Yue Cao's research interests lie at the intersection of intelligent manufacturing, robotics, and physical process modeling, with a focus on enabling autonomous robotic control for complex arc-based manufacturing processes such as welding and wire arc additive manufacturing (WAAM). His work addresses two fundamental challenges: how to endow robots with physical cognition of the manufacturing process, and how to transfer skilled human operators' adaptive expertise to autonomous systems.
On the perception and modeling side, his research spans low-cost electrical signal sensing for penetration control, generative deep learning for cross-process state representation and anomaly detection, and physics-informed neural networks for reconstructing otherwise unmeasurable weld pool flow fields from sparse multi-modal observations. On the control side, he develops virtual-reality-driven human-robot collaborative platforms for systematic demonstration data acquisition, and imitation learning frameworks that extract operators' implicit decision-making criteria for robotic deployment, combined with online optimal control for real-time adaptation beyond the limitations of demonstration data.
His current research aims to unify these threads into a physics-cognition-driven autonomous control architecture, where robots make decisions based on reconstructed physical states rather than raw sensor signals, achieving both physical interpretability and cross-condition generalizability in high-value
Yue Cao's research interests lie at the intersection of intelligent manufacturing, robotics, and physical process modeling, with a focus on enabling autonomous robotic control for complex arc-based manufacturing processes such as welding and wire arc additive manufacturing (WAAM). His work addresses two fundamental challenges: how to endow robots with physical cognition of the manufacturing process, and how to transfer skilled human operators' adaptive expertise to autonomous systems.
On the perception and modeling side, his research spans low-cost electrical signal sensing for penetration control, generative deep learning for cross-process state representation and anomaly detection, and physics-informed neural networks for reconstructing otherwise unmeasurable weld pool flow fields from sparse multi-modal observations. On the control side, he develops virtual-reality-driven human-robot collaborative platforms for systematic demonstration data acquisition, and imitation learning frameworks that extract operators' implicit decision-making criteria for robotic deployment, combined with online optimal control for real-time adaptation beyond the limitations of demonstration data.
His current research aims to unify these threads into a physics-cognition-driven autonomous control architecture, where robots make decisions based on reconstructed physical states rather than raw sensor signals, achieving both physical interpretability and cross-condition generalizability in high-value