Energy Management for Autonomous Underwater Vehicles using Economic Model Predictive Control

Niankai Yang, Dongsik Chang, Mohammad Reza Amini, Matthew Johnson-Roberson, Jing Sun

Abstract

Autonomous Underwater Vehicles (AUVs) face significant endurance challenges due to limited battery capacity. This paper presents a real-time energy management strategy using Economic Model Predictive Control (EMPC). By introducing a two-stage decomposition of vehicle motion—separating maneuvers into dynamic steering and static cruising stages—the authors approximate the "energy-to-go" terminal cost. This approach achieves energy efficiency comparable to global optimization methods while reducing computational overhead by 98%, enabling real-time onboard deployment.

Introduction

Endurance is the primary constraint for AUVs operating in remote, high-pressure environments. Traditional energy management relies either on computationally intensive global optimization (e.g., Direct Collocation), which is too slow for real-time use, or standard path-tracking MPC, which prioritizes trajectory accuracy over energy conservation. This research bridges the gap by developing a controller that is both fast enough for onboard hardware and as energy-efficient as a global optimizer.

Methodology

The proposed strategy utilizes Economic Model Predictive Control (EMPC) to directly minimize energy consumption. A critical challenge in EMPC is defining the "terminal cost"—the estimated energy required to reach the goal beyond the short prediction horizon.

The authors' core innovation is a two-stage decomposition of AUV motion. They observed that trajectories typically consist of a Dynamic Stage, where the vehicle uses thrusters to rotate and align toward the goal, followed by a Static Stage, where it cruises in a straight line.

Two-stage decomposition of AUV thrust into Dynamic and Static stages for energy-to-go approximation.

By modeling these stages analytically, the "energy-to-go" can be calculated rapidly. This allows the controller to make globally optimal decisions within a short local prediction window. Furthermore, the model decouples horizontal motion (surge and yaw) from vertical motion (pitch and heave), focusing optimization where the greatest energy savings occur.

Experimental Setup and Evaluation

The controller was evaluated using a high-fidelity 6-DOF dynamic model of the DROP-Sphere, an open-source, 6000m-rated AUV developed by the DROP Lab at the University of Michigan.

The DROP-Sphere AUV platform used for model identification and simulation, showing thruster configurations.

Simulations were conducted in MATLAB/Simulink, comparing the proposed Energy-Optimal EMPC (EO-EMPC) against two benchmarks: a global Direct Collocation (DC) method and a standard Line-of-Sight (LOS-MPC) controller.

Key Results

  • Energy Efficiency: EO-EMPC consumed 52.89% less energy than LOS-MPC and was within 10% of the theoretically optimal DC method.
  • Computational Speed: Calculations for a full mission took approximately 16 seconds with EO-EMPC, representing a 98% reduction compared to the 1,181 seconds required by DC. This makes the algorithm viable for real-time onboard deployment.
  • Robustness: Unlike open-loop global optimization, the closed-loop EMPC successfully handled initial position perturbations, reaching the target efficiently where DC failed to adapt.

Conclusion and Future Work

The EO-EMPC framework provides a viable path for real-time, energy-efficient AUV control by simplifying complex non-linear energy paths into a "Turn then Cruise" model. Future research will extend this model to 3D maneuvers, incorporate the effects of ocean currents, and validate the system on physical hardware in open-water environments.

Summary updated

Abstract

This paper investigates the problem of energy-optimal control for autonomous underwater vehicles (AUVs), To improve the endurance of AUVs, we propose a novel energy-optimal control scheme based on the economic model predictive control (MPC) framework. We first formulate a cost function that computes the energy spent for vehicle operation over a finite-time prediction horizon. Then, to account for the energy consumption beyond the prediction horizon, a terminal cost that approximates the energy to reach the goal (energy-to-go) is incorporated into the MPC cost function. To characterize the energy-to-go, a thorough analysis has been conducted on the globally optimized vehicle trajectory computed using the direct collocation (DC) method for our test-bed AUV, DROP-Sphere. Based on the two operation modes observed from our analysis, the energy-to-go is decomposed into two components: (i) dynamic and (ii) static costs. This breakdown facilitates the estimation of the energy-to-go, improving the AUV energy efficiency. Simulation is conducted using a six-degrees-of-freedom dynamic model identified from DROP-Sphere. The proposed method for AUV control results in a near-optimal energy consumption with considerably less computation time compared to the DC method and substantial energy saving compared to a line-of-sight based MPC method.
BibTeX
@inproceedings{yang2019energy,
  author = {Niankai Yang and Dongsik Chang and Mohammad Reza Amini and Matthew Johnson-Roberson and Jing Sun},
  title = {{Energy Management for Autonomous Underwater Vehicles using Economic Model Predictive Control}},
  booktitle = {ACC},
  year = {2019},
  url = {https://openreview.net/forum?id=jJm4yhMPgR3v},
}
Figure