Embedded Systems

We develop trustworthy, efficient, and deployable AI systems for edge environments by combining collaborative AI methods with next-generation hardware acceleration. With expertise spanning the entire stack—from software and algorithms to hardware architecture and chip design—the unit pursues an end-to-end co-design approach to edge intelligence.

Our research integrates explainable and verifiable distributed AI, privacy-preserving learning, and hybrid accelerators into the RISC-V hardware/software ecosystem. In this way, we create AI systems that meet requirements for energy efficiency, scalability, security, and suitability for safety-critical and regulated sectors.

The “Virtual Perception & Learning” unit develops next-generation computer-aided methods that combine physical modeling with state-of-the-art machine learning. Our research focuses on enabling reliable, interpretable, and data-efficient characterization of complex technical systems by integrating sparse measurement data, domain expertise, and physically-guided inferences. Our goal is to advance virtual modeling beyond traditional statistical methods and to create physics-based models that remain robust even under data-scarce or rapidly changing conditions.

Our team combines expertise in computational physics and simulation methods, data science, and AI model design. We work at the intersection of theory and application, developing algorithms while collaborating with partners in research and industry to validate our methods in real-world environments.