Xu Chenren
Boya Distinguished Associate Professor (Tenured Track), Peking University
Xu Chenren is a Boya Distinguished Associate Professor (tenured track) at Peking University, Assistant Dean of the School of Computer Science, and Deputy Director of the Institute of Network and High-Efficiency Computing. He received his Ph.D. from Rutgers University in 2014. He previously served as a postdoctoral researcher and research scientist at Carnegie Mellon University, a visiting professor at Keio University in Japan, and a visiting scholar at AT&T Shannon Lab and Microsoft Research. His research focuses on super-perceptual intelligence (embodied intelligent IoT and multimodal-driven scientific and engineering intelligence). He has led projects including the National Science Fund for Excellent Young Scholars, key programs of the National Natural Science Foundation of China, and major national R&D initiatives from the Ministry of Science and Technology. His work on high-performance perception and interconnected systems for mobile IoT has received the CCF First Prize for Technological Invention (2025, ranked 1st), six best paper–type awards at MobiCom, MobiSys, SenSys, and UbiComp, the DAMO Academy Young Fellow Award, Huawei Top Ten Inventions Award, the Okawa Foundation Research Award (Japan), and a gold medal at the Geneva International Exhibition of Inventions. His technologies have been deployed at scale in systems such as Huawei HarmonyOS devices and Alibaba logistics platforms. He currently serves as Associate Editor of ACM IMWUT, Editorial Board Member of Nature Partner Journal Artificial Intelligence, General Chair of UbiComp 2026, Program Chair of UbiComp 2025, Steering Committee Member of UbiComp, General Chair of MobiSys 2025, and has served on the program and organizing committees of SIGCOMM, MobiCom, and NSDI. He is also the Outreach Ambassador of ACM SIGDX (Digital Transformation), Secretary-General of ACM SIGBED China Chapter, and a board member of the Carnegie Mellon University Beijing Alumni Association.
Topic
Breaking the “Visual Blind Spots”: How Multimodal Super-Perception Is Reshaping the Next Generation of Embodied Intelligence
As multimodal embodied intelligence continues to expand into real-world environments, the demand for advanced scene understanding—such as generalization across diverse environments, implicit information perception, and object property detection—keeps increasing. This talk approaches the topic from the perspective of “super-visual intelligent perception,” aiming to overcome the inherent limitations of traditional vision under conditions such as low light, occlusion, uniform materials, and invisible structures. By leveraging penetrative, multi-physics modalities—including electrical, acoustic, magnetic, and radio-frequency signals—it systematically presents recent progress in multimodal super-perception across tasks such as warehouse management, biological behavior tracking, liquid identification, and agent interaction, and discusses how these advances empower the next generation of embodied intelligence. The talk will also address the challenges of insufficient scale and quality in multimodal data. It proposes a data engine tailored for super-visual scenarios to expand coverage across both real and synthetic data, and explores the potential of cross-modal synthesis, weakly supervised learning, and Real2Sim2Real paradigms in building a unified data foundation. Finally, the report presents a preliminary exploration of an end-to-end, full-stack computational architecture for embodied real-time video interaction scenarios, designed to meet the latency and performance requirements of complex multimodal data processing.
