InformatiKOM, Bldg. 50.19, Atrium, Adenauerring 12, 76131 Karlsruhe

06 June 2024, 09:30

Frederike Dümbgen is a junior researcher in the WILLOW group, affiliated with Inria and the Computer Science department of École Normale Supérieure in Paris. From 2022 until April 2024, she was a postdoctoral researcher at the Robotics Institute of University of Toronto with Prof. Tim Barfoot. She received her Ph.D. in 2021 from the Laboratory of AudioVisual Communications with Prof. Martin Vetterli and Dr. Adam Scholefield in Computer Science at École Polytechnique Fédérale de Lausanne (EPFL), Switzerland. Before that, she obtained her B.Sc. and M.Sc. in Mechanical Engineering from EPFL in 2013 and 2016, respectively, with Master's thesis at the Autonomous Systems Lab of ETH Zürich. Her research has ranged from novel localization methods, using in particular acoustic, radio-frequency and ultra-wideband signals, to, most recently, global optimization for robotics.

Abstract:

Reliable state estimation is the foundation of most successful robotics applications. To solve the optimization problems commonly arising in estimation, local gradient-based solvers have become a widespread approach. However, these solvers can converge to poor local estimates that may be far from the globally optimal solution. Relying on such solutions without a verification mechanism may result in performance degradation and even catastrophic consequences.
Recent years have seen exciting developments in so-called certifiably optimal estimation, showing that many problems can in fact be solved to global optimality or certified through the use of semidefinite relaxations. In this talk, I will present our efforts to make such methods accessible for robotics. I will start by presenting a catalogue of problems for which we have developed formulations allowing for a global solution or certificate, extending from range-only localization to non-isotropic SLAM. I will then present our algorithm to automate the problem formulation process, which allows for the quick adoption of these methods to new applications. I will conclude with a discussion of our ongoing work along two other important axes of certifiable solvers: improving solver speed by exploiting sparsity, and improving solution accuracy by using differentiable programming.