Pritam Dash is a PhD Candidate in the Department of Electrical and Computer Engineering at the University of British Columbia (UBC), advised by Prof. Karthik Pattabiraman. His research is in the intersection of AI, control theory and security, focusing on safe autonomy even when sensors are compromised. Pritam has published in top-tier venues such as DSN, AsiaCCS and ACSAC, and presented his work at Usenix Enigma. He has received the Best Paper award at DSN’21.
Robotic Autonomous Vehicles (RAV) are transforming various industries by advancing autonomy. RAVs rely on sensors for perception and decision-making. However, security attacks or faulty sensor inputs (collectively referred to as sensor anomalies) can corrupt perception, and cause erroneous and potentially dangerous decisions. Anomaly detection alone is not enough, as it does not prevent adverse consequences such as incorrect actions or crashes. My research combines AI and control theory to design recovery techniques for safe autonomy in RAVs. Specifically, I have proposed attack detection, diagnosis, and recovery techniques for securing RAVs from sensor anomalies. I have developed a feed-forward control-based approach that enables both proactive attack detection and resilient recovery, allowing RAVs to safely complete tasks despite anomalies. In addition, I have designed a unified anomaly detection and recovery framework to safeguard RAVs under strong multi-sensor anomalies. My recent work proposes a recovery technique that actively monitors potential safety violations (e.g., collisions), even under compromised sensors, and derives actions to minimize disruptions.