Room “Sala Seminari”
Abacus Building (U14)
Toward Zero-Shot Camera-to-LiDAR
Matching for Generalizable Robotic Localization
Speaker
Daniele Cattaneo
Research Group Leader at the Robot Learning Lab, University of Freiburg, Germany
Abstract
Enabling mobile robots to safely and efficiently navigate in previously unseen dynamic environments is a longstanding challenge in robotics. While advances in AI and deep learning have transformed many fields, their application to real-world robotic localization is still limited by the need for environment- and embodiment-specific training.
To bridge this gap, there is a pressing need for learning-based localization methods that can seamlessly transfer across different robots and environments.
In this talk, I will present my work on camera-to-LiDAR matching. By combining deep learning with traditional optimization techniques, the proposed method is independent of sensor-specific parameters, generalizable, and can be deployed in the wild for monocular localization in LiDAR maps and camera-LiDAR extrinsic calibration.
Evaluations on three in-house robots demonstrate that the method effectively generalizes to previously unseen environments and sensor setups in a zero-shot manner.
Short Bio
Daniele Cattaneo received the M.Sc. degree in Computer Science from the University of Milano-Bicocca in 2016, and the Ph.D. degree in Computer Science from the same university in 2020. He is currently a Research Group Leader at the Robot Learning Lab, University of Freiburg, Germany. His research focuses on advancing AI-based methodologies for robotic perception and state estimation, with an emphasis on developing models capable of evolving and adapting to different robotic platforms, unseen environments, and novel object
contact person for this Seminar: domenico.sorrenti@unimib.it