Description

V2X (Vehicle-to-Everything) communication enhances road safety and traffic efficiency by enabling vehicles to exchange real-time information with each other, infrastructure, pedestrians, and networks. This connectivity allows for advanced warning of potential hazards, improved traffic flow, and better integration of autonomous driving technologies. Currently, it remains to investigate the potential of continuous interaction in V2X, and one such example is to train machine learning models among a herd of vehicles via V2X. 

The increasing computational power of vehicles make them perfect for participating in training models together with others, through paradigms such as Federated Learning, so that they can contribute to training models while keeping data private [1]. 

However, one prominent feature of vehicles, compared to servers in a datacenter, is that they are highly dynamic, and this leads to a series of challenges, e.g. unstable connection to road side units (RSU) and other vehicles. A decentralized learning approach via V2X is thus suitable in this scenario [2]. Currently it remains to investigate a robust decentralized learning approach with high-performance in such high dynamism [3].  

You will contribute to a proof-of-concept of the described approach. You will:

• familiarize with the state-of-the-art of decentralized deep learning and V2X communication;
• implement one or more decentralized learning algorithms in simulators such as SUMO or CARLA;
• design, execute and validate an experiment to show the usefulness of the proposed method.

This project requires you to be proficient in python programming and deep learning frameworks, e.g. TensorFlow or PyTorch, and have basic knowledge about V2X and simulators such as SUMO and CARLA. Experience with reinforcement learning and computer vision is advantageous.

References

1. Chellapandi, Vishnu Pandi, et al. "Federated learning for connected and automated vehicles: A survey of existing approaches and challenges." IEEE Transactions on Intelligent Vehicles (2023).
2. Yuan, L., Wang, Z., Sun, L., Philip, S. Y., & Brinton, C. G. (2024). Decentralized federated learning: A survey and perspective. IEEE Internet of Things Journal.
3. Barbieri, L., Savazzi, S., Brambilla, M., & Nicoli, M. (2022). Decentralized federated learning for extended sensing in 6G connected vehicles. Vehicular Communications, 33, 100396.