ugo and FastLabel Launch 'ugo VLA Model Development Training Program' for Physical AI Developers

ugo, Inc. (Headquarters: Chiyoda-ku, Tokyo; CEO: Ken Matsui), a developer of domestic AI robots, and FastLabel, Inc. (Headquarters: Shinjuku-ku, Tokyo; President: Takeshi Suzuki) have announced the launch of the 'ugo VLA Model Development Training Program powered by FastLabel.' This practical program utilizes the domestic humanoid robot 'ugo Pro R&D model' to enable enterprises, universities, and research institutions to engage with Vision-Language-Action (VLA) model development from the early stages.

Participants in the training program can experience the end-to-end process, including task design, data collection for imitation learning, data quality management, VLA model development (fine-tuning), machine evaluation, and reporting. Even companies without internal physical AI expertise or development infrastructure can gain hands-on experience with robot arm development and verification processes, facilitating concrete evaluations for the internal use of physical AI.

[Background] In recent years, amid the aging population and labor shortages, improving efficiency and labor reduction in fields like security, inspection, transportation, and guidance have become societal issues. Consequently, physical AI, including VLA models that integrate vision, language, and behavior to flexibly control robots, is gaining attention as a next-generation technology capable of adapting to various field conditions that were difficult to address with traditional rule-based automation.

However, when companies consider implementing physical AI, they require an environment where they can try the end-to-end process from procuring robots and collecting imitation learning data to model development and machine evaluation. Challenges such as not knowing where to start, lacking internal development/verification knowledge, and the high initial burden of launching a Proof of Concept (PoC) have acted as major barriers to early adoption.

To address these issues, ugo and FastLabel have jointly planned and developed a corporate training program that integrates physical robots, data collection and maintenance, VLA model development, evaluation, and lectures. This allows companies to grasp the potential of physical AI and specific issues for internal application through development and verification processes using actual robots.

[Program Content] The 'ugo VLA Model Development Training Program' comprehensively packages the steps necessary for the initial verification of VLA model development. It provides end-to-end support, including setup and operating environment construction of the 'ugo Pro R&D model,' task design, imitation learning data collection, data quality management, VLA model training, deployment, machine evaluation, analysis, and reporting. The program is structured to accumulate basic knowledge and practical wisdom about physical AI within the company.

The training program emphasizes hands-on support aimed at future in-housing and formal adoption decisions. By leveraging the best practices and formats for VLA model development held by ugo and FastLabel, it is possible to carry out the initial development and verification process and accumulate knowledge in as little as approximately three months from the initial inquiry.

[Future Outlook] ugo and FastLabel have previously collaborated on R&D and sales activities in the AI robotics domain. Through the provision of this training program, they intend to create an environment where more companies can initiate initial verification of physical AI development, contributing to the expansion of robot utilization possibilities in Japan.