GPU-Assisted Learning on an Autonomous Marine Robot for Vision-Based Navigation and Image Understanding

We present a GPU-based integrated robotic platform that enables collision avoidance, navigation, and image understanding on a single underwater vehicle. The platform enables observational tasks such as coral reef health assessment by enabling simultaneous operation of multiple image analysis taskswhile navigating in close proximity to obstacles. The integration of a GPU allows us to leverage deep neural networks for collision avoidance and automated object detection and classification while a general purpose CPU processes images to perform visual Simultaneous Localization and Mapping (SLAM). In this paper, we describe the system architecture and summarize experimental results for coral detection and collision-free navigation.