Chapter 1: Cloud robotics

Cloud robotics is a subfield of robotics that focuses on the advantages of converged infrastructure and shared services for robotics. Cloud robotics is an attempt to make use of cloud technologies, such as cloud computing, cloud storage, and other Internet technologies. When connected to the cloud, robots are able to take advantage of the strong compute, storage, and communication resources of current data centers in the cloud. These centers are able to process and distribute information from a variety of robots or agents, including other machines, smart objects, humans, and so on. Through the use of networks, humans are also able to outsource duties to robots operating remotely. Cloud computing technologies make it possible for robot systems to be afforded tremendous capabilities while simultaneously lowering expenses through the utilization of cloud technology. As a result, it is feasible to construct lower-cost, more intelligent robots that are lightweight and have an intelligent "brain" that is stored in the cloud. The "brain" is made up of factors such as the data center, the knowledge base, the task planners, deep learning, information processing, environment models, communication support, and other similar components.

When it comes to robots, a cloud could theoretically consist of at least six important components:

A strategy known as "lifelong learning." CAS has proposed the utilization of lifelong learning in order to construct a cloud for robots to use as a brain. The dilemma of how to make robots combine and transmit their experience in order to enable them to make good use of existing knowledge and quickly adapt to new situations was the impetus for the author's work. The authors propose a learning architecture for navigation in cloud robotic systems that they call Lifelong Federated Reinforcement Learning (LFRL). This architecture is intended to address the issue. on their study, they present an algorithm for knowledge fusion that may be used to improve a shared model that is put into operation on the cloud. Transfer learning strategies that are proven to be effective in LFRL are then presented. Both LFRL and human cognitive science are compatible with one another, and it works well with cloud robotic systems. For the purpose of robot navigation, experiments have demonstrated that LFRL significantly enhances the effectiveness of reinforcement learning. In addition, the implementation of the cloud robotic system demonstrates that LFRL is able to combine previously acquired information.

The method is known as Federated Learning. In the year 2020, it was predicted that robots may be equipped with a cloud brain by utilizing lifetime learning. Learning a new behavior is something that humans are capable of doing by studying how other people practice the skill. Robots, on the other hand, are also capable of doing this through imitation learning. Moreover, if humans are provided with direction from outside sources, they will be able to grasp the new habit more effectively. Thus, how are robots able to accomplish this? The authors provide a novel framework that they have identified as FIL in order to address the issue. For cloud robotic systems, it offers a framework for the merging of heterogeneous knowledge from several sources. After that, a knowledge fusion algorithm in FIL is suggested as a solution. Consequently, it makes it possible for the cloud to combine the diverse expertise of local robots and to produce guide models for robots that are in possession of service requests. Following that, we present a knowledge transfer mechanism that will make it easier for local robots to acquire knowledge from the cloud. With the help of FIL, a robot is able to make use of the information that it has learned from other robots in order to improve the accuracy and efficiency of its imitation learning. When compared to transfer learning and meta-learning, FIL is a more acceptable option for deployment in cloud robotic systems. They are doing research on a task that requires robots (cars) to drive themselves. The results of the experiments show that the shared model that is developed by FIL brings about a boost in the efficiency of imitation learning for local robots that are part of cloud robotic services.

Peer-assisted learning is the approach chosen. It was proposed by UM that a cloud brain for robots could be constructed through the utilization of peer-assisted learning. With the introduction of data-driven deep learning technologies, the area of robotics is currently undergoing a groundbreaking technological revolution. Nevertheless, the process of constructing datasets for each local robot is cumbersome. On the other hand, data islands existing between local robots prevent data from being exploited in a collaborative manner. The work provides Peer-Assisted Robotic Learning (PARL) in robotics, which is inspired by the concept of peer-assisted learning in cognitive psychology and pedagogy. The purpose of this work is to address the issue that has been presented. Within the framework of cloud robotic systems, PARL is responsible for implementing data collaboration. After completing their training locally and doing semantic computing, robots then upload both their data and their models to the cloud. In addition to performing augmentation, integration, and transfer, the cloud is responsible for bringing together the data. Last but not least, perfect this larger shared dataset in the cloud so that it can be used by local robots. For the purpose of putting the data processing in PARL into action, we also suggest using the DAT Network, which stands for the Data Augmentation and Transferring Network. The augmentation of data from several local robots is something that can be accomplished using DAT Network. Experiments are been carried out by the authors on a simplified version of the self-driving challenge for robots (cars). Within the realm of self-driving scenarios, DAT Network has made a considerable advancement in terms of the enhancement. In addition to this, the findings of the experiments on self-driving vehicles also show that PARL is capable of enhancing learning effects through the collaboration of data from local robots.

RoboEarth was given funding by the Seventh Framework Programme for research and technical development initiatives implemented by the European Union. The purpose of this funding was to especially investigate the topic of cloud robotics. The purpose of RoboEarth is to make it possible for robotic systems to gain knowledge from the experiences of other robots. This will pave the way for rapid advancements in machine cognition and behavior, and ultimately, for human-machine interaction that is more nuanced and intelligent. A Cloud Robotics infrastructure is something that RoboEarth provides. The database that RoboEarth built in the style of the World Wide Web includes information that was created by humans as well as robots in a format that is readable by machines. Software components, maps for navigation (such as object locations and world models), task knowledge (such as action recipes and manipulation methods), and object recognition models (such as photos and object models) are all examples of the types of data that are stored in the RoboEarth knowledge base. There is support for mobile robots, autonomous vehicles, and drones within the RoboEarth Cloud Engine. These types of devices demand a significant amount of processing in order to navigate.

An open-source cloud robotics framework called Rapyuta was developed by a robotics researcher at ETHZ. It is built on RoboEarth Engine and was created by the researcher. In the framework, every robot that is connected to Rapyuta has the potential to have a protected computing environment, which is represented by rectangular boxes. This provides them with the opportunity to relocate their heavy computations to the cloud. Additionally, the computing environments are highly networked with one another and have a connection to the RoboEarth knowledge repository that is capable of delivering a tremendous amount of bandwidth.

One of the extensions of the RoboEarth project is known as KnowRob. It is a knowledge processing system that combines methods for knowledge representation and reasoning with techniques for acquiring knowledge and for anchoring the knowledge in a physical system. Additionally, it has the capability of serving as a common semantic framework for the integration of information from a variety of sources.

RoboBrain is a large-scale computational system that acquires knowledge through the use of resources that are accessible to the general public on the Internet, computer simulations, and live robot experiments. It compiles all of the information pertaining to robotics into a knowledge base that is both comprehensive and interconnected. Prototyping for robotics research, constructing robots for the home, and developing autonomous vehicles are all examples of applications. The objective of the project is as straightforward as its name suggests: to bring about the creation of a centralised, always-online brain that robots can access. Both Stanford University and Cornell University are doing the majority of the work on the project. In addition, the National Science Foundation, the Office of Naval Research, the Army Research Office, Google, Microsoft, Qualcomm, the Alfred P. Sloan Foundation, and the National Robotics Initiative are all contributing to the project. The National Robotics Initiative's objective is to advance robotics in order to assist the United States in becoming more competitive in the global economy.

The Internet can be connected to robots and other intelligent devices through the use of a service called MyRobots. It is possible to think of it as a social network for robots and other intelligent objects...