Autonomous robot-diver assistance through joint intention theory

While monitoring their own life support systems, underwater divers have to perform complex technical tasks in unstructured environments. An Underwater Robotic Assistant (UWRA) could assist a diver during the execution of these tasks by navigating the diver to the worksite, ferrying tools from the surface, and monitoring the diver's physiological responses. However, for the UWRA to be an asset to the diver and not a hindrance, it is proposed that the UWRA should interact with the diver similarly to how a human diver would interact with another human diver. In order to frame the human-robot interaction, Joint Intention Theory (JIT) was used to distinguish between planned and emergent coordination. Still, before JIT can be fully realized, the UWRA has to be capable of interpreting hand signals, recognizing equipment, and detecting the diver. In this paper, seven object-recognition algorithms, that do not require large sets of training data, are assessed for performance in terms of Receiver Operating Characteristic (ROC) plots and processing time using real-world data. It is shown that several of the template-based object recognition algorithms outperformed a state-of-the-art detector.