OceanSim is a high-fidelity underwater simulation framework designed to accelerate the development of robust underwater perception solutions. Leveraging GPU-accelerated rendering and advanced physics-based techniques, OceanSim accurately models both visual and acoustic sensors, significantly reducing the simulation-to-real gap.
Towards scaling marine perception with synthetic data, we are now bringing OceanSim 0.2 with our explorations and solutions for an underwater detection task by synthetic data.
🚀 GPU-accelerated: OceanSim fully leverages the power of GPU-based parallel computing. OceanSim is built on top of NVIDIA Isaac Sim and is part of NVIDIA Omniverse ecosystem, which provide high performance and real-time rendering.
🌊 Physics-based underwater sensor rendering: Experience realistic simulations with advanced physics models that accurately replicate underwater sensor data under varied conditions.
📡 ROS 2 integration: Easily integrate into you existing projects with support for publishing of sensors and control over ROS 2.
🎨 Efficient 3D workflows: Users of OceanSim can enjoy efficient 3D workflows empowered by OpenUSD.
🤝 Built by the community, for the community: OceanSim is an open-source project and we invite the community to join us to keep improving it!
[2026/8]We now incorporate a pipeline for our underwater object detection project in our paper and bringing OceanSim 0.2.[2025/9]OceanSim is now compatible with Isaac Sim 5.0 GA.[2025/4]OceanSim is featured by NVIDIA Robotics![2025/4]🔥 OceanSim 0.1 is released![2025/3]🎉 OceanSim will be presented at AQ²UASIM and the late-breaking poster session at ICRA 2025![2025/3]OceanSim paper is available on arXiv. Check it out here.
- Documentation for OceanSim provided example
- Built your own digital twin documentation
- Code release
- ROS2 Example release, contributed from Tang-JingWei
We divide the documentation into four parts:
- Installation
- Running OceanSim
- ROS 2 [❗NEW❗]
- Building Your Own Digital Twins with OceanSim
- Underwater SDG using standalone scripts [❗NEW❗]
We welcome contributions and discussions from the community!
- Use Discussions to share your ideas and discuss with other users.
- Report bugs or request features by opening an issue in Issues.
- Submit a pull request if you want to contribute to the codebase. Please include the description of your changes and the motivation behind them in the pull request. You can check more details in CONTRIBUTING.md.
OceanSim is an open-source project initiated by the Field Robotics Group (FRoG) at the University of Michigan. We hope to build a vibrant community around OceanSim and invite contributions from researchers and developers around the world! A big shoutout to our contributors:
Jingyu Song, Haoyu Ma, Onur Bagoren, Advaith V. Sethuraman, Yiting Zhang, Anja Sheppard, Tanner Aslan, Elias Fandi, Ashrith Edukulla and Katherine A. Skinner.
If you find OceanSim useful for your research, we would appreciate that you cite our paper:
@misc{song2025oceansim,
title={OceanSim: A GPU-Accelerated Underwater Robot Perception Simulation Framework},
author={Jingyu Song and Haoyu Ma and Onur Bagoren and Advaith V. Sethuraman and Yiting Zhang and Katherine A. Skinner},
year={2025},
eprint={2503.01074},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2503.01074},
}
If you use the sonar model in OceanSim, please also cite the HoloOcean paper as the HoloOcean sonar model inspires our sonar model implementation:
@inproceedings{Potokar22iros,
author = {E. Potokar and K. Lay and K. Norman and D. Benham and T. Neilsen and M. Kaess and J. Mangelson},
title = {Holo{O}cean: Realistic Sonar Simulation},
booktitle = {Proc. IEEE/RSJ Intl. Conf. Intelligent Robots and Systems, IROS},
address = {Kyoto, Japan},
month = {Oct},
year = {2022}
}
OceanSim - A GPU-Accelerated Underwater Robot Perception Simulation Framework