Abstract
This paper presents an electrically driven underwater manipulator named SeaArm-2 and considers the kinematic control approach to perform autonomous fixed-base grasping of an object of known size and shape at unknown positions underwater. SeaArm-2 has an integrated monocular camera near the gripper, which enables perception and object detection capabilities without the need for an external system. Furthermore, this work presents autonomous fixed-base seabed grasping operations utilizing the integrated camera and object detection functionality through a novel open source machine learning framework. A total of 20 trials have been conducted where the object was placed within reach but out of sight of the manipulator. The results from the experimental testing showed that the manipulator was able to detect the object in all trials, and successfully grasp the object in 17/20 trials.