What Should a Robot Remember? Rethinking Scene Representation for Embodied AI

The success of future embodied AI systems will depend not only on how well robots perceive and act in the world, but also on what they choose to remember about it. A robot operating for months or years must build internal memories that capture geometry, semantics, dynamics, and experience, while remaining scalable, adaptable, and useful for decision-making. This talk explores scene representation as a foundational problem in robotics and embodied AI. Drawing on his recent work at MIT and Perceptica Robotics at NUS, Rajat Talak will discuss new approaches that enable robots to construct richer representations of their environments and use them to support long-term autonomy. Through studies spanning mapping, scene understanding, and world modeling, he will illustrate how advances in representation are helping robots move closer to human-like understanding of their surroundings. Rajat will conclude with a perspective on why this challenge matters for Singapore's robotics ecosystem and what opportunities lie ahead for research, industry, and public impact. (Technical Level: 200) More About the Speaker Rajat Talak is an Assistant Professor in the Department of Electrical and Computer Engineering at the National University of Singapore (NUS) and Founder and Principal Investigator of the Perceptica Robotics Lab. He received his PhD from the Massachusetts Institute of Technology (MIT) and subsequently served as a Postdoctoral Associate and Research Scientist at MIT, working at the intersection of robotics, embodied AI, and autonomous systems. His research seeks to enable robots to build rich representations of the world that support robust operation over extended periods in complex, real-world environments. His work spans robot perception, mapping, scene representation, optimization, and learning, with applications ranging from autonomous robots to next-generation embodied AI systems. Rajat serves in leadership roles across the robotics community, including as Associate Editor of IEEE Robotics and Automation Letters (RA-L), Associate Editor for ICRA 2026, and Area Chair for RSS 2026, CoRL 2026. He has also organized workshops at leading venues including RSS, CVPR, and IROS, with a focus on long-term autonomy, safe autonomy, and space robotics. 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