HOME―Our Project

Our Project

A background in launching project

This project is conducted under commissioned projects promoted by the Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO):

Post-5G Information and Communication System Infrastructure Enhancement R&D Program / Development of Data Platforms for Generative AI Foundation Models in the Robotics Field
Post-5G Information and Communication System Infrastructure Enhancement R&D Program / Research and Development on the Establishment of Data Ecosystems (GENIAC)Title: “Research and Development of a Data Ecosystem for the Self-Sustaining Expansion of Robot Action Data in Real Homes”

Data Collection

Large-scale robot motion data collection to support the development and open release of robot foundation models. In FY2025, we completed the collection of approximately 80,000 hours of data with the Human Support Robot (HSR). Going forward, we will expand training-data collection across multiple robot platforms.

Data Collection with Physical Robots

Building the methods and operational structure to collect motion data efficiently in real-world environments. Single-arm mobile manipulators and dual-arm manipulators are teleoperated with a range of controllers to gather data.

Single-arm mobile manipulator (Human Support Robot, HSR)

Combining omnidirectional mobility with a single arm, the HSR enables efficient motion-data collection and validation in real-world environments through both teleoperation and autonomous operation. The knowledge and operational infrastructure developed through large-scale HSR data collection are being extended to data-collection efforts using YUBI, G2, and other platforms. We are also advancing validation and data collection in real home environments, including at Toyota Woven City.

Site Video

YUBI (Yielding Universal Bidigital Interface)

A finger-tracking gripper and a separate 6-DoF mechanism give operators intuitive control of both arms, A general-purpose data-collection interface that combines finger-tracking grippers with separate 6-DoF mechanisms, enabling intuitive dual-arm operation. It supports efficient large-scale robot motion-data collection while reducing the burden on operators.

Paper Site Video

Dual-arm mobile manipulator (G2 and others)

A dual-arm platform capable of advanced bimanual coordination, enabling data collection for complex tasks in homes and other real-world environments.

Domestic General-Purpose Robot Development Competition

A competition aimed at strengthening the competitiveness of Japan’s robotics industry by promoting the development and manufacturing of domestically produced general-purpose robots suited to real-world data collection and deployment. Robot platforms that meet the evaluation criteria will undergo further improvement and preparation for mass production, with the aim of deploying them as robots for data collection.

Site

Model Development

Using the training data we collect, we pursue the full model-development pipeline — from pre-training Vision-Language-Action (VLA) foundation models for general-purpose motion generation to post-training and performance evaluation for deployment on physical robots.

Pre-training

We pre-train our own VLA foundation models, using multi-embodiment data from YUBI, G2 and other platforms, with the aim of improving general-purpose motion generation and generalization.

Site Video

Post-training

We improve VLA foundation model performance for deployment on physical robots, validate motion-adaptation methods, and evaluate performance on real hardware.

AI Robotics Talent Development

We aim to raise the overall level of researchers and engineers in Japan’s AI × robotics field while providing development opportunities for early-career researchers. Through competitions, we cultivate talent with hands-on implementation skills, while overseas placements at leading research institutions help develop highly skilled researchers and engineers.

Developing AI Robotics Talent

Developing practical talent with strong implementation skills through two key initiatives: competitions and overseas research placements. By designing competitions that leverage the software and data developed through this project, we aim to broaden the pool of researchers, further develop top-tier talent, and strengthen the competitiveness of Japan’s AI robotics field.

Physical AI Robot Challenge 2026 (PARC2026)

A Vision-Language-Action (VLA) development competition in which participants tackle cutting-edge challenges in Physical AI, designed to cultivate developers in the field. Co-hosted with the Matsuo-Iwasawa Lab at the University of Tokyo, the competition is part of our efforts to build Japan's Physical AI ecosystem.

Overseas Research Placements

AIRoA will select several candidates, including top performers from our competitions and promising young researchers with proven track records, and place them at leading research institutions overseas. They will work on research topics aligned with AIRoA’s R&D activities and share their findings and insights both within and beyond AIRoA, helping raise the overall level of research in Japan.

Click here

Career

Click here