Jiahui FU 付佳慧

My research focuses on building robots that see, predict, and act by linking deep spatial perception with autonomous manipulation. During my PhD, I developed object-centric 3D representations and neural implicit models for long-term SLAM, enabling robots to maintain persistent, object-aware understanding of changing environments. Recently, I have been leveraging this spatial foundation to treat video generative models as predictive world models for zero-shot robotic manipulation. By grounding high-level visual foresight into physical interaction, my current work develops frameworks that translate generative "commonsense" into robust, closed-loop robot execution.