Simulation of Semantically-Aware Obstacle Avoidance Algorithms for Underwater Robots

Abstract

Researchers working with autonomous underwater vehicles (AUVs) must be able to test their robotic vision-based algorithms on-location in field trials. These trials can be time-consuming and carry the risk of unforeseen hardware and software bugs limiting the amount of data gathered. To this end, being able to test and evaluate algorithms risk-free in a computer simulation beforehand can be invaluable for researchers. Current simulation solutions can provide realistic physics and easily modifiable worlds, however using a 3D graphics engine to create realistic underwater scenarios can improve results considerably and ease the transition into a real-world environment. This research demonstrates the potential of the Unity 3D graphics engine to provide a realistic simulation environment by running and evaluating a vision-based underwater obstacle avoidance algorithm on a simulated Aqua robot. We find that Unity can provide simulated stereo images that can be used by the Semantic Obstacle Avoidance for Robots (SOAR) algorithm to navigate a simple obstacle field en route to a predetermined goal position.