An Industrial Robotics Application with Cloud Computing and High-Speed Networking

Abstract

This paper describes an industrial cloud robotics distributed application that was executed across a high-speed wide-area network. The application was implemented using ROS libraries and packages. The purpose of the application is to enable an industrial robot to perform surface blending. A Kinect sensor, a surface blending tool and a laser scanner are mounted on the robot arm. The arm is moved under software control to scan a work bench on which metal parts of variable size can be laid out at any orientation. The collected point cloud data is processed by a segmentation algorithm to find the surface boundaries. A Cartesian path planning algorithm is executed to determine paths for the robot arm to execute the blending action and a laser scan on a selected surface. A new ROS package was implemented to collect CPU, memory and bandwidth usage for each significant ROS node in this distributed application. To emulate a scenario in which computing resources at a remote datacenter can be used for the segmentation and path planning algorithms in conjunction with the robots located on a factory floor, a software-defined network testbed called GENI was used to distribute compute-heavy ROS nodes. Measurements show that with TCP tuning, and high-speed end-to-end paths, the total execution time in the Cloud scenario can be reasonably close to a local scenario in which computing is collocated with the robot.