Physical AI Challenges: From Perceiving the Physical World to Trustworthy Action
Dr Yuqian Lu
University of Auckland, New Zealand
Abstract
Physical artificial intelligence – systems that must perceive, reason, and act within the physical world – remains markedly less mature than its digital counterpart. Models that achieve strong results on standard benchmarks degrade substantially under the occlusion, variability, and real-time constraints of engineering environments. This tutorial organises the resulting research landscape around five coupled scientific challenges that together form the operating loop of an embodied agent: perception and scene comprehension, skill learning from limited interaction, trustworthy action under uncertainty, adaptation under disruption and distribution shift, and verification of increasingly agentic, multi-agent autonomy. Drawing on work from the Industrial AI Research Group at the University of Auckland, the talk presents evidence of progress against the first four and identifies the verification of autonomous behaviour as an open frontier.
Biography
Dr Yuqian Lu is a Senior Lecturer and Director of the Industrial AI Research Group in the Department of Mechanical and Mechatronics Engineering at the University of Auckland. He received the B.E. (Hons.) degree in mechatronics engineering from Dalian University of Technology, Dalian, China, in 2012, and the Ph.D. degree in mechatronics engineering from the University of Auckland, Auckland, New Zealand, in 2017. His research focuses on industrial AI, spanning assembly scene understanding, human–robot collaboration, robot skill learning, and factory automation. He has authored more than 130 peer-reviewed publications, which have attracted over 16000 citations. Dr. Lu is a recipient of the University of Auckland Early Career Research Excellence Award, and serves as an Associate Editor for the IEEE Transactions on Systems, Man, and Cybernetics: Systems and the IEEE Transactions on Automation Science and Engineering.