Overview
Background
Peyman Moghadam is an Adjunct Associate Professor at the University of Queensland (UQ). He is a Principal Research Scientist at CSIRO Data61 as well as Professor (Adjunct) at the Queensland University of Technology (QUT). He leads the Embodied AI Research Cluster at CSIRO Data61, working at the intersection of Robotics and Machine learning. He is also the Spatiotemporal AI portfolio Leader at the CSIRO's Machine Learning and Artificial Intelligence (MLAI) Future Science Platform and oversees research and development of MLAI methods for scientific discovery in spatiotemporal data streams. In 2022, he served as a Visiting Professor at ETH Zürich. In 2019, he held a Visiting Scientist appointment at the University of Bonn. Peyman has led several large-scale multidisciplinary projects and won numerous awards, including CSIRO's Julius Career Award, National, and Queensland state iAward for Research and Development, CSIRO’s Collaboration Medal and the Lord Mayor’s Budding Entrepreneurs Award. His current research interests include self-supervised learning for robotics, embodied AI, 3D multi-modal perception (3D++), robotics, and computer vision.
Availability
- Dr Peyman Moghadam is:
- Available for supervision
Fields of research
Research interests
-
Embodied Intelligence; Self-Supervised Learning; spatiotemporal learning
-
Robotics, Computer Vision, Machine Learning, Deep Learning
-
Beyond visible Spectrum Perception (Hyperspectral, Thermal)
-
3D LiDAR SLAM; 3D Scene understanding; 3D Segmentation
Works
Search Professor Peyman Moghadam’s works on UQ eSpace
2009
Conference Publication
Online, Self-Supervised Vision-Based Terrain Classification in Unstructured Environments
Moghadam, Peyman and Wijesoma, Wijerupage Sardha (2009). Online, Self-Supervised Vision-Based Terrain Classification in Unstructured Environments. IEEE International Conference on Systems, Man and Cybernetics, San Antonio Tx, Oct 11-14, 2009. NEW YORK: IEEE. doi: 10.1109/ICSMC.2009.5345942
2008
Conference Publication
Improving Path Planning and Mapping Based on Stereo Vision and Lidar
Moghadam, Peyman, Wijesorna, Wijerupage Sardha and Feng, Dong Jun (2008). Improving Path Planning and Mapping Based on Stereo Vision and Lidar. 10th International Conference on Control, Automation, Robotics and Vision, Hanoi Vietnam, Dec 17-20, 2008. NEW YORK: IEEE. doi: 10.1109/ICARCV.2008.4795550
Supervision
Availability
- Dr Peyman Moghadam is:
- Available for supervision
Before you email them, read our advice on how to contact a supervisor.
Available projects
-
Self-Supervised Learning for 3D Multimodal Perception
Potential impact of deep learning is limited due to the lack of large, annotated, and high-quality datasets in domains of interest. Annotating such datasets is laborious, costly and time-consuming. This project proposes to develop self-supervised learning systems to extract and use the relevant context given by strong prior spatio-temporal models (e.g. dense 3D reconstructions) as supervisory signals in training. This new concept will investigate model structures that encodes spatio-temporal data, and show rapid adaptation of models to new domains (few-shot learning) using trained embeddings layers (self-supervised, or prior data).
-
3D Scene Understanding
Simultaneous Localization and Mapping (SLAM) is a key enabling component of driverless vehicles, robotics and augmented reality. The SLAM goal is to estimate pose of the vehicle and simultaneously generate dense 3D scene reconstruction. At CSIRO we have developed and deployed state-of-the-art 3D LiDAR-based SLAM systems for the past decade. There is a new direction of research at the intersection of deep learning and geometry-based 3D SLAM. The research in this PhD programme will develop algorithms for geometry-based Deep Learning SLAM in a dynamic and unstructured environment. The PhD programme will involve the development of self or semi-supervised learning methods to address the significant weakness of most current deep networks.
-
Hyperspectral Deep Learning
Hyperspectral cameras are currently undergoing a change from bulky and expensive equipment towards mobile and portable devices. A hyperspectral camera comprises of hundreds of bands with shortwave dependencies. Compared to conventional colour cameras (RGB bands), one could use these shortwave dependencies to design and develop a deep network for object classification, semantic segmentation and scene understanding. Both spectral and spatial relationship needs to be modelled by the deep networks simultaneously. The research in this PhD programme will develop algorithms for hyperspectral deep learning. The PhD programme will involve the development of learning with self-supervision algorithms to address the significant weakness of most current deep networks.
Supervision history
Current supervision
-
Doctor Philosophy
Generalizing Implicit Representations for Robotics Manipulation of Articulated Objects
Associate Advisor
Other advisors: Associate Professor Mahsa Baktashmotlagh
Media
Enquiries
For media enquiries about Dr Peyman Moghadam's areas of expertise, story ideas and help finding experts, contact our Media team: