Physical AI & Sim-to-Real

Physical AI & Sim-to-Real

AI systems that perceive and act in the physical world, using physics-based simulation to generate training data and bridge the gap to real environments.

• Physics-based simulation and synthetic data generation
• Sim-to-real gap analysis
• Embodied control for robotic manipulators on embedded and neuromorphic hardware

Research Area Overview

We build AI systems that perceive and act in the physical world, using physics-based simulation to generate training data and closing the gap between simulated and real environments. Our work spans patient fall detection for smart hospital beds and energy-efficient control of robotic manipulators, with an emphasis on models light enough to run on embedded and neuromorphic hardware.

Focus Topics

• Physics-based simulation and synthetic data generation for data-scarce physical domains
• Sim-to-real gap analysis between simulated dynamics and real-world sensing
• Embodied control for robotic manipulators on embedded and neuromorphic hardware
• Lightweight, real-time models deployable on embedded clinical and robotic edge devices