Data-Enabled Predictive Control and Guidance for Autonomous Underwater Vehicles

TL;DR

The paper tackles robust waypoint path following for autonomous underwater vehicles without relying on explicit hydrodynamic models. It develops a data-driven framework based on Data-Enabled Predictive Control (DeePC), with a cascaded DeePC depth controller and a predictive 3-D guidance extension (PALOS) that integrates with DeePC for 3-D waypoint tracking. Key contributions include (i) DeePC-based heading control, (ii) cascaded DeePC depth control with inner-outer loop separation, (iii)PALOS to generate predictive references for DeePC, and (iv) extensive REMUS 100 simulations showing improved tracking and robustness against ocean currents and nonlinear dynamics compared to PI/PID controllers. The approach reduces modeling effort while delivering superior anticipatory performance, with practical implications for reliable AUV operation in dynamic ocean environments and clear avenues for future experimental validation.

Abstract

This paper presents a fully data-driven control framework for autonomous underwater vehicles (AUVs) based on Data-Enabled Predictive Control (DeePC). The approach eliminates the need for explicit hydrodynamic modeling by exploiting measured input-output data to predict and optimize future system behavior. Classic DeePC was employed in the heading control, while a cascaded DeePC architecture is proposed for depth regulation, incorporating a loop-frequency separation to handle the different dynamic modes of input and output. For 3-D waypoint path following, the Adaptive Line-of-Sight algorithm is extended to a predictive formulation and integrated with DeePC. All methods are validated in extensive simulation on the REMUS 100 AUV and compared with classical PI/PID control. The results demonstrate superior tracking performance and robustness of DeePC under ocean-current disturbances and nonlinear operating conditions, while significantly reducing modeling effort.

Figures (6)

• Figure 1: Classic AUV control architecture with the PID heading control and the cascaded PI/PID control of the heave position with an inner loop controlling the pitch angle.
• Figure 2: Control architecture of the DeePC-based heading control and the cascaded DeePC control of the heave position with an inner loop controlling the pitch angle.
• Figure 3: Illustration of the sampled points (red and black dots) along the past and future trajectories from the outer loop to illustrate the reference trajectory generation for the inner control loop in pink via linear extrapolation.
• Figure 4: Integration of PALOS into the DeePC framework for 3-D Path following.
• Figure 5: Responses in heave position $z_p$, pitch angle $\theta$, and yaw angle $\psi$ of DeePC and PI/PID to each desired reference. Associated control command angles for the rudder plane and stern plane, $\delta_r$ and $\delta_s$.