An Open-Source Pipeline for Automated Subsea Object Grasping

Subsea manipulation and interaction are core tasks for remotely operated vehicle (ROVs). However, due to constrained perception, complex system dynamic and the adversive marine environment, manipulation remains a risky and time-consuming task for human operators, and is extremely challenging to automate. A core challenge with the development of subsea autonomy is the overhead in accessing both a platform capable of performing manipulation tasks and a suitable test environment. This makes rapid, iterative development, as well as the collection of extensive data sets for the training of machine learning algorithms, prohibitively expensive. We describe here an open source software stack, written in the ROS1 environment, for performing autonomous manipulation of pre-specified but not precisely modelled subsea objects, tested both in simulation and on a low-cost manipulation testbed. The system is provided as a baseline from which more complex and sophisticated manipulation autonomy approaches can be developed.