Localisation and Denoising Modules for Autonomous Systems
Jun.-Prof. Dr. Vasileios Belagiannis
Institute of Measurement, Control and Microtechnology Ulm University
DEIB - Conference Room "E. Gatti" (building 20)
February 14th, 2020
2.00 pm
Contacts:
Matteo Corno
Research Line:
Control systems
Institute of Measurement, Control and Microtechnology Ulm University
DEIB - Conference Room "E. Gatti" (building 20)
February 14th, 2020
2.00 pm
Contacts:
Matteo Corno
Research Line:
Control systems
Sommario
We formulate the process of autonomous driving as a composition of differentiable modules where deep neural networks are the design basis. In this talk, we examine the modules of localisation and denoising. For the localisation module, the goal is to determine the agent’s pose based on landmark measurements and map landmarks. Based on multi-modal sensory information, the map is built by extracting landmarks from the vehicle's field of view in an off-line way, while the measurements are collected in the same way during inference. To map the measurements and map landmarks to the vehicle's pose, we propose an approach that copes with dynamic input.
On denoising, our focus is on sequential and image-based data that is corrupted by sensor noise. We study the problem as few-shot learning-based denoising where the training set contains just a handful of clean and noisy samples. We propose a new meta-learning training approach for few-shot learning-based denoising problems. Our model is meta-trained using known synthetic noise models, and then fine-tuned with the small training set, with the real noise, as a few-shot learning task. Both localisation and denoising modules demonstrate promising results on standard benchmarks.
On denoising, our focus is on sequential and image-based data that is corrupted by sensor noise. We study the problem as few-shot learning-based denoising where the training set contains just a handful of clean and noisy samples. We propose a new meta-learning training approach for few-shot learning-based denoising problems. Our model is meta-trained using known synthetic noise models, and then fine-tuned with the small training set, with the real noise, as a few-shot learning task. Both localisation and denoising modules demonstrate promising results on standard benchmarks.