When: 26th of February, 1:00pm AEDT
Where: This seminar will be partially presented at the ACFR seminar area, J04 lvl 2 (Rose St Building) and partially online via Zoom. RSVP
Speaker: Dr. Joshua Knights
Title: Bridging Domain Gaps in 3D Scene Understanding
Abstract:
3D scene understanding is a pivotal research area in the pursuit of autonomous or semi-autonomous robotics, with recent state-of-the-art approaches for key tasks such as 3D place recognition, metric localisation and semantic segmentation dominated by deep learning-based approaches. However deep learning-based approaches often suffer from poor generalisation: degradation in performance on unseen data when there is a significant distribution shift, or domain gap, relative to the data used to originally train the network. This is a significant challenge for deploying these approaches on robots which will often be deployed into novel environments, environments which change dynamically over time, or onto robotic platforms with different sensor configurations. In this seminar I will be providing an overview of the key challenges in 3D domain generalisation, the key contributions from my doctoral thesis in addressing these challenges, and interesting future research directions for this field of study.
Bio:
Dr Joshua Knights received his PhD in Computer Science from the Signal Processing, Artificial Intelligence and Vision Technologies (SAIVT) lab at the Queensland University of Technology (QUT) in December 2025, where he also collaborated with DATA61 at the CSIRO. His research experience spans Artificial Intelligence, Computer Vision, Continual Learning and Self-Supervised Domain Adaptation. Joshua recently joined the Australian Centre for Robotics (ACFR) at the University of Sydney as a Postdoctoral Research Associate in 3D Deep Learning for Vision and Lidar.