The Matsuo-Iwasawa Laboratory (“Matsuo Lab”), Graduate School of Engineering, The University of Tokyo, in collaboration with the AI Robotics Association of Japan (“AIRoA”), of which Yutaka Matsuo serves as director, has started the development and large-scale data collection of an infrastructure model in the robotics domain under the “Investigation on the Construction of Robot Motion Dataset for Learning and Validation of its Applicability to Robot Infrastructure Models for Accelerating the Development of General-purpose Robots Using AI” of the New Energy and Industrial Technology Development Organization (NEDO). The project started with the development of a basic model and large-scale data collection in the robotics area under the “Study on the Construction of Robot Motion Dataset for Learning and Verification of Applicability to Robot Basic Model for Accelerating the Development of General-purpose Robots Using AI. We aim to collect tens of thousands of hours of data to build a robot AI model and strengthen our technical development system through collaboration with researchers and engineers.
Background
In recent years, the dramatic evolution of large-scale language models (LLMs) and visual language models (VLMs) and their application to the robotics domain has opened up the possibility for robots to flexibly respond to more generic tasks. pi0 (Physical Intelligence) and GR00T (NVIDIA) and other precedents, it is believed that robotics efforts that apply multimodal large-scale models will occupy an important position as the technological foundation for the robotics domain in the future.
Against this backdrop, the construction of versatile basic models and the collection of large-scale data to enable robots to acquire diverse skills and continuously learn and improve in real environments have become important international research themes.
Details of Initiatives
Matsuo-Iwasawa Laboratory will promote the following activities in collaboration with AIRoA in this initiative.
Through the promotion of these activities and future verification and PoC in the field environment, we aim for social implementation of AI robots in the fields of retail, manufacturing, logistics, and others.
- Development of Robot Base Models
To ensure that robots have the same level of contextual understanding as humans and can act flexibly in unknown environments, we build pre-trained models using large-scale multimodal data. The design features both versatility to accommodate a variety of robots and sensor information, and adaptability to on-site needs. - Support for the construction of robot operation data on a scale of tens of thousands of hours
We are also proactively involved in the development of data obtained from large-scale robot operations in Japan and overseas locations in terms of architecture design and evaluation index formulation. We accelerate the development of high-quality, high-density data sets suitable for model learning. - Human Resource Development through Robot Infrastructure Competition
We aim to develop human resources for AI robotics through lectures on large-scale language models (LLM) and other topics held by the Matsuo Lab, as well as by gathering interested members from the industrial world and conducting development in a competition format.
About the Robotics Research Unit, Matsuo-Iwasawa Laboratory, The University of Tokyo
The Matsuo-Iwasawa Laboratory is actively promoting robotics research with the aim of elucidating “what intelligence is” through real machines that interact with the real world.
In addition to actual equipment such as robot arms and mobile manipulators, the laboratory is equipped with a variety of hardware environments including various simulators, VR devices, and multimodal sensors, creating a research infrastructure that enables a wide range of verification from virtual to real environments.
As a research and development community, TRAIL (Tokyo Robotics and AI Lab), in which students and researchers from both inside and outside the university participate, is also in operation and has now expanded to a 50-member scale. The group has won top prizes in both the national and world RoboCup competitions, and has formed a research system equipped with practical technical and team skills.
In April 2025, we will also launch the Physical AI course, an educational course to systematically learn the fusion of AI and robotics, and focus on fostering the next generation of researchers and accelerating applied research. Our research results to date have been accepted by ICRA and other top international conferences, and we have achieved results in a wide range of areas from basic to applied research.
This project will play a central role as a center for the development of robotics infrastructure models, taking advantage of the expertise in large-scale model development and application development that the laboratory has accumulated to date.
Future
This initiative aims to introduce AI robots in Japanese industry by combining cutting-edge research in AI technology with field knowledge in robotics. Matsuo-Iwasawa Laboratory is actively recruiting researchers and research engineers to promote robotics research together. For more information, please see the following job posting: https: //herp.careers/v1/weblab/crh6BsRjCxX1
