Anomaly Detection in LiDAR Data Using Virtual and Real Observations
Keiichiro Hattori, Ranulfo Plutarco BEZERRA Neto, Shotaro Kojima, Yoshito Okada, Kazunori Ohno, Shintaro Ishihara, Kenji Sawada, Satoshı Tadokoro · 2023
With the constant progress of robot integration within society, security remains a paramount concern, particularly due to the increasing potential for damage arising from malicious attacks. However, the inherent challenges of preventing every potential attack vector require innovative security measures. This study presents a unified anomaly detection method employing a virtual environment mirroring real-world observations. By focusing on the discrepancies between real and virtual observational data, anomalies can be effectively detected, the types of which are further identified through a time-series analysis of these discrepancies. Results demonstrated the capacity of our method to successfully detect and categorize anomalies arising from various sources including environmental noise, robotic malfunctions, and communication-based attacks. Furthermore, our anomaly detection method consistently achieved precision, recall, and F1 scores higher than 90%, underscoring its effectiveness.