Visual Control of an AUV for Multi-Robot Intervention Tasks

In the last decades, robots have been used to explore areas hard to reach for humans. Underwater environments fall into this category, since their operating conditions make even simpler operations risky to be carried out by divers, especially when they have to be performed at high depth. A possible approach to overcome this problem is to use a Remotely Operated Vehicle (ROV), although this easily becomes a difficult and expensive solution because it usually requires a sophisticated support infrastructure and specialized staff. In this regard, the project MERBOTS proposes a new robot-based methodology to make intervention tasks safer, simpler and at a lower cost. The proposed system involves two vehicles, being one of them a Hybrid ROV (H-ROV) equipped with an arm and a manipulator, as well as the necessary perception devices, which altogether implement the supervised intervention task, while the other is an Autonomous Underwater Vehicle (AUV) endowed with cameras to provide alternative points of view of the target for the operator in charge of the H-ROV, enabling thus a more robust and reliable operation. In this paper, we focus on the target detection and tracking tasks to be performed to provide this secondary view, which are part of the goals addressed by the SUPERION subproject. To this end, the object to manipulate must be ensured to appear continuously in the field of view of the camera placed in the AUV. This naturally leads to the implementation of a visual servoing task, whose input is the image stream coming from the AUV camera and its outputs are the velocity commands to be sent to the vehicle controller so as to keep the target in the field of view at all times. Due to its well-known robustness and simpler implementation, in this application, we choose an Image-Based Visual Servo-