RoboSub 2023 Computer Vision Project

Welcome to the Duke Robotics Club's RoboSub 2023 dataset, a comprehensive collection of meticulously curated and annotated images designed to train autonomous underwater vehicles for the RoboSub 2023 competition!

This 50,000+ image dataset consists of the four buoy glyphs—Cetus, Taurus, Serpens Caput, and Auriga—and the two gate glyphs for Abydos and Earth. Our dataset consists of equal amounts of real-world and simulated images which accurately represent the diverse underwater environment of RoboSub 2023.

The real-world images were captured in a competition-like environment with different glyph orientations and camera angles while the simulated images were generated using the Unity Perception package. Special emphasis was placed on simulating the specific environmental conditions of the TRANSDEC pool, such as its characteristic murkiness. The combination of these two types of images has resulted in a robust dataset specifically tailored to RoboSub 2023’s unique demands.

Our aim is to provide a rich resource for teams looking to train their models on accurate and extensive data. Access to this dataset is open to all, in line with the Duke Robotics Club's commitment to fostering a collaborative and innovative robotic community. We encourage you to use this data to fine-tune your models, share your findings, and contribute to the ongoing exploration and evolution of autonomous underwater vehicle technology.