HypeFL is a novel machine learning framework built on Hyperledger Fabric that combines machine learning and blockchain to create a decentralized, privacy-preserving fully autonomous vehicle system. Through cooperative perception, each vehicle communicates its LiDAR sensor detections to other vehicles in the network to incorporate knowledge and localize surrounding vehicles in the system, combatting visual occlusions and corner cases. Our system utilizes federated learning to optimize data privacy by only sharing model parameters between vehicles, rather than raw data. The blockchain serves as an immutable, decentralized server that stores vehicles and model parameters as nodes, providing protection against single-point failures and eliminating the risk of malicious attacks.

• Windows 10 PC with at least 165 GB of free space
• 6 GB GPU (for Unreal Engine)
• Install CARLA 0.9.12
• Install Python 3.8
• Install Node.js
• Install Go IDE

chmod +x docker.sh
sudo ./docker.sh
usermod -a -G docker ${USER}

After cloning this Github into your directory, begin by training the federated learning object detection models.

cd Federated Learning
python3 fl_train.py

python3 fl_test.py

go run main.go

Follow this link to view a demo video of the HypeFL network being used on the online CARLA simulator, as well as a physical setup of miniature Raspberry Pi-powered autonomous vehicles. This project won third place overall in the Engineering Technology: Statics and Dynamics category of the 2023 International Science and Engineering Fair (ISEF), hosted in Dallas, Texas, along with a $1,000 grant.

In addition, the project was later recognized as a spotlight project by NeurIPS 2024. The public release can be viewed here.