VSCL graduate students Garrett Jares, Chris Leshikar, and Hannah Lehman will present papers in January at the 2022 AIAA SciTech Forum in San Diego, California.
Garrett Jares ’17 will be presenting the paper “Flight Demonstration and Validation of Control Acquisition Autopilot Attack”. The paper investigated a method of cyber attack by which an attacker might take over control of a vehicle. This paper built upon prior work by demonstrating and validating the attack on a DJI F450 quadrotor running the ArduCopter autopilot. The experiments focused on two scenarios. One in which the victim performed regulation while the attacker performed non-zero setpoint control and another in which the victim performed non-zero setpoint control while the attacker performed regulation of the system. The experimental results show how the attack poses a threat to real-world UAS and evaluates its performance under different control scenarios.
Hannah Lehman ’20 will be presenting the paper “Addressing Undesirable Emergent Behavior in Deep Reinforcement Learning UAS Ground Target Tracking”, which seeks to investigate and further understand the impact of emergent behavior in reinforcement learning controlled UAS. The paper builds on previous work by further investigating a fixed wing tracking a ground target through reinforcement learning and extends the learning environment and possible agent behavior. The emergent behavior is discovered, categorized, and mitigated through a number of algorithmic, reward, and environment modifications. These approaches are evaluated in simulation based on their ability to improve tracking and extend total tracking time.
Chris Leshikar ’20 will be presenting the paper “System Identification Flight Testing of Inverted V-Tail small Unmanned Air System”, which addresses challenges in conducting flight testing an inverted V-Tail fixed-winged vehicle and the results obtained from the flight tests. The goal of the flight tests was to obtain longitudinal, lateral/directional and combined longitudinal lateral/directional linear state-space model for the RMRC Anaconda using the Observer\Kalman Identification (OKID) algorithm. Both manual and automated excitation signals were injected into the Anaconda. Parametric sweeps of the excitation signals were performed using the Developmental Flight Test Instrumentation Two (DFTI2) system. The identified longitudinal linear state-space model modelled the longitudinal dynamics well and the identified lateral/directional reasonably well while the identified combined longitudinal lateral/directional model showed decent correlation with the decoupled models.