There is a need for autonomous exploration robots because vast majority of our oceans are unexplored, and direct human exploration of the deep sea is an expensive and extremely dangerous endeavor.

Our work explores a novel approach to modeling curiosity in a mobile robot, which is useful for exploration, monitoring, and adaptive data collection tasks, especially in the context of long term autonomous missions where pre-programmed missions are likely to have limited utility. The proposed curious robot, through the use of unsupervised and partially supervised machine learning techniques, learns an online, high- level visual model of benthic environment, and then identifies significant interesting phenomena in- situ. Using these detections, the robot can adaptively plan its path, focusing the data collection over interesting regions. Such surveys will enable us to study transient and patchy phenomena such as: continuously moving organisms, reproductive aggregations, predation events, and whale and log falls, which are difficult to study using non-adaptive uniform sampling techniques. We used ROST realtime topic modeling technique to build the semantic perception model of the environment.

We have demonstrated the proposed algorithm, implemented on Aqua robot, to be able to do tasks such as coral reef inspection, diver following, and sea floor exploration, without any prior knowledge about the environment.