Hu Wan

My research lies at the intersection of AI-system co-design, memory and storage systems, machine learning infrastructure, and embodied intelligence. I have extensive experience in storage-efficient architectures and system optimization for large-scale recommendation and deep learning workloads, including non-volatile memory, in-storage computing, SSD-based embedding management, and tail-latency-aware scheduling. My current research focuses on high-performance edge intelligence and self-evolving robotic systems. In particular, I investigate the integration of foundation models (e.g., LLMs and VLA models) with reinforcement learning to enable adaptive and self-improving robotic manipulation, while optimizing their deployment on resource-constrained edge platforms.