I am currently pursuing my Ph.D. in the Machine Learning Research Lab at Volkswagen Group and TUM, under the guidance of Prof. Patrick van der Smagt, Dr. Philip Becker-Ehmck and Dr. Maximilian Karl. Prior to joining MLRL, I obtained my bachelor’s and master’s degrees in Mechatronics Engineering from Harbin Institute of Technology. During my studies, I focused on computer vision and worked on projects related to hand pose estimation, generative models, and more. Later on, I became interested in model-based reinforcement learning after reading the World Model paper. I then worked as a Machine Learning Researcher at Polixir Technology, an AI startup in Nanjing, where I primarily focused on offline reinforcement learning for real-world applications.

As an AI enthusiast, I am interested in a broad range of topics within the field. My current research is focused on incorporating data-driven knowledge, such as pretrained models, into the world model to enhance its sample efficiency and generalization ability to out-of-distribution data. If you would like to discuss these topics or any other related topics, please feel free to contact me at xingyuan.zhang@argmax.ai.

I am also taking master students. If you are interested in writing your master thesis with me, feel free to drop me an e-mail.

Research Interests

• World Models
• Reinforcement Learning
• Transfer Learning
• Foundation Models

News

[2023-09-22] Our paper “Action Inference by Maximising Evidence: Zero-Shot Imitation from Observation with World Models” has been accepted at NeurIPS 2023. Code, datasets and pretrained models are available here. An additional blog post give a nice intro about the paper.