HMI Lab

Embodied AI / Large Vision Language Models

Image citation: Aligning Cyber Space with Physical World: A Comprehensive Survey on Embodied AI

Embodied intelligence refers to the concept within artificial intelligence that emphasizes the importance of a body in developing intelligent systems, drawing inspiration from the way biological organisms interact with the world. Research in this field has evolved from simple robotic tasks to complex interactions involving perception, cognition, and action in dynamic environments.

The primary goal is to develop AI systems that can operate autonomously and effectively in real-world settings by integrating sensory data and physical capabilities.

In our group, we mainly focus on perception, which is the most computation expensive part.

Key research focuses:

- Perception algorithms

- Robust generalization AI algorithms

- Efficient learning algorithms

AI for Science

Image citation: Foundation models for generalist medical artificial intelligence

AI for Science (AI4Science) refers to the application of AI techniques to scientific research, emphasizing the transformation and acceleration of scientific discovery processes by leveraging AI's capabilities. Research in this area has moved from theoretical models to practical applications that enhance data analysis, simulation, and prediction across various scientific domains.

The primary goal is to develop AI systems that can autonomously process and analyze large volumes of scientific data, thereby aiding in hypothesis generation, experiment design, and the interpretation of complex results.

In our group, we focus particularly on improving the efficiency and accuracy of these processes, with a special emphasis on areas that require intensive computational resources.

Key research focuses:

- Biological systems analysis

- PDE Solver enhancements

- Efficient learning algorithms