Reducing AUV Energy Consumption Through Dynamic Sensor Directions Switching via Deep Reinforcement Learning

Jiawei Liu, Yuanbo Xu, Shanshan Song, Lu JiangAAAI 2025 · OpenReview

TL;DR

This paper introduces RECS, a novel deep reinforcement learning approach for Autonomous Underwater Vehicles (AUVs) that dynamically adjusts sensor direction and radius to significantly reduce energy consumption. By intelligently focusing sensing efforts on critical areas, RECS achieves over 53% energy savings without compromising navigation or obstacle avoidance.

What Problem Does This Paper Address?

Autonomous Underwater Vehicles (AUVs) are vital for marine exploration and data collection, but their operational duration is severely limited by onboard energy reserves. A significant portion of this energy is consumed by the AUV's sensing systems, which are crucial for navigating complex and unpredictable underwater environments and avoiding obstacles. While previous research has focused on optimizing AUV movement and computation for energy efficiency, the energy expenditure of sensing itself has often been overlooked. This paper addresses the gap of efficiently managing sensing energy by proposing a flexible sensing system that dynamically adapts its focus, rather than operating with all sensors at full capacity, which leads to wasted energy in "ineffective sensing areas."

Methods

The core of the RECS (Reduce Energy Consumption by AUV Sensing system) method is a Deep Q-Network (DQN) agent that learns to dynamically control the AUV's sensing system. The approach involves several key components:

  • Environment Division and Weighting: The underwater environment is divided into square areas, with the size determined by the maximum sensing radius. Paths are planned, and the edges of these areas are assigned weights (high, medium, low) based on their relevance to the planned path. Edges along the planned path receive high weights, while those not on the path receive low weights and incur penalties if sensed.
  • Dynamic Sensing: The AUV's sensing system is divided into eight directional segments. Based on the current state (AUV position, energy, obstacles, target), the DQN agent selects which edge(s) to sense and dynamically adjusts the sensing direction and radius. This ensures that sensing efforts are concentrated on the most relevant parts of the environment.
  • Obstacle Avoidance: The Artificial Potential Field (APF) method is employed for obstacle avoidance. When an obstacle is detected that blocks the planned path, RECS re-plans the path and reassigns edge weights, allowing the system to adapt to new environmental configurations.
  • Deep Reinforcement Learning (DQN): RECS utilizes DQN to learn an optimal policy for selecting sensing actions. The state space includes AUV energy consumption, obstacle detection, current position, and target position. The action space involves selecting which of the four edges of the current area to sense. Rewards are designed to encourage sensing high-weight edges and reaching the target, while penalizing sensing low-weight edges or colliding with obstacles.

Datasets

The paper describes experiments conducted in a simulated virtual environment. The environment is a 100x100 unit map with randomly placed obstacles. For training, 64 different start and target point pairs were used. For performance evaluation, the authors used two different map sizes (100x100 and 200x200) and tested across 64 points with varying sensing radii. No specific public datasets were mentioned; the experiments were conducted in a custom simulation.

Key Insights and Ideas

  • Sensing is a Major Energy Drain: The paper highlights that sensing systems, often overlooked, are a significant contributor to AUV energy consumption. By focusing on this aspect, substantial energy savings are achievable.
  • Dynamic, Context-Aware Sensing is Key: The effectiveness of RECS stems from its ability to dynamically adapt sensing direction and radius based on the AUV's current state and planned path. This intelligent focus avoids wasting energy on sensing areas that are not immediately relevant.
  • Weighting Edges Guides Sensing: Assigning weights to environmental edges based on the planned path provides a crucial signal for the reinforcement learning agent, guiding it to prioritize sensing critical areas.
  • Integration of RL and APF: The synergistic combination of DQN for intelligent sensing control and APF for robust obstacle avoidance creates a comprehensive and effective navigation and energy management system.
  • Generalizability: The method demonstrates robustness across different map sizes and sensing radii, suggesting it can be effectively applied to various AUV configurations and mission scenarios.

Key Results

  • Significant Energy Reduction: RECS reduces AUV sensing energy consumption by an average of 53.48% compared to using full sensors on. The minimum reduction observed was 50.8%.
  • Efficient Sensing: The method achieves this by selectively sensing only one or two edges per action, with an average of 2.4 sensing directions activated per action.
  • Comparable Travel Distance: The energy savings are achieved without significantly increasing the AUV's travel distance or negatively impacting mission success.
  • Robustness: RECS maintains its performance across different sensing radii (7.07 and 14.14) and map sizes, indicating good transferability.

Limitations and Future Work

  • Static Obstacles: The current implementation assumes all obstacles are static. The authors acknowledge that real-world environments often feature dynamic obstacles, which would require further adaptation.
  • 2D Environment: The current work is limited to a 2D environment. Future work aims to extend the simulation to a 3D space for more realistic scenarios.
  • Movable Obstacles: The inclusion of movable obstacles is identified as a future research direction to enhance the system's realism and applicability.
  • AUV Movement Speed: The AUV's movement speed is kept constant to simplify the problem. Future work could explore optimizing movement speed in conjunction with sensing strategies.

Summary updated

Abstract

Autonomous underwater vehicle (AUV) is crucial for marine applications such as ocean data collection, pollution monitoring, and navigation. However, their limited energy resources constrain their operational duration, posing a significant challenge for long-term operations. Due to the complex and unpredictable nature of the underwater environment, AUVs allocate energy to their sensing systems to sense the surrounding environment and avoid obstacles. Existing methods focus on reducing energy consumption on AUV computing and movement, neglecting sensing energy consumption and few attempts have been made to balance the AUV energy and sensing ability with a flexible sensing system. Along these lines, we consider both AUV energy consumption and flexible sensing abilities, and propose a deep reinforcement learning-based method to Reduce Energy Consumption by AUV Sensing system (RECS). Specifically, we build an AUV sensing system in a 2-dimension space, with controllable 8-direction sensing abilities to collect the environment information dynamically. Then we divide the underwater environment into several areas and assign weights on the edges of areas based on the AUV planned path. Additionally, we dynamically switch the sensors in different directions and radii to sense the edges of the area where the AUV is located. The Artificial Potential Field (APF) method is employed to re-plan the AUV path to avoid obstacles and reach the target point effectively. Experimental results demonstrate that compared to full sensors on, our method reduces energy consumption by 53.48% and is capable of generalizing to varying environments and varying sensing system radii.
BibTeX
@inproceedings{liu2025reducing,
  author = {Jiawei Liu and Yuanbo Xu and Shanshan Song and Lu Jiang},
  title = {{Reducing AUV Energy Consumption Through Dynamic Sensor Directions Switching via Deep Reinforcement Learning}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2025},
  url = {https://openreview.net/forum?id=OQ4TxCNOE8},
}
Figure