Gregory Dudek

This is a cross-section showing selected publications across my career. A more comprehensive and up-to-date list including on-line versions is on the publication list by topic and there is also lab list (not up to date) mobile robotics lab publications page.

One of the early applications of LLMs to robotics. We used a foundation model to create a photo-acquisition robot that could adapt to different domains by using an LLM to produce domain-specific criteria and evaluation metrics. Followup up papers sometimes refer to this as a photoobt, although in much earlier papers (without an LLM) we introduced a more general idea of a ``photo documentarian'' that would seek to document different kinds of phenomenon (final ICRA 2023 version of paper here)
Prediction of 5G traffic on edge-based compute elements by using transfer learing with a new technique we call Similarity-based Elastic Weight Consolidation. One for All: Traffic Prediction at Heterogeneous 5G Edge with Data-Efficient Transfer Learning, IEEE Globecom 2021. This was awarded a best-paper award at GlobeCom 2021.
Our paper "Generating adversarial driving scenarios in high-fidelity simulators", looks at using adversarial methods to allow autonomous driving agents to improve. The paper was an early example of the developing trend to develop more robust agents by making the environment part of the learning cycyle.
Efficient methods for drivers to meet one another or passengers. Fast and Efficient Rendezvous in Street Networks. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS'16).
Searching for novelty is video, and using it to navigate a swimming robot to maximize the stuff that is observed. Autonomous Adaptive Underwater Exploration using Online Topic Modeling with Yogesh my student and Philippe (from Laval University). This combines an automated search for novelty with the behavior of our amphibious robot. A journal version can also be found in the Intl Journal of Robotics Research in 2013.
Human-robot interaction (HRI) and dialog is a key aspect of robotics and the student author (Junead) is now professor at the University of Minnesota. This 2011 paper is an eraly step in validating HRI methods.
Graphical state space programming, from 2011-2009, (GSSP) material and executable code. A generic framework for developing light weight robot controllers.
Our first work on visual barcodes that serve as fiducial targets with variable resolution. We call them Fourier tags. A more recent paper on this looks at elaborating it with an enhanced payload and greater robustness.
With Dimitri Marinakis looked at at inferring the topology of sensor networks using very sparse event information. A more recent paper one this is our 2010 article in the IEEE Transactions on Robotics.
A preliminary paper on our older recommender system work. It's notable feature was a relatively large user-generated tag base for recommendation, a very usualy feature especailly at that time.
Work that compares PCA to landmark-based posed estimation schemes, with Rob Sim: Comparing image-based localization methods (IJCAI 2003).
Work on using visual landmarks for pose estimation. We have look at several variants of this and the following CVPR 2001 paper (postscript, gz) uses generative models.
Initial work on reconstruction of depth images from intensity data with Luz-Abril Torres Mendez, from the proc. IEEE Workshop of Applications of Computer Vision 2002 (PDF file). (Starting paper, see newer papers for better results.)
With Yiannis Rekleitis and Evangelos Milios, we looked at using multiple robots to reduce errors during mapping. This paper is Multi-Robot Collaboration for Robust Exploration (this link is for a PDF reprint from the Annals of Mathematics and Artificial Intelligence, 2001).
An ICCV 1999 paper on learning landmarks for pose estimation (gzipped postscript).
A Taxonomy for Multi-Agent Robotics. Combines a taxonomy, some new results, and a survey of existing work. A preprint of this article can be found as a PDF file (an older journal article that was retreaded and updated for the book Robot Teams: From Diversity to Polymorphism, T. Balch and L. E. Parker (Eds.), 2002).
Global localization: Localizing a Robot with Minimum Travel, with Kathleen Romanik and Sue Whitesides, SIAM J. Comput, 27, 2, April 1998, pp. 583-604. (300 KB).
Localization from sonar models, with Paul MacKenzie, compressed postscript.
Just-in-time sensing from sonar and laser data, ICRA 1996. with Rekleitis and Freedman.
Work on the classification and use of sonar landmarks, with Simon Lacroix who did a postdoc at McGill, IROS 1997 (postscript).
Computing ``shape texture'' for recognition (compressed postscript).
2-D shape-from-darkness and how hidden surfaces can be recovered. Langer, Dudek, Zucker, IROS 1995 (postcript).
Multi-robot exploration, with Rekleitis and Milios, ICRA 1996 (compressed postscript).
3-D shape-from-darkness using iterative ``carving'' and arbitrary light source directions. Daum and Dudek, CVPR 1998.

See lots more at https://www.cim.mcgill.ca/~dudek/publications_by_topic.html and also the (not up to date) mobile robotics lab publications page (www.cim.mcgill.ca/~mrl).

Professor G. Dudek

Research areas:

Address: