Grasping unknown objects using depth gradient feature with eye-in-hand RGB-D sensor

It is a remaining challenge for intelligent robot to interact with human daily living environment; one of the scenario is grasping objects from the table. Because of the massive variety of objects in daily life, it is still a hard task for robots to achieve. In this work, we propose an RGB-D eye-in-hand system and an effective grasp selection algorithm to grasp objects without prior knowledge of the object with a parallel-plate gripper, depending on depth information from the grasping direction. Without modeling, learning, training and segmentation, the robot can find an optimal position to place the gripper from single direction. Experiments of grasping single and multiple objects on a table are conducted to verify the feasibility of our approach. The results show that our approach performs well both in accuracy and efficiency.