TraCR Scholar Webinar - Kemal Akkaya

Leveraging Generative AI for Sensor Data Falsification on Drones and Automated Vehicles
Both drones and automated vehicles employ a large number of sensors to sense their environment. In the past, a lot of research has been dedicated to protection of these sensors from compromise since the attackers can manipulate the sensors readings to be able to control drones or vehicles. While extensive research has been done on anomaly/intrusion detection coming out of these sensors, the advancement of generative AI techniques such as Generative Adversarial Networks (GANs) and variational autoencoders (VAEs) introduces possibilities for more sophisticated falsified sensor data through false data injection (FDI) attacks. In this talk, we first present how multiple drone sensors data can be manipulated using GANs through several cases. We present experiment results on the success of these attacks and the need for better training for the existing filters. We then discuss how these attacks can be applicable to automated vehicles with multiple sensors.