A Consistent Dataset for Dynamic Underwater Proximity Operations

Open data are scarce in underwater robotics, making it more difficult for researchers to develop new methods that address real-world problems. In this work, we present the experimental design, execution, and curation of three datasets that represent different conditions in a realistic underwater dynamic proximity operation. The raw datasets gathered during the deployments are post-processed to improve consistency. We detail and use a joint batch optimization technique that uses a probabilistic approach to iteratively search for the set of optimal agent trajectories that are in best agreement with the relative measurements provided by a USBL positioning system and optical pose measurements from a fiducial light array. Error analysis of the relative measurements with respect to the baseline and optimized trajectories validate our results, effectively providing ground-truth trajectories of the agents. The resulting datasets, together with their documentation, are publicly available at github.com/aldoteran/asko_2024_datasets