UNav-Sim: A Visually Realistic Underwater Robotics Simulator and Synthetic Data-generation Framework

Abdelhakim Amer,Olaya Álvarez-Tuñón,Halil Ibrahim Ugurlu,Jonas le Fevre Sejersen,Yury Brodskiy,Erdal Kayacan

2023-10-18

Abstract:Underwater robotic surveys can be costly due to the complex working environment and the need for various sensor modalities. While underwater simulators are essential, many existing simulators lack sufficient rendering quality, restricting their ability to transfer algorithms from simulation to real-world applications. To address this limitation, we introduce UNav-Sim, which, to the best of our knowledge, is the first simulator to incorporate the efficient, high-detail rendering of Unreal Engine 5 (UE5). UNav-Sim is open-source and includes an autonomous vision-based navigation stack. By supporting standard robotics tools like ROS, UNav-Sim enables researchers to develop and test algorithms for underwater environments efficiently.

Robotics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the deficiency in the rendering quality of existing underwater robot simulators, which limits their ability to transfer from the simulated environment to real - world applications. Specifically, many existing underwater simulators cannot accurately simulate complex underwater scenes due to their low rendering quality, which poses a challenge to the training and testing of AI algorithms that rely on image input. For this reason, the author introduced UNav - Sim, which is the first open - source underwater robot simulator based on Unreal Engine 5 (UE5). It can provide efficient and highly - detailed rendering, thus supporting the development of AI and vision - based navigation algorithms. By supporting standard robotic tools such as ROS, UNav - Sim enables researchers to develop and test algorithms applicable to underwater environments more effectively.