Effective 3D/4D Representations and Spatial Intelligence Built upon Them

PhD Thesis Proposal Defence


Title: "Effective 3D/4D Representations and Spatial Intelligence Built upon Them"

by

Mr. Xinhang LIU


Abstract:

This thesis studies how to endow AI systems with persistent spatial and physical
understanding through effective 3D/4D representations and the spatial
intelligence built upon them. While modern foundation models excel at
appearance and semantics, they lack explicit representations of geometry,
motion, and long-term correspondence, which are essential for reasoning and
action in physical environments. This dissertation proposes universal 4D world
representations that model structure and dynamics as persistent entities rather
than frame-dependent observations. Building on these representations, it
investigates spatially grounded intelligence that enables consistent
perception, correspondence-aware reasoning, and long-horizon spatial
understanding. Together, this work establishes a unified framework in which
explicit 3D/4D representations serve as the foundation for scalable spatial
intelligence, bridging perception, reasoning, and interaction in dynamic
physical worlds.


Date:                   Wednesday, 25 February 2026

Time:                   2:00pm - 4:00pm

Venue:                  Room 2132C
                        Lift 22

Committee Members:      Prof. Chi-Keung Tang (Supervisor)
                        Dr. Dan Xu (Chairperson)
                        Prof. Dit-Yan Yeung