A Microservices-based for Distributed Deep Neural Network of Delta Robot Control System

Abstract

Recent advances in robotics, enable the evolution of the manufacturing industry to adapt to the new environments of the industry itself. A distributed deep neural network (DDNN) has been improved in cloud services as a distributed robot control system that can be accessed over protocols such as HTTP, TCP, LwM2M, etc. The conventional deep neural networks (DNN) method has complex computational matrices that will be affected the training accuracy, training runtime, and so on. Microservices-based DNN is deployed to reduce the training workloads of deep neural networks in a single node that can be improved by Containerized DNN architecture. This paper describes a microservices-based for the Deep Neural Network as a Services (DNNaaS) that has been implemented to encourage an inverse kinematic solution of the delta robot control system. Each container trained with non-identical inverse kinematic motions data of delta robot within the data length used is 200 motions data for each node. The proposed method was trained in containers C1 until C4. The training process of containerized DNN performed with accuracy equals to 95.76% and loss equals to 1.06% in containers C1 until C4 for 13.7279 seconds. The proposed method was tested to transmitted 100 IK motions data over Socket.IO with 1.7kB in 336 milliseconds.