Embodied AI (EAI) Lab

Welcome! The Embodied AI lab are dedicated to developing autonomous, real-world robotic systems. We approach this challenge from three perspectives. The first is Action Control. We train robot policies that allow robots to perform mobile manipulation tasks with high efficiency, effectiveness, and robustness. The second is Learning and Exploration. We develop algorithms that enable physical agents to explore their environments and continually acquire new skills with minimal human intervention. The third is Dynamic Modeling. We build dynamic systems that enhance environmental understanding by simulating the outcomes of physical interactions. Together, these three research topics aim to create robots that function as intelligently as animal beings.

Code for our projects is available at https://github.com/EmbodiedAI-NTU.

We are looking for Ph.D/Master/undergraduate students interested in robot learning, computer vision, and machine learning. If you are an undergraduate at NTU interested in research positions, please fill out this form. If you are interested in Master/Doctoral positions, please fill out this form.

Recent Publications (^∗ Equal contribution. ^† Corresponding author.)

Learning Conditional Source Distribution via Flow Reversal for Temporal Flow Matching
Kuan-Hsun Tu^*, Hsuan-Chi Liu^†*, Jia-Wei Liao^*, Chien-Sheng Chiang, Cheng-Fu Chou, and Tsung-Wei Ke

In submission

arXiv / Code / Project Page
Task Inference Beyond Least Squares in Behavioral Foundation Models
Kuan-Hsun Tu^1*, Chien-Sheng Chiang^1†*, Hsin-Wei Chen¹, Ping-Chun Hsieh², and Tsung-Wei Ke¹

¹National Taiwan University, ²National Yang Ming Chiao Tung University

In submission

arXiv / Code / Project Page
TaRL: Learning General and Physical Rewards from Tactile Demonstrations
Po-Yi Wu¹, Dao-Jan Chang¹, Shang-Ya Hsiao², Hong-Ming Chen², Yu-Cheng Su², Tsung-Wei Ke¹

¹National Taiwan University, ²Delta Electronics

ArXiv 2026

arXiv / Code / Project Page
COT-FM: Cluster-wise Optimal Transport Flow Matching
Chien-Sheng Chiang^*, Kuan-Hsun Tu^*†, Jia-Wei Liao^*, Cheng-Fu Chou, and Tsung-Wei Ke

CVPR 2026

arXiv / Code / Project Page
DexMan: Learning Bimanual Dexterous Manipulation from Human and Generated Videos
Jhen Hsieh¹, Kuan-Hsun Tu¹, Kuo-Han Hung^1,2, Tsung-Wei Ke¹

¹National Taiwan University, ²Stanford University

ArXiv 2025

arXiv / Code / Project Page
Information Seeking for Robust Decision Making under Partial Observability
Djengo Cyun-Jyun Fang¹, Tsung-Wei Ke¹

¹National Taiwan University

ArXiv 2025

arXiv / Code / Project Page