A team of Rutgers ECE researchers led by Associate Professor Dario Pompili (PI) along with co-PIs Saman Zonouz (Associate Professor) and Bo Yuan (Assistant Professor) are the recipients of a $280K supplement award from the National Science Foundation under the DARPA-NSF Real-Time Machine Learning (RTML) program for the project titled “Trustworthy Autonomous UAVs via Compiler-Accelerated Efficient AI,” in collaboration with the University of Washington (UW).
The ECE team will optimize their developed Deep Neural Network (DNN)-based Unmanned Aerial Vehicle (UAV) controller security solutions (via firmware reverse engineering and symbolic execution) that include real-time intrusion detection and response capabilities. This project will enable trustworthy UAV flight operations despite complex security attacks via optimized distributed deployment of data-driven AI-based intrusion resilience on UAVs resource-limited processors. The technical challenges addressed in the project include: (1) Accurate and optimal security monitoring and decision making for flight-time attack detection and response; and (2) Resource-optimization and DNN acceleration for execution of security solutions side-by-side along with the drone’s main controller software stack. Success will be measured via full flight demonstration of security-enhanced UAV controllers against various attack types, e.g., Denial of Service (DoS), false data injection, and vulnerability exploitation. As part of this DARPA-NSF collaborative program, the team will deliver the first working solution and open research directions for use of real-time efficient AI for security protection of cyber-physical safety-critical platforms even beyond UAVs.
More details on the project can be found at the NSF page here.
Congratulations Dario, Saman and Bo!