Cyber Resilient Attitude Determination and Control for Space Vehicles

Attitude determination and control systems (ADCS) represent critical single points of failure for spacecraft, yet their resilience against cyberattacks remains underexplored. This paper evaluates five attitude control architectures including PD, LQR+KF, neural surrogate, median ensemble, and hybrid, under post-compromise conditions using simulations on a genuine spaceflight computer. Through Monte Carlo simulations with sensor corruption, actuator tampering, timing delays, and model mismatch attacks, we demonstrate that learning-based controllers exhibit severe instability with crash rates of 70-100\%, while model-based controllers degrade predictably and maintain attitude authority. These results indicate that cyber-resilient ADCS for contested environments should prioritize model-based designs with stability guarantees, using learning-based methods only as auxiliary components with validated fallback strategies.