Towards Spatially Intelligent and Embodied AI Systems: From Perception to World Models

Abstract: 
AI is rapidly advancing from task-specific perception towards foundation models which are capable of understanding and generating multimodal content. However, intelligent interaction with the physical world requires more than recognising objects or generating realistic images. In this end, AI systems need to understand what is happening, where it is happening, how the environment evolves, and how different entities interact within it. This talk presents three recent works that collectively illustrate the trajectory towards spatially intelligent and embodied AI systems. The first project introduces an LLM-driven framework for training-free referring video object segmentation, which moves from multimodal perception towards reasoning-aware visual understanding, leveraging LLM-driven hierarchical reasoning to identify and segment objects referred to by natural language in dynamic video scenes. 
Secondly, the project of 3D scene-adaptive trajectory-controllable human image animation with camera movement advances from understanding visual content towards spatially and geometrically controllable generation, incorporating 3D scene information, human trajectories and camera movement to enable more structured and controllable visual generation.
Finally, a framework explores lightweight audio-visual segmentation for efficient multimodal perception. By integrating complementary visual and audio information, the work demonstrates how AI systems can move beyond single-modality perception while maintaining computational efficiency.

Short bio: 
Dr Lin Yuanbo Wu is an assistant professor/SL with University of Warwick. She received her PhD from The University of New South Wales, Kensington, Sydney, Australia. Dr Wu is specialized in multi-modal learning, spatial reasoning, embodied visual intelligence, generative AI, and AI-agentic advancement. She has intensively published more than 100 peer-reviewed research articles in premier journals and proceedings. She was a recipient of the 2026 CVPR Transparency Champion Award, Outstanding Nomination Paper Award in the 7th International Conference on Pattern Recognition and Artificial Intelligence 2024, the Award for Growth of 2021 The 4th Eureka International Innovation and Entrepreneurship Competition (Eureka IIEC 2021, Australia).  Dr Wu is currently serving as the Associate Editor with IEEE Trans on Image Processing, IEEE Trans on Circuits and Systems for Video Technology, IEEE Trans on Neural Networks and Learning Systems, IEEE Trans on Emerging Topics in Computational Intelligence, IEEE Trans on Cybernetics, IEEE Trans on Multimedia, IEEE Trans on Big Data, Pattern Recognition. She was invited to serve as Area Chair with AAAI 2027/2026, International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2026/2025), British Machine Vision Conference 2025/2024, ACM Multimedia 2026-2022. She is a Senior Member of IEEE.