This repository contains the code and resources used in our Master's thesis project titled, "Towards Federated Fleet Learning Using Unannotated Data". The project was carried out at Zenseact & AI Sweden under the expert guidance of Johan Östman, PhD, and Mina Alibeigi, PhD.
Our research ventured into the burgeoning field of Federated Learning (FL) within the context of autonomous driving. We confronted the significant challenge of working with scarce or non-existent labelled data, representing a stark departure from previous works that predominantly relied on plentiful labelled data.
Our approach leveraged semi-supervised learning for ego-road segmentation and imitation learning for trajectory prediction. We demonstrated the potential of FL in autonomous driving when utilising on-vehicle generated labels or when employing semi-supervised or unsupervised learning methods. This insight underscores the promise of FL as a robust learning methodology in autonomous driving, especially given the paucity of labelled data.
This repository is a testament to our belief in open research and collaboration, as it allows other enthusiasts and professionals to learn from and build upon our work.
To learn more about our project, we recommend reading our blog post and our Master's thesis. The blog post provides an accessible overview of our research, while the thesis offers a comprehensive look at our methodology, results, and implications for the field.
Blog post: https://my.ai.se/projects/259
Thesis (local): thesis.pdf
Thesis (online @ Chalmers University of Technology): [to-be-added]
Final presentation slide deck: presentation-slide-deck.pdf