Analyzing and improving generalization of vision language action models

• Applicant: Dr.-Ing. Max Argus
• Subject Area: Methods in Artificial Intelligence and Machine Learning ; Automation, Mechatronics, Control Systems, Intelligent Technical Systems, Robotics
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 559307072

Large language models are extremely popular because they are able to perform a wide variety of tasks out-of-the box, or for more unusual tasks, with only a few in-context examples. In the future, we would want robotic models to be similarly intuitive and easy to use, and vision action language (VLA) models are a current area of robotics research to achieve this and our high-level goal of this project is to improve them. There are three objectives that we want to pursue in order to reach this goal: improving the generalization to new robots, improving the interpretation of vision language action models, and training VLA models for imitation from demonstration videos.

• DFG Programme: WBP Fellowship
• International Connection: USA
• Host: Professorin Jeannette Bohg, Ph.D.