The Robot Foundational Model Development Competition, which took place from April 15, 2025, to February 16, 2026, and aimed at improving the autonomous operation capabilities of assistive robots, has successfully concluded.

■ Background and Overview

This competition was conducted by AIRoA with support from GENIAC (Generative AI Accelerator Challenge), which in turn is a project led by the Ministry of Economy, Trade and Industry (METI) and NEDO to strengthen Japan's development capabilities in generative AI.

Participants were provided with a large-scale robot motion dataset comprising tens of thousands of hours collected by AIRoA. A total of six teams took on the challenge of building "Robot Foundation Models."

■ Competition Scope

Based on the vast amount of data provided, each team was tasked with developing models that enable robots to autonomously perform daily tasks in the physical world, such as using utensils to handle food or putting clothes on hangers. Teams were judged not only on simulation accuracy, but also on a rigorous evaluation of the “real-world operational success rate” in physical environments.

■ Final Results and the Winning Team

Following a strict review and comparison process based on AIRoA’s independently-established evaluation metrics, Group 6, a joint team consisting of the National Institute of Advanced Industrial Science and Technology (AIST), SoftBank Corp., and Mitsubishi Electric Corporation, was declared the winner of the competition.

Group 6’s winning strategy centered on leveraging advanced data curation technology that automatically evaluated and selected the highest quality data from the provided datasets. By fine-tuning their foundational model using this refined, high-quality data, they achieved the top score in real-world task success rates.

■ Future Outlook

This competition highlighted that the ability to select high-quality training data from massive datasets significantly impacts the performance and versatility of a model. Based on these insights, AIRoA will continue to advance the development of sophisticated robot foundation models and remain committed to promoting further integration of robots into society.