I develop frameworks that detect, categorize, and recover from attacks across the full UAV control loop.
My work focuses on PWM integrity, multimodal side-channel fusion, and hardening IDS through adversarial data augmentation and reinforcement learning. I validate methods with hardware-in-the-loop, real hardware Trojan experiments, and flight tests.
Areas: UAV security · CPS · IDS · PWM integrity · Side-channels · Adversarial ML
Affiliation: Florida International University · CREPES
Funding & collaborations: AFRL · NSF · DOE · NSA (projects)
Selected highlights
Impact and Stealth Analysis of UAV Attacks
This is the core methodology that guides my experimental and modeling work. Table 1 lists common attack goals and affected signals.
The training and evaluation of intrusion detection systems for UAVs require attack data that capture both
physical impact on flight and the ability to remain stealthy relative to sensor telemetry. I generate and
validate attack vectors using simulation, adversarial ML, GAN-based data augmentation, reinforcement
learning, and hardware-in-the-loop experiments. I evaluate analytics on two axes. First, impact — whether
a vector causes measurable instability, trajectory deviation, or mission failure. Second, stealth — whether
a vector bypasses sensor-only detectors while producing unsafe actuation.
| Attack Goal |
Example Sensor Measurements Affected |
Impacted Actuators / Signals |
| GPS spoofing |
GPS position, fused velocity, heading |
Motor thrust allocation via PWM |
| GPS jamming |
Loss of GPS, increased IMU reliance |
Flight trajectory control; hover stability |
| False data injection |
Accelerometer, gyroscope, barometer |
Controller outputs → PWM commands |
| PWM manipulation / Hardware Trojan |
Controller-to-actuator command path |
Propeller motors; roll/pitch/yaw thrust |
| CPU / battery interference |
Power draw, CPU load, timing jitter |
Actuator timing and repeatability; control latency |