일시: 2025. 12. 15. (월요일), 오전 11시~ 오후 12시

장소: 공학원 452B호

Speaker: Chuan Wen, Ph.D. - Assistant Professor at Shanghai Jiao Tong University

Title: Data Scaling on the Job

Abstract:

Modern robot learning has largely centered on two pillars: collecting massive datasets and training increasingly capable policies. Yet, the story often stops there. The final step, deployment, is treated merely as an evaluation phase, rather than the crucial stage it truly is. In practice, deployment is not just execution—it’s data generation. Each trajectory a robot produces in the real world, whether a success or a failure, encodes valuable information about policy weaknesses, and environment dynamics. Ignoring this stream of experience means missing an opportunity. In this talk, I’ll present our recent efforts under the theme of Data Scaling on the Job, a framework for turning deployment into an engine for continual data accumulation and policy improvement. By leveraging online experience, augmented by strategic human intervention and policy exploration, we move toward a future where robots not only execute what they’ve learned, but learn as they execute, reducing human burden and enabling truly scalable autonomy.

Bio:

Chuan Wen is a tenure-track Assistant Professor at School of Artificial Intelligence, Shanghai Jiao Tong University. His research interest focuses on embodied AI, especially scalable robot learning and multi-modal motor control. He obtained his Ph.D. degree from Tsinghua University, advised by Prof. Yang Gao. He was a visiting scholar at UC Berkeley, advised by Prof. Pieter Abbeel and Dr. Xingyu Lin. He also worked closely with Prof. Dinesh Jayaraman at UPenn GRASP lab during his Ph.D.