Shota Kato

Establishing Foundations for Autonomy-Driven Physical Modeling Using Multimodal Data

Grant No.：JPMJPR25T1

Researcher : Shota Kato

Outline

This project aims to establish an autonomy-driven physical modeling cycle that integrates literature-derived knowledge with small experimental datasets. We will develop three core components: (i) symbolic regression guided by physical constraints, units, and prior relations extracted from documents, (ii) multimodal evaluation that assesses models by combining equations, plots, and natural language explanations to ensure extrapolative validity and interpretability, and (iii) Bayesian optimization capable of proposing new experimental conditions, including language-based actions such as adding new measurements. The framework will be validated on three representative processes—CSTR, polymer rheology, and the Czochralski crystal growth process—demonstrating scalability across complexity levels. The expected outcome is a general methodology that reduces experimental cost and accelerates scientific discovery.

Makoto P. Kato

Client-side Information Retrieval for Research and Development

Grant No.：JPMJPR25T2

Researcher : Makoto P. Kato

Outline

In this research project, we develop technologies that employ client-side AI running on researchers’ PCs to continuously monitor research activities and information-seeking behaviors, autonomously retrieving and providing information necessary for advancing research. This approach reduces privacy and security risks while enabling the timely delivery of information that researchers intend to search for, information they did not explicitly intend to search for but is still useful, and information that contributes to the progress of their research.

Kanako Kumada

Development of an Autonomous Photocatalytic Reaction System

Grant No.：JPMJPR25T3

Researcher : Kanako Kumada

Outline

This research aims to develop an autonomous photoreaction design system that integrates AI, robotics, and machine learning, focusing on the photodegradation of persistent organic pollutants (PFAS) using molecular photocatalysts. By combining a structure–function–condition database based on literature and theoretical calculations, an autonomous experimental system, and a reaction prediction model in a unified framework, the project seeks to achieve highly accurate catalyst design and reaction condition optimization without relying on conventional empirical rules, thereby revolutionizing the methodology of scientific discovery.

Atsushi Shibai

Development of a tailor-made tool for automating cell culture experiments

Grant No.：JPMJPR25T4

Researcher : Atsushi Shibai

Outline

This project aims to develop software that enables life-science researchers to automate experiments with flexible, modular configurations. The design centers on a unified Python API layer with plug-in device adapters, providing integrated control of existing instruments, with built-in AI to assist experimental design and planning. The platform is designed to let researchers phase in automation stepwise, tailored to their goals, expertise, and laboratory environment. The project plans to validate utility through two proof-of-concept studies in bacterial research and to release the system as an open platform that others can adopt, adapt, and redesign.

Yaonan Zhu

Teleoperation and Foundation Models based Humanoid Research Support System

Grant No.：JPMJPR25T5

Researcher : Yaonan Zhu

Outline

This project leverages foundation models for language interpretation and task planning, together with teleoperation for skill teaching. It aims to enable the autonomous execution of burdensome R&D tasks, such as transporting experimental materials and operating laboratory equipment. Building on these efforts, the project envisions establishing a humanoid research support platform that can be seamlessly introduced into existing research environments, thereby facilitating researchers’ creative activities.

Naoya Chiba

Unified 3D Data Representation with Neural Fields

Grant No.：JPMJPR25T6

Researcher : Naoya Chiba

Outline

3D data appears in research and development across a wide range of fields. This proposal aims to extend Neural Fields and establish them as a versatile 3D data representation applicable across variable research fields. By developing Neural Fields as a canonicalized and physically consistent data representation, and by exploring diverse applications, I will try to expand the range of usable applications of Neural Fields. Through this, we aim to enable sharing methods and task collaboration on 3D data.

Shinya Mine

Data-Driven Catalyst Discovery Integrating Transfer Learning with Automated Experimentation

Grant No.：JPMJPR25T7

Researcher : Shinya Mine

Outline

In this study, I aim to establish a broadly applicable catalyst activity prediction model through transfer learning, enabling adaptation across diverse catalytic reaction systems. In parallel, I will construct a closed-loop catalyst discovery platform that integrates automated synthesis robotics with high-throughput evaluation to accelerate data acquisition and enhance reproducibility. The effectiveness of this approach will be demonstrated through the development of catalysts for CO2 reduction reactions, thereby validating the practical utility of the system. Ultimately, my goal is to advance a new paradigm in catalyst research by uniting state-of-the-art machine learning with automated experimentation.

Hiroki Yasuga

A Liquid‑Handling Platform Based on Microactuators for Grabbing and Releasing Microdroplets

Grant No.：JPMJPR25T8

Researcher : Hiroki Yasuga

Outline

In this study, I endeavor to realize a liquid-handling platform characterized by minute sample volumes, a compact footprint, and automation by means of a novel liquid actuation mechanism that harnesses the microscale phenomenon whereby surface tension drives liquids to converge into spherical droplets. Liquid-handling operations—including dispensing, transport, and mixing of liquid samples—are important in drug discovery and clinical testing settings, and this project seeks to enable innovations that fully automate these processes.

Daisuke Yamazaki

Development and Validation of DNA language model-based biomolecule designing algorithms

Grant No.：JPMJPR25T9

Researcher : Daisuke Yamazaki

Outline

Large Language Models (LLMs) have been extensively developed in recent years, and they have started to be utilized in designing biomolecules such as genes and proteins. In this project, DNA language model-based biomolecule designing algorithms will be built from small, but high quality datasets acquired from automated measuring devices. Furthermore, generated models will be validated and applied to develop the new digital bioanalyses for clinical diagnostics.