GTI 5G and Cloud Robotics White Paper 1 http://www.gtigroup.org Confidentiality: This document may contain information that is confidential and access to this document is restricted to the persons listed in the Confidential Level. This document may not be used, disclosed or reproduced, in whole or in part, without the prior written authorization of GTI, and those so authorized may only use this document for the purpose consistent with the authorization. GTI disclaims any liability for the accuracy or completeness or timeliness of the information contained in this document. The information contained in this document may be subject to change without prior notice. Document History Date DD-MM-YYYY Meeting # Version # Revision Contents NA DD-MM-YYYY DD-MM-YYYY DD-MM-YYYY DD-MM-YYYY 3 4 Cloud Robotics: Trends, Technologies, Communications Abstract Cloud robots are controlled from a “brain” in the cloud. The brain, located in a data center, makes use of Artificial Intelligence and other advanced software technologies to deal with tasks that in traditional robots were undertaken by a local, on-board controller. Compared to local robots, cloud robots will generate new value chains, new technologies, new architectures, new experiences and new business models, this white paper will explore these aspects. 1. Introduction Cloud robotics is a relatively recent concept. Early work dates back to 2010, when the European Commission’s i RoboEarth project began. This aimed to establish a “World Wide Web for robots”. RoboEarth and later ii iii projects such as Rapyuta and Robohow formalized the basic concept and technologies, and are still influencing cloud robotic research today. There are three core advantages of cloud robots compared to stand-alone robots: Information sharing Many cloud robots can be controlled from one brain, and the brain can accumulate visual, verbal, and environmental data from all connected robots. Intelligence derived from this data can be used by all the robots controlled by the brain. As with other cloud services, information collected and processed on each robot will always be up-todate and backed-up safely. Developers also benefit, as they can build reusable solutions for all cloud-connected robots. Offloaded computation Some robot tasks require more computational power than a local controller can economically deliver. Offloading to the cloud data-intensive tasks such as voice and image recognition, voice generation, environmental mapping and motion planning will lower the hardware requirements and power consumption of robots, making them lighter, smaller, and cheaper. Collaboration Cloud robots do not need to work alone. Using the cloud as a common medium, two robots can work together to carry an object too heavy for one, or a group of simple worker robots can work with a local map, provided by a leader robot with costly sensors. 5 Figure 1: Large scale data collection with an array of robots (14 robots are sharing experiences of machine learning for grasping ) [Source: https://research.googleblog.com/2016/03/deep-learning-for-robots-learning-from.html] x Distributed version of AlphaGo exploited 40 search threads, 1202 CPUs and 176 GPUs , no ordinary robot can install inside. But cloud robot can make use of it. 2. Applications for cloud robots Using cloud resources empowers robots and gives them new capabilities in many areas: Intelligent visual processing: image classification, target detection, image segmentation, image description, character recognition. Natural language processing: semantic understanding based on depth learning, accurate identification of user intent, multiintention analysis, emotional analysis. Makes use of a powerful background knowledge base. Facial recognition: face detection algorithm based on depth learning; In the real-time video stream to accurately detect the face; Any face mask and real-time detection under the viewing angle; To overcome: the side face, half obscured, blurred face; Extension from current robot applications: outdoor map navigation, indoor positioning and navigation, typical product identification, universal item identification, environmental understanding, text reading, voice prompts. 6 The applications that will emerge for cloud robots are of many kinds; some are emerging now – others are at an early stage of development. Logistics Amazon, Jingdong, S.F. Express and other companies have deployed logistics robot systems. The wheele