|
|
|
|
Physics-Driven Continuous and Differentiable Representations for Perception, Planning, and Learning
|
**RSVP by Tuesday 16 June for catering purposes.
Accept this calendar invitation if you plan to attend the seminar in person and join us for a light lunch at approximately 12.30pm.
|
|
|
|
|
In this talk, I will first present our work on continuous distance fields for localisation, mapping, safe control, planning,
rendering, grasping, human-robot interaction and multi-robot coordination using models that aim to approximate the solution of the regularised Eikonal equation. I will then turn to our work on continuous magnetic vector fields for active perception, localisation,
and mapping, which rely on a divergence-free kernel. Finally, I will present our recent work on differentiable skill representations for learning by demonstration through deep reinforcement learning. Simulations and experimental results will demonstrate the
performance of these representations.
|
|
Teresa Vidal-Calleja is Professor and Research Director at the Robotics Institute at UTS. Her research spans robotics perception,
unconventional sensing, SLAM, continuous mapping, and autonomous navigation, with applications across manufacturing, space, defence, construction, and logistics. She currently leads an ARC Discovery grant on flow field robotic perception and co-leads Program
1 of the Australian Robotics Centre. She is an IEEE Senior Member, Senior Editor for IROS, and Associate Editor of IEEE Transactions on Robotics and IEEE Transactions on Field Robotics. She obtained her PhD in automatic control, computer vision, and robotics
from the Technical University of Catalonia, and was a postdoctoral fellow at LAAS-CNRS and the ACFR before joining UTS. She has been a visiting academic at ETH Zürich, the German Aerospace Centre, and the Tokyo University of Science.
|
|
|
|
|
|
