< ICML 2026 受賞者の写真。左から：ナ・ビョンフ（KAIST KIロボット研究所）、 Jiseok Kwak (データサイエンス研究科博士課程学生); Suhyeon Jo (産業システム工学科博士課程学生);キム・テウ (マスター >
A KAIST research team won the International AI Challenge held in conjunction with ICML 2026 for developing a system that can interpret images and text, apply the laws of physics, and explain their reasoning. The results demonstrate KAIST’s world-leading ability in AI to understand the physical world, with potential applications in aircraft design, robotics, autonomous vehicles, and space systems.
KAIST (Chairman Bae Choong-sik) announced on July 23 that a team led by Professor Il-Chul Moon of the KAIST AX division won the Visual Grounded Physics Problem Solving Challenge hosted by the AI4Math Workshop, the official workshop of the 2026 International Conference on Machine Learning (ICML).
The competition was held from May 1st to June 16th, and 139 teams participated, including participants from ETH Zurich, Fudan University, and Shanghai Innovation Research Institute. The KAIST team received the highest score in the final evaluation and won the top award at the ICML awards ceremony held in Seoul on July 11th.
This challenge assessed whether an AI can understand physics problems presented through images and text, apply the laws of physics, and generate both a correct answer and an inference process. Participating systems had to integrate multimodal information and use principles such as Newtonian mechanics to logically solve problems. The focus was not on simple calculations, but on the ability to understand and explain complex physical phenomena.
This result suggests that AI can go beyond solving test problems to understanding the real physical environment and optimizing its design and operation. This technology has the potential to support aircraft, robotics, autonomous driving, and space systems where decisions must take into account real-world physical conditions.
Five researchers from KAIST’s Applied Artificial Intelligence Laboratory (AAILab) participated, including Jiseok Kwak, a doctoral student in the Department of Industrial Systems Engineering. The team built a multi-agent AI architecture in which several underlying models act as independent agents, validating and discussing each other’s answers.
< AI4Math Track 3 の物理コンテストの問題の例 >
This approach reduced errors from individual models, increased inference reliability, and enabled teams to achieve the best performance in the competition. We also demonstrated new possibilities for AI reasoning about complex physical phenomena and agent AI systems.
“By converting the Foundation model into an agent and watching multiple agents arrive at the correct answer through discussion and verification, we realized that no single Foundation model will monopolize the future,” said Professor Moon.
“Although this competition took the form of a physics problem-solving test, the underlying technology is about the development of AI for the optimal design and operation of physical systems, such as the AI-based aircraft design and operation that Boeing is researching,” he added.
This research was supported by the Information and Communication Technology Planning and Evaluation Institute and the ITRC Defense Group Systems Research Center.
/Open to the public. This material from the original organization/author may be of a contemporary nature and has been edited for clarity, style, and length. Mirage.News does not take any institutional position or position, and all views, positions, and conclusions expressed herein are those of the authors alone. Read the full text here.
