Federated Learning-Based Driving Strategies Optimization for Intelligent Connected Vehicles

Thanks to smart manufacturing and artificial intelligence technologies, intelligent connected vehicles (ICVs) is emerged as a main transportation means. However, due to the limitations of finiteness and privacy of driving data, ICVs may not be able to share their data with other vehicles, which limits the development of ICVs. To overcome aforementioned challenges, we propose a federated learning-based driving strategy optimization scheme for ICVs. Conditional imitation learning is employed to obtain a single-vehicle intelligent driving strategy. To improve the driving ability while ensuring data privacy, federated learning is leveraged to aggregate driving policies of different ICVs. Finally, the experimental results based on the Carla platform show that the single-vehicle intelligent driving strategy achieves a high level of accuracy, and the federated learning vehicle model achieves a significant 15% increase in the success rate of turning tasks and a 21% increase in the success rate of going straight, which verifies the effectiveness of the method in this paper.