Hacking Autonomous Vehicles?! The Importance of Sensor Security
While humans possess excellent sensors and decision-making mechanisms, they sometimes make incredibly elementary mistakes. When driving a car, these mistakes can lead to irreversible consequences. In 2021, traffic accidents in Japan resulted in 2,636 fatalities and 361,768 injuries. To reduce such harm, various companies and research institutions around the world are conducting research and demonstration experiments to achieve fully autonomous driving. Within a few years, many of you reading this article may have the opportunity to test-drive one of these vehicles.
The rapid evolution of autonomous driving technology in recent years is due not only to advances in recognition AI but also to significant contributions from the increased precision of 3D sensors. LiDAR (Light Detection and Ranging) sensors, which obtain high-precision 3D data in real time, play a major role in current Level 4 autonomous driving. LiDAR sensors are used for core autonomous driving functions such as self-localization, which determines which lane the autonomous vehicle is in, and object detection, which identifies surrounding cars and pedestrians.
On the other hand, sensor spoofing attacks—which exploit vulnerabilities in sensors like LiDAR to inject false data through hacking (attacks)—have been demonstrated and pose a significant threat to autonomous driving. By exploiting this attack, it is possible, for example, to trick an autonomous vehicle into misrecognizing a wall appearing in front of it, inducing sudden braking and causing harm to passengers and following vehicles. Because such accidents could fundamentally undermine trust in autonomous driving technology, the Yoshioka Laboratory is conducting sensor security research to prevent these kinds of sensor attacks. For instance, we aim to realize safe and secure autonomous vehicles by discovering new LiDAR vulnerabilities and proposing sensor structures that prevent hacking.
Sensor security research is not straightforward, as it requires spanning a wide range of fields, including LiDAR sensors, autonomous driving control, and AI. However, it is a field where the flexible thinking unique to undergraduate and graduate students can be leveraged, and it provides practical experience across a broad spectrum of sensors and software. In particular, LiDAR sensor security is a pioneering field where established defense methods have not yet been developed, offering the potential to make a significant societal impact depending on one's efforts.