While still in the United States, Duan led the creation of MolmoAct2, an AI model that helps robots understand their surroundings. The model achieved the highest average score on 13 robotics‑related tests, outperforming both GPT‑5 and Google’s Gemini Robotics‑ER 1.5.

Since returning, Duan secured S$2 million (US$1.6 million) under Singapore’s National Robotics Programme. The funding will support a project that gathers real‑world robot data and builds common evaluation methods for diverse, complex tasks beyond controlled labs.

His team works with the data company FrodoBots to collect information from humanoid robots used in homes across Southeast Asia. Duan hopes the research will eventually produce robots that assist older adults at home and perform physically demanding industrial work, raising people’s quality of life.

The significance of Duan’s work lies in linking cutting‑edge AI research with everyday applications. By moving robot learning into real environments, Singapore can strengthen its AI ecosystem and create technologies that directly improve daily life.