I recently graduated with a B.Eng. in Software Engineering from Beihang University, with a minor in Law, and will join the School of Information, Renmin University of China as an incoming PhD student. My current research focuses on embodied intelligence, especially vision-language-action models and latent action supervision. Previously, I worked on spatial-temporal data modeling and imputation.
A systematic study of latent action supervision for vision-language-action models. The paper compares image-based and action-based latent action formulations in a unified VLA setting and shows the effectiveness of directly supervising VLMs with discrete latent action tokens.
ACM Transactions on Intelligent Systems and Technology
This work studies spatial-temporal imputation with an LLM-native framework that converts spatial, temporal, and cross-variable correlations into semantic tokens, improving correlation-aware reasoning on real-world imputation tasks.