UUVSim: Intelligent Modular Simulation Platform for Unmanned Underwater Vehicle Learning

Zekai Zhang,Jingzehua Xu,Jun Du,Weishi Mi,Ziyuan Wang,Zonglin Li,Yong Ren

DOI: https://doi.org/10.1109/ijcnn60899.2024.10650129

2024-01-01

Abstract:Unmanned underwater vehicles (UUVs) face challenges such as high hardware costs, security concerns, a lack of training data in the actual development and debugging. Creating a simulation platform for simulation verification, training, and learning presents a potential solution to address these challenges. However, this area has seen limited prior work, and existing underwater platforms lack accuracy, user-friendliness, and intelligence. Therefore, this paper introduces an intelligent simulation platform “UUVSim” based on the robot operating system and Gazebo. UUVSim modular integrates basic modules such as high-precision simulation scenarios, dynamic models, sensors and controllers, while reserving programming interfaces. In addition, UUVSim provides reinforcement learning environment for UUV intelligent learning, supplemented with scenario transfer training, multi-agent reinforcement learning, offline reinforcement learning techniques to realize efficiently training for complex tasks, multi-robot coordination, and simulation to reality (sim2real) deployment. Further, we validate these technologies through underwater target tracking benchmarks and sim2real experiments, demonstrating the platform’s practicality.