Data-Driven Control Methods for High-Velocity Vehicle Control Surfaces

Author(s):

D. Johann Djanal-Mann and Muthuvel Murugan

Abstract:

This report investigates data-driven control strategies for high-velocity vehicle (HV) control surfaces, focusing on the demanding requirements imposed by high-velocity flight environments. Traditional control methods, reliant on analytical models, often struggle with the nonlinearities and uncertainties inherent in HV dynamics. To address these challenges, the study develops and compares two data-driven approaches: a classic proportional–integral–derivative (PID) controller and a modern differentiable predictive controller (DPC). Both are designed using machine-learned models trained on experimental data from a rapid control prototyping platform. The PID controller, tuned via neural state-space model, demonstrates robust, fast, and accurate reference tracking with negligible overshoot and steady-state error. The DPC uses a neural ordinary differential equation model to enable gradient-based optimization and end-to-end policy learning, achieving even faster rise and settling times than the PID in closed-loop simulations. While both controllers deliver high-performance actuation under the tested conditions, the DPC offers greater flexibility and efficiency for complex, nonlinear systems. These findings highlight how data-driven control and learning-based modeling are advancing the reliability and responsiveness of HV control surfaces, supporting future high-velocity maneuverability and survivability.