Optimized LQR-Based Depth-Pitch Control of an Autonomous Underwater Vehicle

This paper investigates optimization-based tuning of a linear-quadratic regulator (LQR) for depth-pitch regulation of a REMUS-type autonomous underwater vehicle (AUV) with coupled vertical-plane dynamics. A linear depth-pitch model is employed to design and compare three state-feedback controllers: a genetic algorithm (GA)-tuned LQR, a constrained model predictive controller (MPC), and a classical pole-placement controller. The controllers are evaluated in simulation under nominal conditions, input disturbances, measurement noise, and a combined worst-case scenario using RMSE, IAE, settling time for both depth and pitch, overshoot, peak pitch excursion, and control-effort indices. Results show that MPC typically achieves the strongest depth-tracking accuracy (lowest RMSEz). In contrast, GA-LQR achieves comparable depth accuracy and often faster settling than MPC in several disturbance scenarios while maintaining robust performance. Control-effort analysis indicates that MPC demands the highest stern-plane activity, GA-LQR requires moderate-to-high effort (lower than MPC but substantially higher than pole placement), and pole placement is the most actuator-efficient but exhibits weaker disturbance tolerance and poorer transients. Overall, the study highlights trade-offs among tracking performance, pitch excursion, and actuator usage, with GA-LQR offering a practical middle ground compared to MPC.