Junhao Shi

Junhao Shi is a PhD student at Fudan University, expected to graduate in 2029, affiliated with the Shanghai Innovation Institute. He researches Vision-Language-Action models and Embodied Artificial Intelligence, with work on task-agnostic pretraining for robot learning and world-aware planning for large vision-language model planners. He is known for proposing the Task-Agnostic Pretraining (TAP) framework, which uses self-supervised inverse dynamics on unlabeled interaction data to learn transferable motor priors, and for the World-Aware Planning Enhancement (WAP) framework for LVLM planners.