Date: 16:00-18:00, Friday, July 17, 2026
講演者：小林 大輝
Speaker: Hiroki Kobayashi
題目：DAC実験の自動化、そしてその先にあるもの
Title: Automated DAC experiments, and beyond
要旨：「自分の実験をもっと効率化できるのではないか？」実験家なら誰しもこう感じたことがあるのではないでしょうか。近年、材料科学分野における実験自律化の動きは凄まじく、自律的に考え昼夜動き続ける実験室を目の当たりにして、「実験の設計」という概念が以前とは比べ物にならないほど広範になってきていると感じます（Szymanski
et al., Nature 2023, 624,
86）。その一方で、実験を自動化、ひいては自律化しようという動きは、まだ一部の分野のみにとどまっているような印象も受けます。少なくとも私の周りでは、あまりそのような動きは感じられません。それには、既存の設備の延長としての自動化が、直接的な成果（論文）にも、卓越した科学的成果にも、必ずしもつながらないという事実、それに反して実装にかかるそれ相応の手間、さらにはセル組みをロボットにやらせるのが難しいという高圧実験特有の障壁の存在などが要因かもしれません。
しかし、私はこの5年間、多くの方々の多くの種類の実験に貢献しうる基盤環境の構築は、装置開発からソフト面まで、幅広く価値の高いものだと感じてきました。そこで、私は最近、実験の効率化や自動化のための環境整備に取り組んでいます。全自動実験を実現するため、まずさまざまな装置を自前のプログラムを通じて制御できるようにし、次にそれらの装置を自由自在に組み合わせて取り扱えるようにするという二つの段階を踏んで実装を進めています。また、自動化の副産物として、これまでできなかった発展的な実験操作がいくつかできるようになり、全自動でなくても、より多様で高度な実験の実現の足がかりも作ることができました。ただ、「良い結果」を合成された物質の物理量から定義可能な新規材料合成と違い、私たちの分野では、どのような条件が達成されれば実験が成功なのかが曖昧です。そこに、自動化からさらに進んで、自律化へ向けて私たちが超えなければならない壁があると感じます。こうした基盤環境整備の未来について、目先の成果にとどまらない広い可能性を見据えながら、皆さんと一緒に議論させていただければ幸いです。
参考資料へのリンク:
(1) BL-18C controller, a Python-based platform for advanced and
automated experiments at BL-18C
https://github.com/khsacc/bl18c_controller
(2) “PaceMaker”, a Druck PACE5000 pressure controller
https://github.com/khsacc/PaceMaker
(3) “FluoraPressée”, a spectroscopic measurement/analysis platform
designed for rapid pressure determination in high-pressure experiments
Abstract: “Could my experiments be made more efficient?” This is a
question that almost every experimentalist has probably asked at some
point. In recent years, the rapid progress of autonomous
experimentation in materials science has been remarkable. Seeing
laboratories that can autonomously design, execute, and refine
experiments around the clock has made me realise that the very concept
of experimental design has expanded far beyond what it once was
(Szymanski et al., Nature 2023, 624, 86).
At the same time, however, efforts toward experimental automation, and
ultimately autonomy, still seem to be confined to relatively few
research fields. At least in my own research community, such
activities remain relatively uncommon. This may be because automation
built upon existing experimental infrastructure does not necessarily
translate directly into scientific publications or groundbreaking
scientific discoveries, despite requiring substantial development
effort. In high-pressure research, additional obstacles also exist,
such as the inherent difficulty of automating pressure-cell assembly.
Throughout the five years of experience in this group, I have become
convinced that developing a shared experimental infrastructure, from
hardware to software, can provide lasting value by supporting
researchers working across a broad range of related experimental
fields. Motivated by this perspective, I have been developing an
integrated environment that enables efficient and automated
high-pressure experiments.
The implementation followed two stages. First, every experimental
device was brought under the control of a unified software platform
developed in-house. Second, these independently controlled devices
were integrated into a flexible framework that allows them to be
freely combined into complex experimental workflows. As a by-product
of this automation effort, several advanced experimental procedures
that had previously been impractical have become possible.
Consequently, even without fully autonomous operation, my platform
already enables more sophisticated and versatile experiments than
before.
Nevertheless, unlike autonomous materials synthesis, where « good
results » can often be quantified through measurable physical
properties of synthesised materials, the success criteria in our field
are far less well defined. Determining what constitutes a successful
experiment remains an inherently ambiguous problem. I believe this
ambiguity represents one of the fundamental challenges that must be
overcome in moving beyond automation toward truly autonomous
experimentation.
In this presentation, I would like to share our progress and plans to
build this experimental infrastructure and discuss its future
potential, looking beyond immediate research outputs toward the
broader possibilities that automation and autonomy may offer to many
areas of experimental research.