Learning-Based Underwater Autonomous Grasping via 3D Point Cloud

Underwater autonomous grasping is a challenging task for robotic research. In this paper, we propose a learning-based underwater grasping method using 3D point cloud generated from an underwater stereo camera. First, we use Pinax-model for accurate refraction correction of a stereo camera in a flat-pane housing. Second, dense point cloud of the target is generated using the calibrated stereo images. An improved Grasp Pose Detection (GPD) method is then developed to generate the candidate grasping poses and select the best one based on kinematic constraints. Finally, an optimal trajectory is planned to finish the grasping task. Experiments in a water tank have proved the effectiveness of our method.

I. Introduction

Underwater manipulation is challenging because of low accurate control performance of the underwater manipulator, poor lighting, more noise of the sensors and harsh underwater environment. Learning-based manipulation in the air has got many amazing achievements, such as learning ambidextrous robot grasping [1], learning complex manipulation skills [2]. However, most of the current underwater manipulation research focuses on the basic task, like grasping a box [3] –[5], catching a ball [6] or remote control [7]. There exists a huge gap between underwater grasping and the latest research of robotics. Therefore, how to improve the ability of underwater autonomous grasping is important.