SLAM-based grasping framework for robotic arm navigation and object model construction

A typical grasping system consists of three subtasks: object model acquisition, grasping point calculation and navigation of the robotic arm. These tasks are usually considered separately. In this paper, we present a framework that combines these steps together. Our main motivation is that as the robot are moving, new information should be obtained from the sensor and these information should be used to increase accuracy of the model of the object and the current position of the robot. In other words, our framework employs SLAM approach. We also provide several real world implementations of our framework and compare them to illustrate the benefit of our framework. In particular, we install a depth camera DepthSense DS325 on a Katana robotic arm and use this system to simulate the navigation of the robotic arm for grasping. The comparison of our implementation confirms effectiveness of our framework.