Moonshot Goal 3
General-Purpose Autonomous Humanoids that Grow Through Multi-Layered Teaching and Inheritance
Robots that are taught, learn, and pass skills on. A coalition of young PIs building Physical AI foundations from Japan.
A Future Scenario for 2050
Vision1Day 1: Teaching before opening
In the 30 minutes before opening, the shop owner teaches the robot the day’s tasks.
2Growth: The robot runs the job
Through learning, it comes to handle multiple tasks autonomously.
3Inheritance: Teaching new robots and people
It passes tips and knowledge on to newly arrived robots and human staff, raising the whole team.
4Continuity: The owner’s knowledge lives on
Knowledge is received from the experienced owner and passed on to the next generation.
News
- Preview version of the project website is now available
- [Dummy] Kickoff meeting held
- [Dummy] Multiple papers accepted at IROS
Team
Young PI Collective
Nine PIs integrate their research on a shared platform.
Kento Kawaharazuka
Masaki Murooka
Shuhei Kurita
Asako Kanezaki
Satoshi Yagi
Kazuki Miyazawa
Kohei Honda
Chie Hieida
Taisuke Kobayashi
Research Architecture
3 research themes, 9 research projects
Projects are not a fixed hierarchy. Every PI works across the three themes, integrating results on a shared platform.
Theme 1
Humanoids Suited for Teaching and Inheritance
- 01Body Design and Learning Control
- 05Inheritance Humanoid Systems
- 06Teaching Humanoid Systems
Theme 2
Foundation Systems Enabling Teaching and Inheritance
- 02Behavior Foundation Models
- 03Robot Foundation Models
- 04Spatial Foundation Models
- 09Robot Learning Theory
Theme 3
Task Execution and Social Implementation Based on Teaching and Inheritance
- 07Growth through Real-World Experience
- 08Emotion Models
Roadmap
Milestones toward 2030
2026
Prototypes and initial foundations
Initial research platforms and foundation models.
2028
Demonstration under known conditions
Teaching and inheritance in known environments, tasks, and bodies.
2030
Real-robot demonstration under unknown conditions
Adaptation to unknown environments, tasks, and bodies; spot work; skill inheritance on real robots.
Platforms
Research Platforms
Early exploration runs on off-the-shelf and joint platforms, in parallel we develop MEVA — our own open-source humanoid — and ultimately integrate the results.
- Developed in this project
MEVA
An open-source humanoid to be developed by this project.
- Joint research
PHF
A compact humanoid developed by Toyota Motor Corporation, cooperating on housework and moving-task demonstrations. Content will be confirmed with the company before publication.
- Off-the-shelf
Unitree G1
An off-the-shelf platform used for early exploration. Not a robot developed by this project.
Join Us
Join the project
Together with researchers, students, and industry, we are building Physical AI foundations for teaching and inheritance.
Researchers & students
Research participation, RA positions, and internships in PI labs.
Student community
Student community activities around Physical AI.
Industry
Co-creation of demonstration tasks, joint research, and technology transfer.
Institutions
Participating institutions
The University of Tokyo, AIST, National Institute of Informatics, Tohoku University, The University of Osaka, Nagoya University, The University of Electro-Communications
Toyota Motor Corporation
Japan Science and Technology Agency (JST)