Recruitment Begins for Student AI Foundation Model Development Contest: "Operating Robots with LLMs"

The University of Tokyo's Matsuo-Iwasawa Laboratory will be responsible for the planning and operation of the "GENIAC-PRIZE 2026 AI Development Contest," which is being conducted as part of the "GENIAC-PRIZE 2026 Theme 2: Development of Open Foundation Models for Developer Training (Students) Towards Physical AI" mandated by the Ministry of Economy, Trade and Industry and NEDO. Recruitment of participating members will begin in preparation for the contest's launch around November 2026.

[Background and Overview]

We are entering an era of "Physical AI," where AI recognizes, judges, and acts in the real world. The "AI Robotics Strategy" formulated by the government in March 2026 estimates that the multi-purpose robot market will expand to 60 trillion yen by 2040, while a shortage of 3.39 million personnel to support this is projected, making the training of next-generation developers a national challenge.

"GENIAC-PRIZE" (*) is a program that aims to boost domestic AI development capabilities and promote social implementation by awarding prizes based on results. As part of this initiative, this contest will be held as a team-based, open foundation model development competition to discover and nurture young developers (students) who will lead the next generation of LLM and Physical AI technologies.

[Key Features of This Contest]

1. Opportunity to Challenge Full-Scale Foundation Model Development as a Student

Practical Task Design Participants will work on developing a robot control code generation model using LLMs. The task design allows both LLM development veterans and those aspiring to enter Physical AI to leverage their respective strengths.

Development Environment Valued at Approximately 400 Million Yen We will provide the latest GPUs, such as NVIDIA B300 (planned) and RTX PRO 6000 Blackwell for simulators, which are difficult for individuals to access.

Team Development with Students Nationwide Planned for up to 500 participants across 17 teams, with each team comprising about 20 members. Participants will gain experience in practical team development, including role allocation and progress management.

2. Total Prize Money of Approximately 30 Million Yen Approximately 30 million yen in total will be awarded to top teams (top 3), individual awards, and special awards for free-topic challenges.

[Contest Overview]

Participants will work on developing a "robot operation agent" that generates robot control code from language instructions and environmental information using LLMs.

Example: Develop an AI that can understand instructions like "Open the drawer and put the bowl inside," and generate appropriate control code for the robot to execute the action.

For those who have been involved in LLM development, this task is a challenge to apply their accumulated LLM development skills in inference, code generation, and post-training to the new challenge of "operating robots." For those interested in robotics and Physical AI, it is an opportunity to practically develop the entire process from language instruction to environmental recognition and action generation.

This contest allows both those who want to leverage their LLM development experience and those who want to challenge Physical AI to utilize their respective strengths.

The evaluation will use "CaP-X (Code-as-Policy eXtended)," an evaluation platform for Physical AI. Development and evaluation will be completed within simulator environments such as Robosuite, LIBERO-PRO, and BEHAVIOR, allowing students without practical robot experience to gain hands-on experience in cutting-edge technology development.

* For details on GENIAC-PRIZE, click here.

[Recruitment Overview]

Target Audience: Students of Japanese nationality aged 12 to 29 (※1)

Contest Task: "CaP-X: Robot Operation Evaluation by Code Generation Agent" (Generating robot control code from language instructions and environmental information using LLMs, and evaluating task achievement rate and failure factors based on execution results)

Number of Participants: Approximately 500 participants expected

Number of Teams: 17 teams planned (approximately 20 members per team) (※2)(※3)

Event Period: November 2026 - March 2027 (planned)

Application Deadlines:

Team Leader Application Deadline: July 31, 12:00 PM

Individual Member Application Deadline: August 31, 12:00 PM

[Public Information Session]

A public information session will be held to provide details about the contest, the computational resources provided, and preparations for participation. We encourage those considering participating in the contest to attend.

Date and Time: July 13, 2026 (Mon)

How to Apply: Please check the details here and apply.

[Details and Application Method]

For detailed application requirements (such as required course completion and operational requirements), application for the public information session, and entry into this contest, please visit the dedicated contest page below.

Dedicated Contest Page (Check Details, Apply for Information Session, Enter Contest Here)

(※1) Individuals born between April 2, 1996, and April 1, 2014, who are enrolled in schools under the School Education Act (Law No. 26 of 1947) or equivalent overseas educational institutions, or training institutions established by national or local governments or similar public organizations, and who hold Japanese nationality as of April 1, 2026.

(※2) It is not necessary to form a team at the time of application. Team formation will take place after the applicants are confirmed.

(※3) All of the above are subject to change depending on the circumstances.