AI, ML and Friends is a weekly seminar series within the School of Computing on Artificial Intelligence, Machine Learning, and related topics. We are open to attendees and presenters external to the school. Please sign up to the mailing list to receive weekly announcements including zoom details, and email the seminar organiser to schedule a talk.
Upcoming Seminars #
09 December 2025, 10:00 #
User-centric Assessment of Robotic Systems via POMDP Planning of Adversaries #
Speaker: Ivan Ang
Abstract: As autonomous systems continue to proliferate across industrial, commercial, and domestic settings, users face the challenge of choosing among many robotic systems that differ in safety, effectiveness, and robustness. While research has largely focused on improving controllers and developer-side safety guarantees, far less work provides end-users with transparent tools to compare these systems. This work proposes a general, black-box assessment mechanism that evaluates and ranks robotic systems using only observable behavior, enabling comparison across user-defined criteria. The proposed framework evaluates systems according to user-defined criteria by identifying the minimal adversarial conditions that induce failure, yielding a unified and interpretable measure of performance: the distance to failure. To probe system limits, the mechanism generates adversarial trajectories to test safety and adversarial environments to test effectiveness and robustness. The closer these adversarial conditions are to nominal behavior, the narrower the system’s safety margins and the lower its resilience to uncertainty, disturbances, and unexpected interactions. To compute these minimally adversarial conditions efficiently under partial observability, we introduce Constraint Aware Tree (CAT), an online planner that reformulates the multi-objective problem as a series of related Constrained POMDPs and solves them jointly. Experiments across autonomous driving, robotic guide dog, and high-DOF manipulation—validated in simulation and on hardware—show that the mechanism reliably distinguishes systems and controller configurations. This provides users with a practical, interpretable tool for selecting safe and robust robotic systems.
Bio: Ivan is a final-year PhD student in the Robotics Group at the ANU School of Computing. His research interests include decision-making under uncertainty, robot motion planning, and assurance of autonomous systems.
Where: Building 145, room 1.33
12 December 2025, 14:00 #
From Perception to Action: Multimodal Embodied AI #
Speaker: Prof. Winston Hsu
Abstract: In this talk, we will showcase a series of research works bridging ‘perception’ and ‘action,’ demonstrating advanced scenarios supporting robotics and automotive applications while incorporating human-in-the-loop methodologies. We will envision future possibilities for multimodal embodied AI and emphasize the importance of efficiency and sustainability in cloud-edge-hybrid inference systems.
Bio: Dr. Winston Hsu is a Professor at National Taiwan University and an active researcher in computer vision, robot learning, and machine intelligence. From 2021 to 2024, he served as CTO and Vice President of MobileDrive, a joint venture between Foxconn and Stellantis, where he led the development of HPC-based solutions for digital cockpits and ADAS. Dr. Hsu received his Ph.D. from Columbia University in 2007. He is the Founding Director of the NVIDIA AI Lab at NTU—the first NVIDIA AI Lab in Asia—and has been a strategic AI advisor to several leading ICT companies in Taiwan. His prior research appointments include visiting scientist roles at Microsoft Research Redmond (2014) and IBM T.J. Watson Research Center (2016). In addition to academia, Dr. Hsu co-founded thingnario, a global AI/IoT company providing renewable energy SaaS solutions, and ARDGE, a systems startup focused on on-premise knowledge management. Earlier, he was a founding engineer and R&D manager at CyberLink Corp. He served as the Associate Editor for IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) and IEEE Transactions on Multimedia, two premier journals, and was in the Editorial Board for IEEE Multimedia Magazine (2010 - 2017).
Where: Building 145, room 3.41