Cloud Robot Vision Services Extend High-Performance Computing Capabilities of Robot Systems

The paper describes a computational solution with cloud implementing, extending robot-vision capabilities of real-time multiple articulated objects recognition for on-the-fly robot grasping. Articulated objects are recognized by matching the unknown object's skeleton computed from the input image in a cloud virtual machine (VM) with a set of learned skeleton signatures. This High Performance Computing (HPC) process represents a powerful capability for qualitative shape matching because it unambiguously synthesizes and helps estimating the topology of the object and its shape. The skeleton-based matching process is performed as an application-driven robotic service in a private cloud, ten times faster than the robot controller is able to do it and nearly twice faster than two PC-type robot terminals for multiple parts moving on conveyor belts. The parameters of the virtualization process and experimental results which confirm the solution are presented.