What happens when an AI tells a robot to take an action that looks reasonable, but is physically impossible, unsafe, or simply fails?

Harold will introduce the idea of action hallucination to explore what it takes to ground intelligence in the physical world, where actions are constrained by physics, often require high precision, and need to remain reliable across complex tasks.

He'll explore why simply scaling data or compute may not solve these challenges, and how robots can instead use tactile feedback to sense and correct errors through interaction.

He'll also discuss how making the assumptions behind learned skills more explicit can help agentic systems reason about when and where those skills should be used, and how they can be combined to tackle longer tasks.

More About the Speaker

Harold Soh is an Associate Professor of Computer Science at the National University of Singapore, where he leads the Collaborative Learning and Adaptive Robots (CLeAR) lab.

His research focuses on trustworthy embodied AI: reliable, fluent, and safe robots that people can depend on. His work connects understanding human trust with building dependable robots, spanning mathematical foundations, learning algorithms, and physical systems.

His recent research explores touch perception for contact-rich interactions and robot policies that can be guided by human goals and constraints.

He was awarded an NRF Investigatorship in 2026 and a Robotics: Science and Systems Early Career Spotlight in 2023. His work has been recognized with best paper awards and nominations at IROS’21, IEEE TAFFC’21, RSS’18, HRI’18, RecSys’18, and IROS’12. He is also a co-founder of TacnIQ, a startup developing touch-enabled intelligence. Harold obtained his PhD from Imperial College London in 2013, with a thesis on online learning for assistive robotics.

He lives in Singapore with his wife and two children, and enjoys reading science fiction with a glass of wine.

More About the Series

Singapore Embodied AI Colloquium (SEAIC) is a focused technical research series built around researcher-driven discussions and sharing on frontier topics in embodied AI, robot learning, and physical AI.

Get involved: Learn more about Lorong AI | Speaker Sign-up | WhatsApp Community | LinkedIn | X