Model Predictive Control (MPC) is the gold standard for constrained multi-variable control, yet its computational burden often precludes deployment on low-cost embedded hardware. As a primary contribution to artificial intelligence, this work proposes distilling complex control policies into a symbolic Kolmogorov-Arnold network (KAN), creating a computationally efficient, explicit neural controller.

Designed an adaptive control architecture achieving a significant speedup and reduction of upwards of 1.5x in actuator Total Variation compared to Linear Time-Varying Model Predictive Control. Features rigorous experimental validation and proof of Lyapunov stability by ISS/GUUB and LaSalle-Yoshizawa Theorem.