A Stochastic Compound Failure Model for Testing Resilience of Autonomous Fixed-Wing Aircraft
Our paper presents a simulation process to dynamically emulate the effects of certain adversarial flight conditions on fixed-wing, autonomous aircraft system actuators. We implement a PX4 Autopilot flight stack module that replaces the generated attitude control inputs with perturbed inputs to the plane's actuator mixer. The perturbed inputs rely on a Markov chain to model failure states that emulate adversarial (failing) actuator flight conditions. Simulated flight failures on a fixed-wing autonomous aircraft test the controller response to a stochastic failure sequence on a range of turning radii. Statistical measures of the differences between target and simulated flight paths demonstrate that a well-tuned PID controller remains competitive in the cascading, compound, transient failure regime.
Thelonious Abraham Cooper
MIT EAPS ESSG
Undergraduate Researcher
Cambridge
Website: theloniouscoop.dev
Thelonious Cooper is a junior in the electrical engineering department and an undergraduate researcher for the Earth Signals and Systems Group at MIT. Thelonious has broad interests but at this time he is particularly fascinated by systems science and embedded applications of machine learning.
Sai Ravela
Earth Signals and Systems Group
Principal Research Scientist
Sai Ravela directs the Earth Signals and Systems group, MIT where he is a PRS. Sai earned a Ph.D. (CS, UMASS Amherst, vision and robotics). His research includes autonomous observatories, cyclone risk, image recognition, seismic monitoring, exoplanet detection, and co-active learning. Dr. Ravela has over 125 articles, is a co-founder of Windrisktech LLC, an instrument-rated and hobby pilot. He received the 2016 MIT Infinite Kilometer award for outstanding research and exceptional mentorship.