Sharron LIU, Yu YAN (Intel)
Convolutional Neural Network (CNN) based Artificial Intelligent methods are widely used in intelligent robot. However, the CNN inference consumes large amount of CPU loads and introduces unacceptable latency. Existing acceleration solutions require specific knowledge on CNN frameworks, acceleration libraries, and parallel programming. We introduced OpenVINO™ for CNN inference acceleration. We proved OpenVINO™ acceleration for industrial robot firstly, by establishing a platform solution consists of the OpenVINO™ Grasp Library and the MoveIt based intelligent visual grasp application running on real robot. OpenVINO™ supports self-converted or 100+ pretrained CNN models – using common APIs for all devices.
Video presented at ROSCon 2019. More details and slides are available at: roscon.ros.org/2019/.
