Enhancing real-time drone trajectory tracking performance using model predictive control on FPGA-SoC

Abstract

This study presents a robust control framework for enhancing real-time trajectory tracking of drones, overcoming the limitations of conventional proportional - integral - derivative (PID) controllers. The proposed architecture integrates a model predictive controller (MPC) with a Kalman filter to achieve accurate state estimation under noisy measurements, while explicitly handling system constraints. To address the computational bottleneck of MPC, we employ a fast gradient method-based solver with warm-start initialisation and fixed-iteration updates, ensuring deterministic and efficient execution. Experimental results on a Zynq-7000 FPGA-SoC demonstrate that the optimised software-only implementation achieves real-time feasibility, with an average computation time of 0.18 ms (≈1.4% of a 12.8 ms control cycle). Compared to PID, the proposed MPC reduces the maximum absolute axis tracking error from ≈0.5 to ≈0.2 m and generates significantly smoother control signals (~13x lower variance and =7x lower mean rate of change). These improvements enhance trajectory accuracy and flight stability while extending actuator lifetime, all achieved without requiring hardware acceleration.