Under-ice camera and sonar simulation for visual navigation

Navigation and mapping in relatively featureless underwater and under-ice environments is a difficult challenge. Using common sensors already onboard many unmanned underwater vehicles to aid in this process provides a self-contained and cost-effective solution. Evaluation of such visual navigation and mapping methods under development requires multiple UUVs to gather the required sonar and video data in underwater and under-ice environments, which is costly and prohibitive. It is also difficult to ensure accurate ground truth is available during such data collection missions. Alternatively, a simulation approach is presented here to provide datasets with full ground truth and fine-grained controllability, requiring much fewer resources. This can enable faster and more robust development of these algorithms prior to costly field testing. The simulation methods and datasets developed are presented herein. The code and datasets themselves will be made available as part of our open source commitment.