Fog Robotics: A Summary, Challenges and Future Scope
Siva Leela Krishna Chand Gudi, Benjamin Johnston, Mary-Anne Williams
22 Industrial Robots
M1htary
Robots / Gateway
Agricultural
Robots
Rehabilation
Robots Social Robots
Fig. I. Architecture of
Fog
Robotics f11]Our main goal is to provide a summary of Fog
Robotics by demonstrating the importance and checking the feasibility for wide adoption of the Fog
Robotics.
Coming to the paper,Section TI shows the need for a unique Fog
Robotics fieldinstead of considering it as Fog Computing based robotics along with a comparison of applications. Next, we describethe architectures and a Rescue Robots scenario in Section III.
Later in Section
IV, we discuss the evaluation setup and results of the architectures. Subsequently, we present the advantages, challenges and future scope of Fog
Robotics.
II.
WHY
FOG ROBOTICS?
The necessity of Fog
Robotics instead of Cloud Robotics is clearly demonstrated by Gudi et al., with a comparison in between them [11]. So, in this section, we discuss on why there is a need for specific field dubbed Fog
Robotics insteadof considering it as Fog Computing based robotics with respect to applications. Reasons of a robot include but not limited to the working conditions in near real-time environments and the computing power. For instance, if loT devices are utilising Fog Computing then the equipment considered can be of the low
TABLE I
COMPARISON
BETWEEN APPLICATIONS OF Foo ROBOTICS
AND
Foo
COMPUTINGParameters
Fog
Robotics
Fog ComputingStorage
High
Low
Storage Type
Transient Transient
Location
Distributed Distributed
Response
Time Milliseconds
Milliseconds
Topology Mostly one hop Mostly one hop
Coverage
Local Local
Security
Protocols
Specific Specific
CPU/Number of Cores
High
LowNumber of Tasks
High
Low
Power
Consumption
High
Low
GPU
High
Low
Latency/Jitter Unacceptable AcceptableMobility
Unstable
StableReal-time Interaction
Highly
Required
Less-likely Required
Bandwidth
High
Low
Data Transfer Rate
High
Low specification whereas for robotics, the requirement should be high due to the massive usage of Al/ML algorithms [22] [23] [24].
However,
Fog
Robotics shares some of the characteristics Gateway
Gateway / D20 - Robot
Robol Robol Robol 2
Fig. Case A) Basic FR Architecture,
Case B) FR Architecture with D2D
Communication,
Case C) FR Architecture with
Multiple
Fog
Robot
Servers [ 11] of Fog Computing such as the deployment of fog servers and low latency communication. In addition, type of storage, response time, topology, coverage and security protocols are based on the concepts of Fog Computing [25] [26].
Another issue that is concerned about Fog
Robotics (FR) system is the need for a high amount of storage with more number of CPU/GPU cores due to its data usage. In contrast, general applications of Fog Computing (FC) does not require such high computing requirements.
Also, most robots that are currently available in the market comprise of low-level
GPU power and it can limit their potential capabilities.
Therefore, robots can utilise GPU of the FR system for most of its tasks while it is impossible for FC. For better understanding, we listed the main differences and similarities between Fog
Computing and
Fog
Robotics in Table: 1.
Further, the working process of FR consumes a large amount of energy. On the other hand, FC uses low power as it manages mostly the deviceswhich consume less energy [27]. Besides, latency is unac ceptable in FR systems as robots need to perform their tasks concurrently and always assumed as hard real-time systems. This requires the need for higher bandwidth because there will be a rapid data transfer between robots and FR system. Conversely, latency is somewhat acceptable when there are less amount of real-time interactions for
FC.
Therefore, FC can manage with low bandwidth and data transfer rate. Moreoverrobots are mobile, moving around from one place to another place to finish their assigned tasks. It makes FR system to havehandovers in between FRS and robots. Unlike the situation of robot movements, FC applications are most likely to be immobile, staying at one point while performing its task. Thus, due to specific needs and standards required by Fog
Robotics, they are not in a position to use the exist ing infrastructure of Fog Computing.
Also, based upon the above comparison of Fog
Robotics and the
Fog Computing applications perspective, we believe that a specific field of Fog
Robotics is essential rather than
Fog
Computing basedrobotics. Ill. FOG ROBOTICS
ARCHITECTURES
Earlier,
Fog
Robotics architectures (Fig. are proposed by Gudi et al., [ 11] and in this section, we provide a snapshot of the three different types of models. In each and every architecture, fog robot server (FRS) provides information to the robots and enquires the cloud based on the unavailability of data. For Case A and Case B architecture, only one fog robot server can be used with multiple robots whereas CaseC can utilise multiple fog robot servers.
Possessing several
FRS, robots can receive information from the adjacent
FRSwhen needed.
By comparing all of the models, only
Case B can have an advantage of edge processing using Device to Device communication (D2D). It can be applicable based on the distance between the robots. If the distance is short then they can use D2D or else an FRS will be utilised.
Based on the necessity and the area of usage such as homes, hotels, airports or parks, any of the three architectures can be employed. If there is a rise in traffic for a particular area then a sub fog robot server can be introduced. To further examine the importance of Fog
Robotics, we explore a realistic scenario in the upcoming section.Rescue Robots ScenarioHuman-robot interaction can be mostly observable during the robot collaboration and communication either with humans or other robot peers [28] [29] [30]. Therefore, we consider a realistic scenario where rescue robots seek to collaborate with others during a fire mishap assisting fire brigades. For instance, if a fire accident happens, there are various problems that firefighters can encounter. They include visibility issues due to smoke, lack of blueprint of the fire affected zone, stress, health hazards from proximity such as victims being idle due to a panic attack and unconsciousness [31) [32). Also, firefighters need to search room by room to ensure everyone is evacuated. This time-consuming process may hinder firefighter saving lives efficiently.
Therefore, robots can be utilised for assisting firefighters to work more efficiently such as detecting victims beforehand or to map • • • • • • • • • • • • • • • • • • • • Short
Range •• AWS Server
FRS,
Sydney
Pepper
Robot
Fig. Architecture of
Fog
Robotics w.r.t
AWS and
Pepper
Robot
India (Mumbai)
Max • Avg • Min ------------------·10328 .---···---------------1116.9 ??-------·405.8
Singapor Austral .. {Sydn,,y) South {Seoul) South America (Sao Paulo) USA {Ohio) Canada {Central) i c: Germ.Joy {Frankfurt) ? :,_g United Krngdom (London) Laten?ms) 800 1000 1100 Fig. Latency: AWS w.r.t Robot. architectures are discussed by Gudi et al., [l J ]. They claim that FR performs better than Cloud Robotics (CR) using an assumption of latency value. Therefore in this section, we discuss the advanced analysis of results using real-time latency. For examining the FR architectures, Jet us consider a social robot Pepper [45], a Fog robot server (FRS) and the Cloud. Firstly, a scenario of FR is considered as shown in Fig. 3. For Cloud, we considered Amazon Web Services (AWS) servers [46] from various locations of the world. They includeAustralia (Sydney), South Korea (Seoul), Singapore, India (Mumbai), Germany (Frankfurt), United Kingdom (London), South America (Sao Paulo), United States of America (Ohio) and Canada (Central) while a local server is regarded as FRS. Usually for any kind of scenario, mostly robots exchange information such as images, maps and analysis of speech. Thisis processed in the form of data either with fog robot server or the cloud. Thus, we considered sending packets of data for calculating the latency. For better understanding, latency is tested with the help of the Pepper robot and are plotted as shown in Fig. 4. Based on the obtained latency results of cloud sever concerning the robot, we can observe different latency values across various countries. For validation purpose, only the highest, average and lowest latency of particular countries are considered. As an average after several attempts, we can see that South America (Sao Paulo) has the highest latency of maximum l an average of andminimum of 390.16ms. On the other hand, lowest latency is seen at Australia (Sydney) with as maximum, as average and 32.19ms of minimum. Alongside, a median of latency is observed at South Korea (Seoul) with as maximum, 26 L.85 as average and 246.76 as a minimum. For performance evaluation of FR, an iFogsim toolkit [47] is chosen for predicting the latency with various conditions. Further, the impact of latency on the proposed three • D2D • Fog • Cloud (Sao Paulo) • Cloud (Seoul) • Cloud (Sydney) I 525 1001 I I I -- -- Number of Robot? I 387 8S8 ]00800400 Fig. 5. Results of Architecture A/B Scenarios • Fog • Cloud (Sao Paulo) • Cloud (Seoul) • Cloud (Sydney) zo Number of Fog Robot Servers •• ? CC1I -3000 .,, E Fig. 6. Results of Architecture C Scenario architectures are as shown below. A. FR Architectures(AIB) For evaluating the Fog Robotics (FR) architectures (A/B), we chose to validate the latency with a variation of 1-5robots. These robots send packets of data to Fog Robot Server and the Cloud. Upon measuring the latency w.r.t architecture description, we can say that the latency of FR raised from to I hiking to ms. Also, a sudden spike for FR shows tolerance by a rise in the usage of robots. WhileD2D has an increase in latency from 3.82ms to with an internal lag at Coming to the Cloud, there is a smallrise of latency from 1085.7 lms to Sao Paulo, to Seoul, to n the internet of things, edge and fog computing environments,"Software: Practice and Experience, vol. no. pp. [48] R. M. I. Mykhalchychyna and L. Buchhold, "Bring me a beer from the fridge: Object segmentation and manipulation on a domestic servicerobot extended with a nvidia jetson tx2." [49] S. L. K. C. Gudi, S. Ojha, B. Johnston, M.-A. Williams et al., "A proactive robot tutor based on emotional intelligence," in Interna tional Conference on Robot intelligence Technology and Applications. Springer, pp. [50] J. W. DeCew, In pursuit of privacy: Law, ethics, and the rise oftechnology. Cornell University Press, [51] A. N. Joinson, U.-D. Reips, T. Buchanan, and C. 8. P. Schofield, "Pri vacy, trust, and self-disclosure online," Human-Computer lnteraction, vol. no. I, pp. [52] D.S. Syrdal, M. L. Walters, N. Otero, K. L. Koay, and K. Dautenhahn, "He knows when you are sleeping-privacy and the personal robot companion," in Proc. Workshop Human Implications of Human-Robot Interaction, Association for the Advancement of Artificial Intelligence (AAA!aAZ07), pp. [53] M. R. Calo, "12 robots and privacy," Robot ethics: The ethical and social implications of robotics, p. l. [54] J. Langston, "Uw researchers hack a teleoperated surgical robot to reveal security flaws," UW Today, [55) T. Denning, C. Matuszek, K. Koscher, J. R. Smith, and T. Kohno, "A spotlight on security and privacy risks with future household robots:auacks and lessons," in Proceedings of the l l th international conference on Ubiquitous computing. ACM, 2009, pp. 105-114. Benjamin Johnston is a Senior Lecturer in the Faculty of Engineering and Information Technology at the University of Technology Sydney. He con ducts research in commonsense reasoning and social robotics, with a particular interest in using Al to cre ate natural, fluent human-robot interactions. He alsoteaches classes on social robotics, entrepreneurship and enterprise software development. Siva Leela Krishna Chand Gudi is a research scholar focusing on Social Robotics at the Magic Lab within Center for Artificial Intelligence of the University of Technology Sydney CUTS), Australia.He is recognized as among the top 200 of the most qualified young researchers globally to attend the prestigious Heidelberg Laureate Forum. He also got invited to Global Solutions Summit which pro vides policy recommendations to G20ff20 summitsand to deliver talks at top-notch venues from MIT CSAIL/Sloan, USA to an audience of Nobel Laure ates, Germany. Besides, he worked for various research organizations which include the Commonwealth Bank of Australia (CBA), WENS Lab, Indian Space Research Organization (ISRO), Defence Research and Development Organization (DRDO) along with several patents and publications. Also, he served as a chair/keynote speaker/reviewer for conferences/journals namely RSS IEEE Access and few more. ln addition, he won the world's reputed competitions, various scholarships as well as fellowships including from the Australian Academy of Science. Mary-Anne Williams is Director of the Innovationand Enterprise Research Laboratory (The Magic Lab) at University of Technology Sydney. Mary Anne has a Masters of Laws and a Ph.D. in Knowl edge Representation and Reasoning with transdisci plinary strengths in AJ, disruptive innovation, design thinking, data analytics, JP law and privacy law. Mary Anne is a Faculty Fellow at Stanford University and a Guest Professor at the University of Science and Technology China where she gives intensive courses on disruptive innovation. Her current re search mainly focuses on social robotics while covering wide range of disciplines like belief, perception and risk assessment in roboticrobotic