Multi-Head PPO-Based Path Planning for AUV Data Collection under Dynamic Ocean Currents
Abstract
This paper proposes a reinforcement learningbased path planning framework for autonomous underwater vehicle (AUV) data collection in dynamic ocean currents, simultaneously considering propulsion and communication energy, buffer capacity, and Age of Information (AoI). By employing a multi-head PPO network, the method jointly learns node selection and velocity control. Simulation results indicate that the proposed approach outperforms the greedy baseline in energy efficiency and mission success rate, with approximately $\mathbf{1 2 \%}$ higher offloading efficiency and 11% lower AoI.
Fields of Study & Subjects
Underwater Vehicles and Communication Systems
Age of Information Optimization
Reinforcement Learning in Robotics
Publication details
- Journal / source
- International Conference on Electronics, Information, and Communication (ICEIC)
- Pages
- 1–2
- Publication date
- 18 January 2026
- Language
- EN