Adversarial Sensor Attacks on LiDAR-based Cooperative Perception in Autonomous Driving Environments

Project Objective

The cooperative perception in autonomous driving utilizes the Lidar prediction results generated by multiple connected vehicles to enhance the prediction accuracy. However, the cooperative perception system could be compromised by the fake prediction results conducted by the attacker. To address this issue, this work aims to study the security of LiDAR-based cooperative perception in autonomous driving. We will design methods to generate the adversarial samples to fool the cooperative perception system to predict the wrong results. Meanwhile, the defense strategy will be proposed and the simulation will be conducted to evaluate the attacking method and defense work.

Development Tools Tutorials

• Pytorch tutorial: ​https://pytorch.org/tutorials/
• Tensorflow tutorial: ​https://www.tensorflow.org/tutorials

Machine Learning Models for Autonomous Vehicles

• SECOND for KITTI/NuScenes object detection example: ​https://github.com/traveller59/second.pytorch
• Baidu Apollo: ​https://github.com/ApolloAuto/apollo
• The open-source code of Paper[2]: ​https://github.com/Aug583/F-COOPER
• Example of a physical adversary attack on self-driving models: ​https://github.com/xz-group/AdverseDrive

Reading Material

• *​Cooper: Cooperative Perception for Connected Autonomous Vehicles based on 3D Point Clouds. (IEEE ICDCS'19)
• *​FCooper: Feature based Cooperative Perception for Autonomous Vehicle Edge Computing System Using 3D Point Clouds. (IEEE SEC’19)
• *​Adversarial Sensor Attack on LiDAR-based Perception in Autonomous Driving. (ACM CCS'19)

Week 1 Activites

Get ORBIT/COSMOS account and familiarize oneself with the testbed procedures