/detail?code=0530085&year=2025https://catalog.he.u-tokyo.ac.jp/detail?code=0530085&year=2025https://www.tac-mi.titech.ac.jp/en/https://www.tac-mi.titech.ac.jp/en/curriculum/https://www.tac-mi.titech.ac.jp/en/curriculum/https://doi.org/10.1109/AUTEST.2001.948916https://doi.org/10.1109/AUTEST.2001.948916https://sila-standard.com/http://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387jPerspective Digital DiscoveryOpen Access Article. Published on 27 May 2025. Downloaded on 6/18/2025 2:18:29 AM.  This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.View Article Online145 OPC 30500-1 Laboratory and Analytical Device Standard,https://opcfoundation.org/documents/30500-1/, (accessedApril 2025).146 Y. Jiang, H. Fakhruldeen, G. Pizzuto, L. Longley, A. He,T. Dai, R. Clowes, N. Rankin and A. I. Cooper, DigitalDiscovery, 2023, 2, 1733–1744.147 P. Quijano Velasco, K. Y. A. Low, C. J. Leong, W. T. Ng,S. Qiu, S. Jhunjhunwala, B. Li, A. Qian, K. Hippalgaonkarand J. J. W. Cheng, Digital Discovery, 2024, 3, 1011–1020.© 2025 The Author(s). Published by the Royal Society of Chemistry148 S. Pai, K. Takahashi, S. Masuda, N. Fukaya, K. Yamane andA. Ummadisingu, IEEE/RSJ International Conference onIntelligent Robots and Systems (IROS), 2024, DOI: 10.1109/IROS58592.2024.10801608.149 Issues in the Development of Advanced Research Tools andEquipment, https://www.jst.go.jp/crds/pdf/2024/RR/CRDS-FY2024-RR-03.pdf, in Japanese (accessed April 2025).150 T. Fujimoto, Evol. Inst. Econ. Rev., 2007, 4, 55–112.151 R. Nuwer, Nature, 2023, 615, S92–S94.Digital Discovery, 2025, 4, 1384–1403 | 1403https://opcfoundation.org/documents/30500-1/https://doi.org/10.1109/IROS58592.2024.10801608https://doi.org/10.1109/IROS58592.2024.10801608https://www.jst.go.jp/crds/pdf/2024/RR/CRDS-FY2024-RR-03.pdfhttps://www.jst.go.jp/crds/pdf/2024/RR/CRDS-FY2024-RR-03.pdfhttp://creativecommons.org/licenses/by/3.0/http://creativecommons.org/licenses/by/3.0/https://doi.org/10.1039/d4dd00387j Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan Self-driving laboratories in Japan