398

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 2, APRIL 2015

A Survey of Research on Cloud Robotics

and Automation
Ben Kehoe, Student Member, IEEE, Sachin Patil, Member, IEEE, Pieter Abbeel, Senior Member, IEEE, and
Ken Goldberg, Fellow, IEEE

AbstractThe Cloud infrastructure and its extensive set of Internet-accessible resources has potential to provide signicant benets to robots and automation systems. We consider robots and
automation systems that rely on data or code from a network to
support their operation, i.e., where not all sensing, computation,
and memory is integrated into a standalone system. This survey
is organized around four potential benets of the Cloud: 1) Big
Data: access to libraries of images, maps, trajectories, and descriptive data; 2) Cloud Computing: access to parallel grid computing
on demand for statistical analysis, learning, and motion planning;
3) Collective Robot Learning: robots sharing trajectories, control
policies, and outcomes; and 4) Human Computation: use of crowdsourcing to tap human skills for analyzing images and video, classication, learning, and error recovery. The Cloud can also improve
robots and automation systems by providing access to: a) datasets,
publications, models, benchmarks, and simulation tools; b) open
competitions for designs and systems; and c) open-source software.
This survey includes over 150 references on results and open challenges. A website with new developments and updates is available
at: http://goldberg.berkeley.edu/cloud-robotics/
Note to PractitionersMost robots and automation systems still
operate independently using onboard computation, memory, and
programming. Emerging advances and the increasing availability
of networking in the Cloud suggests new approaches where
processing is performed remotely with access to dynamic global
datasets to support a range of functions. This paper surveys
research to date.
Index TermsBig data, cloud automation, cloud computing,
cloud robotics, crowdsourcing, open source.

I. INTRODUCTION

S illustrated in Fig. 1, the Cloud has potential to enhance

a broad range of robots and automation systems. The
National Institute of Standards and Technology (NIST) denes
the Cloud as a model for enabling ubiquitous, convenient,
on-demand network access to a shared pool of configurable
resources (e.g., servers, storage, networks, applications, and
Manuscript received March 15, 2014; revised September 15, 2014; accepted
November 06, 2014. Date of publication January 12, 2015; date of current version April 03, 2015. This paper was recommended for publication by Associate
Editor S. Sarma and Editor S. Sarma upon evaluation of the reviewers' comments. This work was supported in part by Google and Cisco and by the U.S.
National Science Foundation under Award IIS-1227536: Multilateral Manipulation by Human-Robot Collaborative Systems.
The authors are with the University of California, Berkeley, CA
94720 USA (e-mail: benk@berkeley.edu; sachinpatil@berkeley.edu;
pabbeel@berkeley.edu; goldberg@berkeley.edu).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TASE.2014.2376492

services) that can be rapidly provisioned and released with

minimal management effort or service provider interaction
[115]. An example is the online word processing capabilities
offered by Google Docs. One can send Microsoft Word documents over the Internet, but Google Docs differs in that the
document and software does not reside locally. The data and
code is stored in the Cloud using remote server farms with
shared processors and memory. This is helpful because one
does not have to worry about maintenance, outages, and software or hardware updates. The Cloud also provides economies
of scale and facilitates sharing of data across applications and
users [122].
Cloud Robot and Automation systems can be broadly dened as follows: Any robot or automation system that relies
on either data or code from a network to support its operation, i.e., where not all sensing, computation, and memory is
integrated into a single standalone system. This denition is intended to include future systems and many existing systems that
involve networked teleoperation or networked groups of mobile robots such as UAVs [100], [116] or warehouse robots [44],
[96] as well as advanced assembly lines, processing plants, and
home automation systems, and systems with computation performed by humans [132], [155]. Due to network latency, variable quality of service, and downtime, Cloud Robot and Automation systems often include some capacity for local processing for low-latency responses and during periods where network access is unavailable or unreliable. This is not a binary
denition; there are degrees to which any system will t under
this denition.
The Google self-driving car exemplies the idea. It indexes
maps and images collected and updated by satellite, Streetview,
and crowdsourcing from the Cloud to facilitate accurate localization. Another example is the Kiva Systems pallet robot for
warehouse logistics. These robots communicate wirelessly with
a local central server to coordinate routing and share updates on
detected changes in the environment.
In 2010, James Kuffner coined the term Cloud Robotics
and described a number of potential benets [98]. An article in
IEEE Spectrum quickly followed [70] and Steve Cousins summarized the concept as No robot is an island. The next section
considers the history of this important idea.
This survey is organized around four potential benets from
the Cloud: 1) Big Data: access to remote libraries of images,
maps, trajectories, and object data; 2) Cloud Computing: access
to parallel grid computing on demand for statistical analysis,
learning, and motion planning; 3) Collective Robot Learning:
robots sharing trajectories, control policies, and outcomes; and

1545-5955 2014 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/
redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

KEHOE et al.: A SURVEY OF RESEARCH ON CLOUD ROBOTICS AND AUTOMATION

399

Fig. 1. The Cloud has potential to enable a new generation of robots and automation systems to use wireless networking, Big Data, Cloud Computing, statistical
machine learning, open-source, and other shared resources to improve performance in a wide variety of applications including assembly, inspection, driving,
warehouse logistics, caregiving, package delivery, housekeeping, and surgery.

4) Human computation: using crowdsourcing access to remote

human expertise for analyzing images, classication, learning,
and error recovery. This survey also cites examples where the
Cloud can enhance robotics and automation systems by facilitating access to: a) datasets, publications, models, benchmarks,
and simulation tools; b) open competitions for designs and systems; and c) open-source software.
II. A BRIEF HISTORY
The value of networking to connect machines in manufacturing automation systems was recognized over 30 years ago.
In the 1980s, General Motors developed the Manufacturing Automation Protocol (MAP) [80]. A diverse set of incompatible
proprietary protocols were offered by vendors until a shift began
in the early 1990s when the World Wide Web popularized the
HTTP over IP protocols [121].
In 1994, the rst industrial robot was connected to the Web
with an intuitive graphical user interface that allowed visitors
to teleoperate the robot via any Internet browser [63]. In the
mid and late 1990s, researchers developed a series of web interfaces to robots and devices to explore issues such as user interfaces and robustness [64], [65] that initiated the subeld of
Networked Robotics [66], [112].
In 1997, work by Inaba et al. on remote brained robots described the advantages of remote computing for robot control
[79].
In May 2001, the IEEE Robotics and Automation Society
established the Technical Committee on Networked Robots
[10] which organized a number of workshops. Chapters of the
Springer Handbook on Robotics were focused on Networked
Tele-robots (where robots are operated remotely by humans
using global networks) and Networked Robots (where robots
communicate with each other using local networks), respectively [99], [143].
In 2009, the RoboEarth project was announced. It envisioned
a World Wide Web for robots: a giant network and database
repository where robots can share information and learn from
each other about their behavior and environment [22], [156]

Fig. 2. The RoboEarth systems architecture designed to allow robots to share

data and learn from each other [22], [156]. (Image reproduced with permission
from authors.)

as illustrated in Fig. 2. Under a major European Union grant,

the RoboEarch research team developed a series of system architectures for service robotics [30], [53], developing Cloud
networking [74], [87], and computing resources [78] to generate 3D models of environments, speech recognition, and face
recognition [148].
As noted in the previous section, James Kuffner introduced
the term Cloud Robotics in 2010. This broader term supplanted earlier terminology and has been adopted by many
researchers including the organizers of this Special Issue of
the IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND
ENGINEERING.
Cloud Robotics and Automation is related to several other
new initiatives. The Internet of Things [32], a term also introduced in 2010, describes how RFID and inexpensive processors

400

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 2, APRIL 2015

Fig. 3. Data can be collected from many sources as shown in this schematic architecture for the Mobile Millennium, a Cloud-based transportation system that
combines streaming data from taxis, maps, and road-based sensors [77]. Mobile Millennium uses the Big Data and Collective Robot Learning aspects of Cloud
Robotics and Automation. (Image reproduced with permission from authors.)

could be incorporated into a vast array of robots and physical

objects from inventory items to household appliances [110] to
allow them to communicate and share information.
The term Industry 4.0, introduced in Germany in 2011,
predicts a fourth industrial revolution that will use networking
to follow the rst (mechanization of production using water
and steam power), the second (mass production with electric
power), and the third (use of electronics to automate production) industrial revolutions [11].
In 2012, General Electric introduced the term Industrial Internet, to describe new efforts where industrial equipment such
as wind turbines, jet engines, and MRI machines connect over
networks to share data and processing for industries including
energy, transportation, and healthcare [55], [93]. For example,
GE is using sensor readings from aircraft engines to optimize
fuel consumption under a myriad of conditions [59]. The power
of Big Data and Cloud Computing are extensively being used
by many industries [25], [113], and are being harnessed to optimize water usage for irrigation [52].
Many related projects are emerging. In August 2014,
Ashutosh Saxena announced his goal to build a RoboBrain:
a large-scale computational system that learns from publicly
available Internet resources, computer simulations, and real-life
robot trials.
III. BIG DATA
The Cloud can provide robots and automation systems with
access to vast resources of data that are not possible to maintain
in onboard memory. Big Data describes data that exceeds
the processing capacity of conventional database systems [54]
including images, video, maps, real-time network and nancial
transactions [102], and vast networks of sensors [159].
A recent U.S. National Academy of Engineering Report summarizes many research opportunities and challenges created by
Big Data [41] and other challenges are summarized in [29] and
[165]. For example, sampling algorithms can provide reasonable approximations to queries on large datasets to keep run-

ning times manageable [38], but these approximations can be

seriously affected by dirty data [58].
Hunter et al. [77] presents algorithms for a Cloud-based transportation system called Mobile Millennium, which uses the GPS
in cellular phones to gather trafc information, process it, and
distribute it and also to collect and share data about noise levels
and air quality (see Fig. 3).
Large datasets can facilitate machine learning, as has been
demonstrated in the context of computer vision. Large-scale
image datasets such as ImageNet [50], PASCAL visual object
classes dataset [56], and others [142], [151] have been used for
object and scene recognition. By leveraging Trimble's SketchUp
3D warehouse, Lai et al. reduced the need for manually labeled training data [101]. Using community photo collections,
Gammeter et al. created an augmented reality application with
processing in the Cloud [57]. Combining internet images with
querying a local human operator, Hidago-Pena et al. provided
a more robust object learning technique [72]. Deep learning is
a technique using many-layered neural networks that can take
advantage of Big Data [49], and has been used for computer vision [97], [140] and grasping [104].
Grasping is a persistent challenge in robotics: determining
the optimal way to grasp a newly encountered object. Cloud
resources can facilitate incremental learning of grasp strategies
[40], [120] by matching sensor data against 3D CAD models in
an online database. Examples of sensor data include 2D image
features [75], 3D features [68], and 3D point clouds [39], as well
as demonstrations [94].
Google Goggles [9], a free image recognition service for mobile devices (see Fig. 4), has been incorporated into a Cloudbased system for robot grasping [89], as illustrated in Fig. 5.
The RoboEarth project stores data related to objects and maps
for applications ranging from object recognition to mobile navigation to grasping and manipulation (see Fig. 2) [156]. The Columbia Grasp dataset [67], the MIT KIT object dataset [88], and
the Willow Garage Household Objects Database [40] are available online and have been used to evaluate different aspects of

KEHOE et al.: A SURVEY OF RESEARCH ON CLOUD ROBOTICS AND AUTOMATION

Fig. 4. Google's object recognition system combines an enormous dataset of

images and textual labels with machine learning to facilitate object recognition
in the Cloud [9], [98]. (Image reproduced with permission.)

grasping algorithms, including grasp stability [45], [46], robust

grasping [162], and scene understanding [128]. Dalibard et al.
attach manuals of manipulation tasks to objects [43].
One research challenge is dening cross-platform formats for
representing data. While sensor data such as images and point
clouds have a small number of widely-used formats, even relatively simple data such as trajectories have no common standards yet but research is ongoing [129], [147], [149]. Another
challenge is working with sparse representations for efcient
transmission of data, e.g., algorithms for sparse motion planning for robotic and automation systems [51], [105].
Large datasets collected from distributed sources are often
dirty with erroneous, duplicated, or corrupted data [6], [158],
such as 3D position data collected during robot calibration
[111]. New approaches are required that are robust to dirty data.

IV. CLOUD COMPUTING

Massively-parallel computation on demand is now widely
available [29] from commercial sources such as Amazon's
Elastic Compute Cloud [1], [2], Google's Compute Engine [8],
and Microsoft's Azure [12]. These systems provide access to
tens of thousands of remote processors for short-term computing tasks [105], [106]. These services were originally used
primarily by web application developers but have increasingly
been used in scientic and technical high-performance computing (HPC) applications [20], [86], [114], [152].
Uncertainty in sensing, models, and control is a central issue
in robotics and automation [62]. Such uncertainty can be modeled as perturbations in position, orientation, shape, and control. Cloud Computing is ideal for sample-based Monte-Carlo
analysis. For example, parallel Cloud Computing can be used
to compute the outcomes of the cross-product of many possible
perturbations in object and environment pose, shape, and robot
response to sensors and commands [154]. This idea is being explored in medicine [157] and particle physics [141].
Cloud-based sampling can be used to compute robust grasps
in the presence of shape uncertainty [90][92] (see Fig. 6). This
grasp planning algorithm accepts as input a nominal polygonal
outline with Gaussian uncertainty around each vertex and the
center of mass and uses parallel-sampling to compute a grasp
quality metric based on a lower bound on the probability of
achieving force closure.

401

Cloud Computing has potential to speed up many computationally-intensive robotics and automation systems applications
such as robot navigation by performing SLAM in the Cloud
[134], [135] as illustrated in Fig. 7 and next-view planning
for object recognition [125]. Cloud-based formation control of
ground robots has also been demonstrated [153].
For optimal sampling-based motion planning methods such
as RRT*, Cloud Computing is useful to generate the graphs; it is
also important to recognize that these graphs can grow rapidly
so algorithms for graph reduction are needed to facilitate data
transfer, as illustrated in Fig. 8.
The Cloud also facilitates video and image analysis [123],
[137], and mapping [119], [136] (see Fig. 7. Image processing in
the Cloud has been used for assistive technology for the visually
impaired [36] and for senior citizens [58].
Bekris et al. [33] propose an architecture for efciently planning the motion of new robot manipulators designed for exible
manufacturing oors in which the computation is split between
the robot and the Cloud.
It is important to acknowledge that the Cloud is prone to
varying network latency and quality of service. Some applications are not time sensitive, such as decluttering a room or precomputing grasp strategies or ofine optimization of machine
scheduling, but many applications have real-time demands [83]
and this is an active area of research [26], [27], [95], [109].
V. COLLECTIVE ROBOT LEARNING
The Cloud facilitates sharing of data for robot learning by
collecting data from many instances of physical trials and environments. For example, robots and automation systems can
share initial and desired conditions, associated control policies
and trajectories, and importantly: data on the resulting performance and outcomes.
The Lightning framework (see Fig. 10), proposes a framework for Collective Robot Learning by indexing trajectories
from many robots over many tasks and using Cloud Computing
for parallel planning and trajectory adjustment [35].
Such systems can also be expanded to global networks to facilitate shared path planning, including trafc routing, as shown
in Fig. 9.
For grasping [37], grasp stability of nger contacts can be
learned from previous grasps on an object [45]. Sharing data
through Collective Robot Learning can also improve the capabilities of robots with limited computational resources [69].
The MyRobots project [13] from RobotShop proposes a social network for robots: In the same way humans benet from
socializing, collaborating and sharing, robots can benet from
those interactions too by sharing their sensor information giving
insight on their perspective of their current state [21].
The RoboEarth and RoboBrain databases in Section III are
designed to be updated with new information from connected
robots. The RoboBrain project learns from publicly available
Internet resources, computer simulations, and real-life robot
trials. [16]
KIVA Systems [44], [96] uses hundreds of mobile platforms
to move pallets in warehouses using a local network to coordinate motion and update tracking data.

402

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 2, APRIL 2015

Fig. 5. System Architecture for Cloud-based object recognition for grasping. The robot captures an image of an object and sends via the network to the Google
object recognition server. The server processes the image and returns data for a set of candidate objects, each with precomputed grasping options. The robot
compares the returned CAD models with the detected point cloud to rene identication and to perform pose estimation, and selects an appropriate grasp. After
the grasp is executed, data on the outcome is used to update models in the Cloud for future reference [89]. This project uses the Big Data, Cloud Computing, and
Collective Robot Learning aspects of Cloud Robotics and Automation. (Image reproduced with permission.)

Fig. 6. A Cloud-based approach to geometric shape uncertainty for grasping.

(Top) Uncertainty in object pose and shape. (Bottom) Computed push grasps.
Kehoe et al. use sampling over uncertainty distributions to nd a lower bound
on the probability of success for grasps [90][92].

VI. HUMAN COMPUTATION: CROWDSOURCING

CALL CENTERS

AND

Human skill, experience, and intuition is being tapped

to solve a number of problems such as image labeling for
computer vision [40], [87], [98], [155], learning associations
between object labels and locations [138], and gathering data
[48]. Amazon's Mechanical Turk is pioneering on-demand
crowdsourcing with a marketplace where tasks that exceed
the capabilities of computers can be performed by human
workers. In contrast to automated telephone reservation systems, consider a future scenario where errors and exceptions
are detected by robots and automation systems which then
contact humans at remote call centers for guidance.
Research projects are exploring how this can be used for path
planning [73], [82], to determine depth layers, image normals,
and symmetry from images [61], and to rene image segmentation [84]. Researchers are working to understand pricing models
[145] and apply crowdsourcing to grasping [144] (see Fig. 12).
Knowledge-based solutions are being explored for industrial automation as well [146].

Fig. 7. A Cloud framework for robot navigation using cooperative tracking and
). Riazuelo et al. demonstrate computer intensive bundle admapping (
justment for navigation using simultaneous localization and mapping (SLAM)
performed in the Cloud [134][136]. (Image reproduced with permission.)

Networked robotics has a long history of allowing robots to

be remotely teleoperated by humans over the web [63], and the
expanded resources of the Cloud enables new research into remote human operation [103], [144], [164] (see Fig. 11).
VII. OPEN-SOURCE AND OPEN-ACCESS
The Cloud supports the evolution of Cloud Robotics and Automation by facilitating human access to: a) datasets, publica-

KEHOE et al.: A SURVEY OF RESEARCH ON CLOUD ROBOTICS AND AUTOMATION

Fig. 8. Distributed sampling-based motion planning. A roadmap of trees for

motion planning in high-dimensional spaces. Plaku et al. show that their planner
can easily solve high-dimensional problems that exhaust resources available to
single machines [126]. (Image reproduced with permission.)

Fig. 9. Schematic architecture of CloudThink. Wilhem et al. developed an

open-standard for self-reporting sensing devices such as sensors mounted in automobiles. Cloud-enabled storage of sensor network data can enable collaborative sharing of data for trafc routing and other applications [163]. CloudThink
uses the Collective Robot Learning aspect of Cloud Robotics and Automation.
(Image reproduced with permission from authors.)

tions, models, benchmarks, and simulation tools; b) open competitions for designs and systems; and c) open-source software.
The success of open source software [42], [124] is now
widely accepted in the robotics and automation community. A
primary example is ROS, the Robot Operating System, which
provides libraries and tools to help software developers create
robot applications [14], [18], [131]. ROS has also been ported
to Android devices [19]. ROS has become a standard akin
to Linux and is now used by almost all robot developers in
research and many in industry, with the ROS Industrial project
created to support these users [17].
Additionally, many simulation libraries for robotics are now
open source, which allows students and researchers to rapidly
set up and adapt new systems and share the resulting software.
There are many open source simulation libraries, including
Bullet [4], a physics simulator originally used for video games,

403

OpenRAVE [15] and Gazebo [7], simulation environments

geared specically towards robotics, OOPSMP, a motion-planning library [127], and GraspIt!, a grasping simulator [117].
The open source nature of these libraries allows them to be
modied to suit applications and they were not originally
designed for.
Another exciting trend is in open source hardware, where
CAD models and the technical details of construction of devices are made freely available [47]. The Arduino project [3] is
a widely-used open source microcontroller platform with many
different sensors and actuators available, and has been used in
many robotics projects. The Raven [71] is an open-architecture laparoscopic surgery robot developed as a research platform an order of magnitude less expensive than commercial surgical robots [23]. Recent advances in 3D printing (also known
as additive manufacturing) are poised to have a major impact
on many elds, including development of open source hardware
designs [60], [81], [108].
The Cloud facilitates open challenges and design competitions that can draw on a diverse and geographically distributed
population of innovators.
The DARPA Robotics Challenge (DRC) is a competition
of robot systems and software teams vying to develop robots
capable of assisting humans in responding to natural and
man-made disasters, supported by NIST and the Southwest
Robotics Institute (SwRI) [24]. The DRC simulator is provided to all contestants through CloudSim, an open-source
Cloud-based simulation platform for testing the performance
of the Atlas humanoid robot (shown in Fig. 13) on a variety
of disaster response tasks [5], [7]. The Cloud permits running
interactive, real-time simulation tasks in parallel for purposes
such as predicting and evaluating performance, validating
design decisions, optimizing designs, and training users [28].
Another example of an open competition is the Ultra-Affordable Educational Robot Challenge organized by the
African Robotics Network with support from the IEEE Robotics
and Automation Society in the Summer of 2012. It attracted
28 designs from around the world including the Grand Prize
winning design shown in Fig. 14, where a modied surplus
Sony game controller uses the vibration motors to drive wheels
and lollipops as inertial counterweights for contact sensing by
the thumb switches. This robot can be built from surplus parts
for US $8.96 [150].
VIII. CHALLENGES AND FUTURE DIRECTIONS
Using the Cloud for robotics and automation systems introduces many new challenges. The connectivity inherent in the
Cloud raises a range of privacy and security concerns [133],
[139]. These concerns include data generated by Cloud-connected robots and sensors, especially as they may include images or video or data from private homes or corporate trade
secrets [130], [161]. Cloud Robotics and Automation also introduces the potential of robots and systems to be attacked remotely: a hacker could take over a robot and use it to disrupt
functionality or cause damage. For instance, researchers at University of Texas at Austin demonstrated that it is possible to
hack into and remotely control UAV drones via inexpensive

404

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 2, APRIL 2015

Fig. 10. (Left) Schematic architecture of the Lightning path planning framework. Berenson et al. show a system that is able to learn from experience from precomputed motion plans, which could be stored in the Cloud. The planner attempts to nd a brand-new plan as well as nd an existing plan for a problem similar to
the current one. Whichever nishes rst is chosen [35]. Lightning uses the Big Data, Cloud Computing, and Collective Robot Learning aspects of Cloud Robotics
and Automation. (Image reproduced with permission from authors.)

Fig. 11. Tiered human assistance using Cloud-based resources for teleoperation. Leeper et al. developed an interface for operators to control grasp execution using
a set of different strategies. The results indicate humans are able to select better and more robust grasp strategies [103], [164]. (Image reproduced with permission.)

Fig. 12. Crowdsourcing object identication to facilitate robot grasping. Sorokin et al. developed a Cloud robot system that incorporates Amazon's Mechanical
Turk to obtain semantic information about the world and subjective judgments [144]. This work uses the Human Computation aspect of Cloud Robotics and
Automation. (Image reproduced with permission from authors.)

GPS spoong systems in an evaluation study for the Department of Homeland Security (DHS) and the Federal Aviation
Administration (FAA) [76]. These concerns raise new regulatory, accountability and legal issues related to safety, control,
and transparency [107], [130]. The We Robot conference is
an annual forum for ethics and policy research [160].
On the technical front, new algorithms and methods are
needed to cope with time-varying network latency and
Quality-of-Service. Faster data connections, both wired Internet connections and wireless standards such as LTE [31], are
reducing latency, but algorithms must be designed to degrade
gracefully when the Cloud resources are very slow, noisy,
or unavailable [34]. For example, anytime load balancing
algorithms for speech recognition on smart phones send the
speech signal to the Cloud for analysis and simultaneously

process it internally and then use the best results available

after a reasonable delay. Similar algorithms will be needed for
robotics and automation systems [35].
New algorithms are also needed that scale to the size of
Big Data, which often contain dirty data that requires new
approaches to clean or sample effectively [6], [158]. When the
Cloud is used for parallel-processing, it is vital that algorithms
oversample to take into account that some remote processors
may fail or experience long delays in returning results. When
human computation is used, algorithms are needed to lter
unreliable input and balance the costs of human intervention
with the cost of robot failure.
Moving robotics and automation algorithms into the Cloud
requires frameworks that facilitate this transition. The Cloud
provides three possible levels at which a framework could be

KEHOE et al.: A SURVEY OF RESEARCH ON CLOUD ROBOTICS AND AUTOMATION

Fig. 13. The DARPA Robotics Challenge (DRC) used CloudSim, an opensource Cloud-based simulation platform for testing the performance of the Atlas
humanoid robot (shown) on a variety of disaster response tasks [5], [7]. The
Cloud permits running interactive, real-time simulation tasks in parallel for purposes such as predicting and evaluating performance, validating design decisions, optimizing designs, and training users. This competition also resulted in
enabling sharing of robotics research efforts. (Image reproduced with permission.)

Fig. 14. Lollibot, designed by Tom Tilley of Thailand, won the Grand Prize in
the $10 Educational Robot Design Challenge organized by the African Robotics
Network. This design can be built from surplus parts for US $8.96. [150]. (Image
reproduced with permission.)

implemented [115]. The lowest level is Infrastructure as a Service (IaaS), where bare operating systems are provided on (possibly virtualized) machines in the Cloud. The second level, Platform as a Service (PaaS), provides more structure, including
application frameworks and database access, while restricting
the choice of programming languages, system architectures, and
database models that can be used. Software as a Service (SaaS),
the highest level of structure, is exemplied by the difference
between Google Docs, a Cloud-based word processor, and Microsoft Word, which must be downloaded and installed locally.
The RoboEarth project includes a Cloud Computing platform
called Rapyuta [118], which is a Platform as a Service (PaaS)
framework for moving computation off of robots and into the
Cloud. It also connects to the RoboEarth knowledge repository,
integrating the Big Data aspect. We believe that this PaaS approach can be extended to use the Software as a Service (SaaS)
paradigm, which offers many potential advantages for robots

405

and automation systems [85]. With SaaS, an interface allows

data to be sent to a server that processes it and returns outputs,
which relieves users of the burden of maintaining data and software and hardware and allows companies to control proprietary
software.
We call this approach Robotics and Automation as a Service
(RAaaS). To illustrate the concept, consider two scenarios for a
graduate student setting up a robot workcell. The workcell contains a 7-DoF Fanuc industrial arm with parallel-jaw gripper and
a Microsoft Kinect RGBD sensor. The purpose of the workcell
is to pick up and inspect parts as they come down an assembly
line, requiring object recognition and localization, grasp planning, and motion planning.
In Scenario 1 (today with ROS), the software runs locally.
ROS (Robot Operating System), the well-known open-source
library of robotics software [131], provides access to over 2000
open-source ROS packages. Currently however, ROS is only
supported on the Ubuntu Linux operating system. While Ubuntu
is popular, the computers available to the graduate student run
OSX. Many stable ROS packages are provided as packages,
which simplies installation, but some software is only available as a source distribution, which requires the download and
installation of dependencies. The graduate student must set up a
new machine with Ubuntu and resolve all library dependencies,
including those that conict with other packages.
In contrast, Scenario 2 (in the future with RAaaS), the
analysis and planning software runs in the Cloud. The graduate
student visits a website to input the robot, sensor, and gripper
models. She then selects her desired object recognition and
localization, motion planning, and grasping algorithms, and
uses a graphical interface to connect these algorithms into a
pipeline. Her robot begins sending up data in the form of point
clouds from the Kinect. The robot receives and executes motion
plans and grasps, reporting back outcomes to the Cloud-based
pipeline, which are combined with feedback from other robots
to improve the Cloud-based software parameters over time. We
are excited about the potential of such a system and actively
working with others on developing its components.
This survey is based on research available in August 2014.
A repository for new developments and updates is available at:
http://goldberg.berkeley.edu/cloud-robotics/
ACKNOWLEDGMENT
The authors thank C. Agero, A. Aydemir, A. Bayen,
George Bekey, K. Bekris, D. Berenson, G. Bradski, M.
Ciocarlie, J. Civera, S. Cousins, D. Culler, R. D'Andrea,
S. Gentner, B. Gerkey, E. Guizzo, T. Hunter, M. Ani Hsieh,
V. Isler, R. Katz, J. Kuffner, V. Kumar, J. Malec, M. Mason,
D. Portugal, B. Recht, L. Riazuelo, J. Salmern-Garca, S. Sarma,
A. Saxena, D. Song, C. Sutter, R. Voyles, A. Waibel, L. Wang,
and J. Wiegley for ongoing insights and advice on this topic.
REFERENCES
[1] Amazon Elastic Cloud (EC2). [Online]. Available: http://aws.
amazon.com/ec2/
[2] Amazon Web Services. [Online]. Available: http://aws.amazon.com
[3] Arduino. [Online]. Available: http://www.arduino.cc
[4] Bullet Physics Library. [Online]. Available: http://bulletphysics.org

406

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 2, APRIL 2015

[5] CloudSim.
[Online].
Available:
http://gazebosim.org/wiki/
CloudSim/
[6] For Big-Data Scientists, Janitor Work Is Key Hurdle to Insights.
[Online]. Available: http://www.nytimes.com/2014/08/18/technology/
for-big-data-scientists-hurdle-to-insights-is-janitor-work.html
[7] Gazebo. [Online]. Available: http://gazebosim.org/
[8] Google Compute Engine. [Online]. Available: https://cloud.google.
com/products/compute-engine
[9] Google Goggles. [Online]. Available: http://www.google.com/ mobile/goggles/
[10] IEEE Networked Robots Technical Committee. [Online]. Available:
http://www-users.cs.umn.edu/_isler/tc/
[11] Industry 4.0. [Online]. Available: http://www.bmbf.de/en/19955.php
[12] Microsoft
Azure.
[Online].
Available:
http://www.windowsazure.com
[13] MyRobots.com. [Online]. Available: http://myrobots.com
[14] Open Source Robotics Foundation. [Online]. Available: http: //www.
osrfoundation.org
[15] OpenRAVE. [Online]. Available: http://openrave.org/
[16] RoboBrain. [Online]. Available: http://robobrain.me
[17] ROS-Industrial. [Online]. Available: http://rosindustrial.org
[18] ROS (Robot Operating System). [Online]. Available: http://ros.org
[19] Rosjava, an Implementation of ROS in Pure Java With Android Support. [Online]. Available: http://cloudrobotics.com
[20] TOP500. [Online]. Available: http://www.top500.org/list/2012/06/
100
[21] What is MyRobots? [Online]. Available: http://myrobots.com/wiki/
About
[22] What is RoboEarth? [Online]. Available: http://www.roboearth.org/
what-is-roboearth
[23] An Open-Source Robo-Surgeon, The Economist, 2012. [Online].
Available: http://www.economist.com/node/21548489
[24] DARPA Selects Southwest Research Institute to Support DARPA
Robotics Challenge, Market Watch, 2013.
[25] Accenture Inc., A New Era for Energy Companies: Cloud Computing
Changes the Game. 2012. [Online]. Available: http://www.accenture.com/SiteCollectionDocuments/PDF/Accenture-New-Era-Energy-Companies-Cloud-Computing-changes-Game. pdf
[26] B. Addad, S. Amari, and J.-J. Lesage, Analytic calculus of response
time in networked automation systems, IEEE Trans. Autom. Sci. Eng.
(T-ASE), vol. 7, no. 4, pp. 858869, Oct. 2010.
[27] B. Addad, S. Amari, and J.-J. Lesage, Client-server networked automation systems reactivity: Deterministic and probabilistic analysis,
IEEE Trans. Autom. Sci. Eng. (T-ASE), vol. 8, no. 3, pp. 540548, Jul.
2011.
[28] C. Aguero, N. Koenig, I. Chen, H. Boyer, S. Peters, J. Hsu, B. Gerkey,
S. Paepcke, J. Rivero, J. Manzo, E. Krotkov, and G. Pratt, Inside
the virtual robotics challenge: Simulating real-time robotic disaster response, IEEE Trans. Autom. Sci. Eng. (T-ASE): Special Issue Cloud
Robot. Autom., vol. 12, no. 2, pp. 494506, Apr. 2015.
[29] M. Armbrust, I. Stoica, M. Zaharia, A. Fox, R. Grifth, A. D. Joseph,
R. Katz, A. Konwinski, G. Lee, D. Patterson, and A. Rabkin, A view
of cloud computing, Commun. ACM, vol. 53, no. 4, p. 50, Apr. 2010.
[30] R. Arumugam, V. Enti, L. Bingbing, W. Xiaojun, K. Baskaran, F.
Kong, A. Kumar, K. Meng, and G. Kit, DAvinCi: A cloud computing
framework for service robots, in Proc. Int. Conf. Robot. Autom.
(ICRA), 2010, pp. 30843089.
[31] D. Astly, E. Dahlman, A. Furuskr, Y. Jading, M. Lindstrm, and S.
Parkvall, LTE: The evolution of mobile broadband, Commun. Mag.,
vol. 47, no. 4, pp. 4451, 2009.
[32] L. Atzori, A. Iera, and G. Morabito, The Internet of things: A survey,
Comput. Netw., vol. 54, no. 15, pp. 27872805, Oct. 2010.
[33] K. Bekris, R. Shome, A. Krontiris, and A. Dobson, Cloud automation: Precomputing roadmaps for exible manipulation, IEEE Robot.
Autom. Mag.: Special Issue on Emerging Advances and Applications
in Automation, 2014, under review.
[34] W. Beksi and N. Papanikolopoulos, Point cloud culling for robot vision tasks under communication constraints, in Proc. Int. Conf. Intell.
Robot. Syst. (IROS), Chicago, IL, USA, 2014.
[35] D. Berenson, P. Abbeel, and K. Goldberg, A robot path planning
framework that learns from experience, in Proc. Int. Conf. Robot.
Autom. (ICRA), May 2012, pp. 36713678.
[36] B. Bhargava, P. Angin, and L. Duan, A mobile-cloud pedestrian
crossing guide for the blind, in Proc. Int. Conf. Adv. Comput.
Commun., 2011.
[37] J. Bohg, A. Morales, T. Asfour, and D. Kragic, Data-driven grasp
synthesisA survey, IEEE Trans. Robotics (T-RO), no. 99, pp. 121,
2013.

[38] V. Chandrasekaran and M. I. Jordan, Computational and statistical

tradeoffs via convex relaxation, Proc. Nat. Acad. Sci. USA, vol. 110,
no. 13, pp. E1181E1190, 2013.
[39] M. Ciocarlie, K. Hsiao, E. G. Jones, S. Chitta, R. Rusu, and I. Sucan,
Towards reliable grasping and manipulation in household environments, in Proc. Exp. Robot. Int. Symp. (ISER), Delhi, India, 2010.
[40] M. Ciocarlie, C. Pantofaru, K. Hsiao, G. Bradski, P. Brook, and E.
Dreyfuss, A side of data with my robot, IEEE Robot. Autom. Mag.,
vol. 18, no. 2, pp. 4457, Jun. 2011.
[41] Committee on the Analysis of Massive Data; Committee on Applied
and Theoretical Statistics; Board on Mathematical Sciences and Their
Applications; Division on Engineering and Physical Sciences; National
Research Council, Frontiers in massive data analysis, The National
Academies Press. Washington, DC, USA, 2013.
[42] L. Dabbish, C. Stuart, J. Tsay, and J. Herbsleb, Social coding in
github: Transparency and collaboration in an open software repository, in Proc. ACM Conf. Comput. Supported Cooperative Work,
2012, p. 1277.
[43] S. Dalibard, A. Nakhaei, F. Lamiraux, and J.-P. Laumond, Manipulation of documented objects by a walking humanoid robot, in Proc.
IEEE-RAS Int. Conf. Humanoid Robot., Dec. 2010, pp. 518523.
[44] R. D'Andrea, Guest editorial: A revolution in the warehouse: A Rretrospective on KIVA systems and the grand challenges ahead, IEEE
Trans. Autom. Sci. Eng. (T-ASE), vol. 9, no. 4, pp. 638639, Oct. 2012.
[45] H. Dang and P. K. Allen, Learning grasp stability, in Proc. Int. Conf.
Robot. Autom. (ICRA), May 2012, pp. 23922397.
[46] H. Dang, J. Weisz, and P. K. Allen, Blind grasping: Stable robotic
grasping using tactile feedback and hand kinematics, in Proc. Int.
Conf. Robot. Autom. (ICRA), May 2011, pp. 59175922.
[47] S. Davidson, Open-source Hardware, IEEE Design Test Comput.,
vol. 21, no. 5, pp. 456456, Sep. 2004.
[48] G. de Croon, P. Gerke, and I. Sprinkhuizen-Kuyper, Crowdsourcing
as a methodology to obtain large and varied robotic data sets,
in Proc. Int. Conf. Intell. Robot. Syst. (IROS), Chicago, IL,
USA, 2014.
[49] J. Dean, G. Corrado, R. Monga, K. Chen, M. Devin, M. Mao, M. Ranzato, A. Senior, P. Tucker, K. Yang, Q. V. Le, and A. Y. Ng, Large
scale distributed deep networks, in Advances in Neural Information
Processing Systems 25, F. Pereira, C. Burges, L. Bottou, and K. Weinberger, Eds., 2012, pp. 12231231.
[50] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, Imagenet: A large-scale hierarchical image database, in Proc. IEEE Conf.
Comput. Vision Pattern Recognit., 2009, pp. 248255.
[51] A. Dobson, A. Krontiris, and K. E. Bekris, Sparse roadmap spanners,
Algorithmic Foundations of Robotics X, pp. 279296, 2013.
[52] Droplet, Cloud Powered Water Sprinkler System, 2014. [Online].
Available: http://smartdroplet.com/
[53] Z. Du, W. Yang, Y. Chen, X. Sun, X. Wang, and C. Xu, Design of
a robot cloud center, in Proc. Int. Symp. Auton. Decentralized Syst.,
Mar. 2011, pp. 269275.
[54] E. Dumbill, Dening Big Data, [Online]. Available: Forbes Data
Driven Blog, 2014
[55] P. C. Evans and M. Annunziata, Industrial Internet: Pushing the
boundaries of minds and machines, General Electric, Tech. Rep.,
2012.
[56] M. Everingham, L. Van Gool, C. K. Williams, J. Winn, and A. Zisserman, The PASCAL visual object classes (VOC) challenge, Int. J.
Comput. Vision, vol. 88, no. 2, pp. 303338, 2010.
[57] S. Gammeter, A. Gassmann, L. Bossard, T. Quack, and L. Van Gool,
Server-side object recognition and client-side object tracking for mobile augmented reality, in Proc. IEEE Conf. Comput. Vision Pattern
Recognit., Jun. 2010, pp. 18.
[58] J. Garca, Using cloud computing as a HPC platform for embedded
systems, 2011, Unpublished work. [Online]. Available: http://www.
atc.us.es/descargas/tfmHPCCloud.pdf
[59] General Electric, The case for an industrial big data platform: Laying
the groundwork for the new industrial age, 2014. [Online]. Available:
https://gesoftware.com/Industrial Big Data Platform.pdf
[60] N. Gershenfeld, Fab: The Coming Revolution on Your Desktop-from
Personal Computers to Personal Fabrication. New York, NY, USA:
Basic Books, 2007.
[61] Y. Gingold, A. Shamir, and D. Cohen-Or, Micro perceptual human
computation for visual tasks, ACM Trans. Graphics, vol. 31, no. 5,
pp. 112, Aug. 2012.
[62] J. Glover, D. Rus, and N. Roy, Probabilistic models of object geometry for grasp planning, Robotics: Science and Systems (RSS), vol. IV,
pp. 278285, 2008.
[63] K. Goldberg, Beyond the Web: Excavating the real world via mosaic,
in Proc. 2nd Int. World Wide Web Conf., 1994, pp. 112.

KEHOE et al.: A SURVEY OF RESEARCH ON CLOUD ROBOTICS AND AUTOMATION

[64] K. Goldberg and B. Chen, Collaborative control of robot motion: Robustness to error, in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst.,
2001, vol. 2, pp. 655660.
[65] K. Goldberg, M. Mascha, S. Gentner, N. Rothenberg, C. Sutter, and J.
Wiegley, Desktop teleoperation via the world wide web, in Proc. Int.
Conf. Robot. Autom. (ICRA), Nagoya, Japan, 1995, pp. 654659.
[66] Beyond Webcams: An Introduction to Online Robots, K. Goldberg and
R. Siegwart, Eds. Cambridge, MA, USA: MIT Press, 2002.
[67] C. Goldfeder, M. Ciocarlie, and P. Allen, The Columbia grasp
database, in Proc. Int. Conf. Robot. Autom. (ICRA), May 2009, pp.
17101716.
[68] C. Goldfeder and P. K. Allen, Data-driven grasping, Auton. Robot.,
vol. 31, no. 1, pp. 120, Apr. 2011.
[69] B. Gouveia, D. Portugal, D. Silva, and L. Marques, Computation
sharing in distributed robotic systems: A case study on SLAM, IEEE
Trans. Autom. Sci. Eng. (T-ASE): Special Issue Cloud Robot. Autom.,
vol. 12, no. 2, pp. 410422, Apr. 2015.
[70] E. Guizzo, Cloud robotics: Connected to the cloud, robots get
smarter, IEEE Spectrum Automaton Blog, 2011.
[71] B. Hannaford, J. Rosen, D. Friedman, H. King, P. Roan, L. Cheng, D.
Glozman, J. Ma, S. Kosari, and L. White, Raven II: Open platform for
surgical robotics research, IEEE Trans. Biomed. Eng., pp. 954959,
Apr. 2013.
[72] E. Hidago-Pena, L. F. Marin-Urias, F. Montes-Gonzalez, A.
Marin-Hernandez, and H. V. Rios-Figueroa, Learning from the
Web: Recognition method based on object appearance from Internet
images, in Proc. ACM/IEEE Int. Conf. Human-Robot Interaction,
2013, pp. 139140.
[73] J. Higuera, A. Xu, F. Shkurti, and G. Dudek, Socially-driven collective path planning for robot missions, in Proc. Conf. Comput. Robot
Vision, May 2012, pp. 417424.
[74] G. Hu, W. Tay, and Y. Wen, Cloud robotics: Architecture, challenges
and applications, IEEE Network, vol. 26, no. 3, pp. 2128, 2012.
[75] K. Huebner, K. Welke, M. Przybylski, N. Vahrenkamp, T. Asfour, and
D. Kragic, Grasping known objects with humanoid robots: A boxbased approach, in Proc. Int. Conf. Adv. Robot., Munich, Germany,
2009.
[76] Humphreys, Todd, Cockrell School researchers demonstrate rst successful spoong of UAVs, 2012. [Online]. Available: http://www.
engr.utexas.edu/features/humphreysspoong
[77] T. Hunter, T. Moldovan, M. Zaharia, S. Merzgui, J. Ma, M. J. Franklin,
P. Abbeel, and A. M. Bayen, Scaling the mobile millennium system in
the cloud, in Proc. 2nd ACM Symp. Cloud Compu. (SOCC), Cascais,
Portugal, 2011, pp. 28:128:8.
[78] D. Hunziker, M. Gajamohan, M. Waibel, and R. D'Andrea, Rapyuta:
The roboearth cloud engine, in Proc. IEEE Int. Conf. Robot. Autom.
(ICRA), Karlsruhe, Germany, 2013, pp. 438444.
[79] M. Inaba, Remote-brained robots, in Proc. Int. Joint Conf. Artif. Intell., 1997, pp. 15931606.
[80] J. D. Irwin, The Industrial Electronics Handbook. Boca Raton, FL,
USA: CRC Press, 1997.
[81] P. F. Jacobs and D. T. Reid, Rapid Prototyping & Manufacturing: Fundamentals of Stereolithography. Dearborn, MI, USA: Soc. Manuf.
Eng., 1992.
[82] A. Jain, D. Das, and A. Saxena, PlanIt: A crowdsourcing approach
for learning to plan paths from large scale preference feedback, Tech.
Rep., 2014.
[83] N. K. Jangid, Real time cloud computing, in Data Manage. Security,
2011.
[84] M. Johnson-Roberson, J. Bohg, G. Skantze, J. Gustafson, R. Carlson,
B. Rasolzadeh, and D. Kragic, Enhanced visual scene understanding
through human-robot dialog, in Proc. Int. Conf. Intell. Robot. Syst.
(IROS), Sep. 2011, pp. 33423348.
[85] S. Jordan, T. Haidegger, L. Kovacs, I. Felde, and I. Rudas, The rising
prospects of cloud robotic applications, in Proc. Int. Conf. Comput.
Cybern. (ICCC), Jul. 2013, pp. 327332.
[86] G. Juve, E. Deelman, G. B. Berriman, B. P. Berman, and P. Maechling,
An evaluation of the cost and performance of scientic workows on
amazon EC2, J. Grid Comput., vol. 10, no. 1, pp. 521, Mar. 2012.
[87] K. Kamei, S. Nishio, N. Hagita, and M. Sato, Cloud networked
robotics, IEEE Network, vol. 26, no. 3, pp. 2834, 2012.
[88] A. Kasper, Z. Xue, and R. Dillmann, The KIT object models database:
An object model database for object recognition, localization and manipulation in service robotics, Int. J. Robot. Res. (IJRR), vol. 31, no.
8, pp. 927934, May 2012.
[89] B. Kehoe, A. Matsukawa, S. Candido, J. Kuffner, and K. Goldberg,
Cloud-based robot grasping with the google object recognition engine, in Proc. Int. Conf. Robot. Autom. (ICRA), Karlsruhe, Germany,
2013, pp. 42634270.

407

[90] B. Kehoe, D. Berenson, and K. Goldberg, Estimating part tolerance

bounds based on adaptive cloud-based grasp planning with slip, in
Proc. IEEE Int. Conf. Autom. Sci. Eng. (CASE), Seoul, Korea, 2012,
pp. 11061113.
[91] B. Kehoe, D. Berenson, and K. Goldberg, Toward cloud-based
grasping with uncertainty in shape: Estimating lower bounds on
achieving force closure with zero-slip push grasps, in Proc. Int. Conf.
Robot. Autom. (ICRA), St. Paul, MN, USA, May 2012, pp. 576583.
[92] B. Kehoe, D. Warrier, S. Patil, and K. Goldberg, Cloud-based grasp
planning for toleranced parts using parallelized Monte Carlo sampling, IEEE Trans. Autom. Sci. Eng. (TASE): Special Issue Cloud
Robot. Autom., vol. 12, no. 2, pp. 455470, Apr. 2015.
[93] J. Kelly, The industrial internet and big data analytics: Opportunities and challenges, 2013. [Online]. Available: http://wikibon.org/wiki/v/The Industrial Internet and Big Data Analytics%3A
Opportunities and Challenges
[94] D. Kent and S. Chernova, Construction of an object manipulation
database from grasp demonstrations, in Proc. Int. Conf. Intell. Robot.
Syst. (IROS), Chicago, IL, USA, 2014.
[95] W.-J. Kim, K. Ji, and A. Ambike, Real-time operating environment
for networked control systems, IEEE Trans. Autom. Sci. Eng. (T-ASE),
vol. 3, no. 3, pp. 287296, 2006.
[96] KIVA, KIVA Systems, 2014. [Online]. Available: www.kivasystems.com
[97] A. Krizhevsky, I. Sutskever, and G. E. Hinton, ImageNet classication with deep convolutional neural networks, in Advances in Neural
Information Processing Systems 25, F. Pereira, C. Burges, L. Bottou,
and K. Weinberger, Eds., 2012, pp. 10971105.
[98] J. Kuffner, Cloud-enabled robots, in Proc. IEEE-RAS Int. Conf. Humanoid Robot., Nashville, TN, USA, 2010.
[99] V. Kumar, R. Daniela, and G. S. Sukhatme, Networked robotics, in
Springer Handbook of Robotics, B. Siciliano and O. Khatib, Eds.
Berlin, Germany: Springer-Verlag, 2004.
[100] V. Kumar and N. Michael, Opportunities and challenges with autonomous micro aerial vehicles, Int. J. Robot. Res. (IJRR), vol. 31,
no. 11, pp. 12791291, 2012.
[101] K. Lai and D. Fox, Object recognition in 3D point clouds using web
data and domain adaptation, Int. J. Robot. Res. (IJRR), vol. 29, no. 8,
pp. 10191037, May 2010.
[102] Q. Le, M. Ranzato, R. Monga, M. Devin, K. Chen, G. Corrado, J. Dean,
and A. Ng, Building high-level features using large scale unsupervised
learning, in Proc. Int. Conf. Mach. Learn., Vancouver, BC, Canada,
2012, pp. 85958598.
[103] A. E. Leeper, K. Hsiao, M. Ciocarlie, L. Takayama, and D. Gossow,
Strategies for human-in-the-loop robotic grasping, in Proc. ACM/
IEEE Int. Conf. Human-Robot Interaction, 2012, pp. 18.
[104] I. Lenz, H. Lee, and A. Saxena, Deep learning for detecting robotic
grasps, Int. J. Robot. Res. (IJRR), to be published.
[105] Z. Li, L. O'Brien, H. Zhang, and R. Cai, On the conceptualization
of performance evaluation of IaaS services, IEEE Trans. Services
Comput., to be published.
[106] Z. Li, H. Zhang, L. O'Brien, R. Cai, and S. Flint, On evaluating commercial Cloud services: A systematic review, J. Syst. Softw., vol. 86,
no. 9, pp. 23712393, 2013.
[107] P. Lin, K. Abney, and G. A. Bekey, Robot Ethics: The Ethical and
Social Implications of Robotics. Cambridge, MA, USA: MIT Press,
2011.
[108] H. Lipson and M. Kurman, Fabricated: The New World of 3D
Printing. New York, NY, USA: Wiley, 2013.
[109] B. Liu, Y. Chen, E. Blasch, and K. Pham, A holistic cloud-enabled
robotics system for real-time video tracking application, in Future Inform. Technol., 2014, pp. 455468.
[110] C.-H. Lu and L.-C. Fu, Robust location-aware activity recognition
using wireless sensor network in an attentive home, IEEE Trans.
Autom. Sci. Eng. (T-ASE), vol. 6, no. 4, pp. 598609, 2009.
[111] J. Mahler, S. Krishnan, M. Laskey, S. Sen, A. Murali, B. Kehoe,
S. Patil, J. Wang, M. Franklin, P. Abbeel, and K. Goldberg,
Learning accurate kinematic control of cable-driven surgical
robots using data cleaning and Gaussian process regression,
in Proc. IEEE Int. Conf. Autom. Sci. Eng. (CASE), Taipei,
Taiwan, 2014.
[112] G. McKee, What is networked robotics?, Informatics in Control Automation and Robotics, vol. 15, pp. 3545, 2008.
[113] McKinsey Inc., Big data and the opportunities it creates for
semiconductor players, 2012. [Online]. Available: http://www.mckinsey.com/~/media/McKinsey/dotcom/client_service/Semiconductors/Issue\Autumnn\2012/PDFs/Big_data_and_the_opportunities_it_creates_for_semiconductor_players.ashx

408

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 2, APRIL 2015

[114] P. Mehrotra, J. Djomehri, S. Heistand, R. Hood, H. Jin, A. Lazanoff,

S. Saini, and R. Biswas, Performance evaluation of Amazon EC2 for
NASA HPC applications, in Proc. 3rd Workshop on Scientific Cloud
Computing DateScienceCloud12, 2012, pp. 4150.
[115] P. Mell and T. Grance, The NIST denition of cloud computing,
National Institute of Standards and Technology, 2009.
[116] N. Michael, D. Mellinger, Q. Lindsey, and V. Kumar, The GRASP
multiple micro UAV testbed, IEEE Robot. Autom. Mag., vol. 17, no.
3, pp. 5665, 2010.
[117] A. Miller and P. Allen, GraspIt! A versatile simulator for robotic
grasping, IEEE Robot. Autom. Mag., vol. 11, no. 4, pp. 110122, Dec.
2004.
[118] G. Mohanarajah, D. Hunziker, M. Waibel, and R. D'Andrea, Rapyuta:
A cloud robotics platform, IEEE Trans. Autom. Sci. Eng. (T-ASE), vol.
12, no. 2, pp. 481493, Apr. 2015.
[119] G. Mohanarajah, V. Usenko, M. Singh, M. Waibel, and R. D'Andrea,
Cloud-based collaborative 3D mapping in real-time with low-cost
robots, IEEE Trans. Autom. Sci. Eng. (T-ASE): Special Issue Cloud
Robot. Autom., vol. 12, no. 2, pp. 423431, Apr. 2015.
[120] M. Moussa and M. Kamel, An experimental approach to robotic
grasping using a connectionist architecture and generic grasping
functions, IEEE Trans. Syst., Man, Cybern., Part C, vol. 28, no. 2,
pp. 239253, May 1998.
[121] M. Narita, S. Okabe, Y. Kato, Y. Murakwa, K. Okabayashi, and S.
Kanda, Reliable cloud-based robot services, in Proc. Conf. IEEE Ind.
Electron. Soc., 2013, pp. 83178322.
[122] National Science Foundation, Enabling a new future for cloud computing, 2014. [Online]. Available: http://nsf.gov/news/news summ.
jsp? cntn id=132377
[123] D. Nister and H. Stewenius, Scalable recognition with a vocabulary
tree, in Proc. IEEE Conf. Comput. Vision Pattern Recognit. (CVPR),
2006, vol. 2, pp. 21612168.
[124] D. Nurmi, R. Wolski, C. Grzegorczyk, G. Obertelli, S. Soman, L.
Youseff, and D. Zagorodnov, The eucalyptus open-source cloud-computing system, in Proc. IEEE/ACM Int. Symp. Cluster Comput. Grid,
2009, pp. 124131.
[125] G. Oliveira and V. Isler, View planning for cloud-based active object
recognition, Dept. Comput. Sci., Univ. Minnesota, Minneapolis, MN,
USA, Tech. Rep. 13-025, 2013.
[126] E. Plaku, K. E. Bekris, B. Y. Chen, A. M. Ladd, and L. E. Kavraki,
Sampling-based roadmap of trees for parallel motion planning, IEEE
Trans. Robotics (T-RO), vol. 21, no. 4, pp. 597608, 2005.
[127] E. Plaku, K. E. Bekris, and L. E. Kavraki, OOPS for motion planning:
An online, open-source, programming system, in Proc. Int. Conf.
Robot. Autom. (ICRA), Apr. 2007, pp. 37113716.
[128] M. Popovic, G. Kootstra, J. A. Jorgensen, D. Kragic, and N. Kruger,
Grasping unknown objects using an early cognitive vision system for
general scene understanding, in Proc. Int. Conf. Intell. Robot. Syst.
(IROS), Sep. 2011, pp. 987994.
[129] E. Prestes, J. L. Carbonera, S. Rama Fiorini, V. A. M. Jorge, M. Abel,
R. Madhavan, A. Locoro, P. Goncalves, M. E. Barreto, M. Habib, A.
Chibani, S. Grard, Y. Amirat, and C. Schlenoff, Towards a core ontology for robotics and automation, Robot. Auton. Syst., vol. 61, no.
11, pp. 11931204, 2013.
[130] A. A. Proia, D. Simshaw, and K. Hauser, Consumer cloud robotics and
the fair information practice principles: Recognizing the challenges and
opportunities ahead, Minnesota J. Law, Sci. Technol., to be published.
[131] M. Quigley, B. Gerkey, K. Conley, J. Faust, T. Foote, J. Leibs, E.
Berger, R. Wheeler, and A. Ng, ROS: An open-source robot operating
system, in Proc. ICRA Workshop on Open Source Software, Kobe,
Japan, 2009.
[132] A. J. Quinn and B. B. Bederson, Human computation: A survey and
taxonomy of a growing eld, in Proc. SIGCHI Conf. Human Factors
Comput. Syst., 2011, ser. CHI '11, pp. 14031412.
[133] K. Ren, C. Wang, and Q. Wang et al., Security challenges for the
public cloud, IEEE Internet Comput., vol. 16, no. 1, pp. 6973, 2012.
[134] L. Riazuelo, J. Civera, and J. Montiel, C2TAM: A cloud framework
for cooperative tracking and mapping, Robot. Auton. Syst., vol. 62,
no. 4, pp. 401413, 2013.
[135] L. Riazuelo, M. Tenorth, D. D. Marco, M. Salas, L. Mosenlechner, L.
Kunze, M. Beetz, J. D. Tardos, L. Montano, and J. Montiel, RoboEarth
web-enabled and knowledge-based active perception, in Proc. IROS
Workshop on AI-Based Robotics, Tokyo, Japan, 2013.
[136] L. Riazuelo, M. Tenorth, D. Marco, M. Salas, D. Galvez-Lopez, L.
Mosenlechner, L. Kunze, M. Beetz, J. Tardos, L. Montano, and J. Montiel, RoboEarth semnatic mapping: A cloud enabled knowledge-based
approach, IEEE Trans. Autom. Sci. Eng. (T-ASE): Special Issue Cloud
Robot. Autom., vol. 12, no. 2, pp. 432443, Apr. 2015.

[137] J. Salmeron-Garcia, F. Diaz-del Rio, P. Inigo-Blasco, and D.

Cagigas, A tradeoff analysis of a cloud-based robot navigation
assistant using stereo image processing, IEEE Trans. Autom.
Sci. Eng. (T-ASE): Special Issue Cloud Robot. Autom., vol. 12,
no. 2, pp. 444454, Apr. 2015.
[138] M. Samadi, T. Kollar, and M. Veloso, Using the web to interactively
learn to nd objects, in Proc. AAAI Conf. Artif. Intell., 2012, pp.
20742080.
[139] C. Schmitt, Security and privacy in the era of big data, 2014. [Online]. Available: http://www.renci.org/wp-content/uploads/2014/02/
0313WhitePaper-iRODS.pdf
[140] P. Sermanet, D. Eigen, X. Zhang, M. Mathieu, R. Fergus, and Y.
LeCun, OverFeat: Integrated recognition, localization and detection
using convolutional networks, CoRR, 2013.
[141] M. Sevior, T. Field, and N. Katayama, Belle Monte-Carlo production
on the amazon EC2 cloud, J. Phys.: Conf. Series, vol. 219, no. 1, Apr.
2010.
[142] N. Snavely, S. M. Seitz, and R. Szeliski, Photo tourism: Exploring
photo collections in 3D, ACM Trans. Graphics (TOG), vol. 25, no. 3,
pp. 835846, 2006.
[143] D. Song, K. Goldberg, and N. Y. Chong, Networked telerobots, in
Springer Handbook of Robotics, B. Siciliano and O. Khatib, Eds.
Berlin, Germany: Springer-Verlag, 2004.
[144] A. Sorokin, D. Berenson, S. S. Srinivasa, and M. Hebert, People
helping robots helping people: Crowdsourcing for grasping novel
objects, in Proc. Int. Conf. Intell. Robot. Syst. (IROS), Oct. 2010, pp.
21172122.
[145] A. Sorokin and D. Forsyth, Utility data annotation with amazon mechanical turk, in Proc. IEEE Conf. Comput. Vision Pattern Recognit.
(CVPR) Workshops, Jun. 2008, pp. 18.
[146] M. Stenmark, J. Malec, K. Nilsson, and A. Robertsson, On distributed
knowledge bases for small-batch assembly, IEEE Trans. Autom. Sci.
Eng. (T-ASE): Special Issue Cloud Robot. Autom., vol. 12, no. 2, pp.
519528, Apr. 2015.
[147] M. Tenorth and M. Beetz, KnowRob: A knowledge processing infrastructure for cognition-enabled robots, Int. J. Robot. Res. (IJRR), vol.
32, no. 5, pp. 566590, 2013.
[148] M. Tenorth, A. Clifford Perzylo, R. Lafrenz, and M. Beetz, The
RoboEarth language: Representing and exchanging knowledge about
actions, objects, environments, in Proc. Int. Conf. Robot. Autom.
(ICRA), May 2012, pp. 12841289.
[149] M. Tenorth, A. C. Perzylo, R. Lafrenz, and M. Beetz, Representation
and exchange of knowledge about actions, objects, environments in the
roboearth framework, IEEE Trans. Autom. Sci. Eng. (T-ASE), vol. 10,
no. 3, pp. 643651, Jul. 2013.
[150] The African Robotics Network (AFRON), Ten dollar robot design
challenge winners, 2012. [Online]. Available: http://robotics-africa.
org/design challenge.html
[151] A. Torralba, R. Fergus, and W. T. Freeman, 80 million tiny images: A
large data set for nonparametric object and scene recognition, IEEE
Trans. Pattern Anal. Mach. Intell., vol. 30, no. 11, pp. 19581970,
2008.
[152] R. Tudoran, A. Costan, G. Antoniu, and L. Boug, A performance
evaluation of azure and nimbus clouds for scientic applications, in
Proc. Int. Workshop on Cloud Computing PlatformsCloudCP'12,
2012, pp. 16.
[153] L. Turnbull and B. Samanta, Cloud robotics: Formation control of a
multi robot system utilizing cloud infrastructure, in Proc. IEEE Southeastcon, 2013, pp. 14.
[154] J. van den Berg, P. Abbeel, and K. Goldberg, LQG-MP: Optimized
path planning for robots with motion uncertainty and imperfect state
information, Int. J. Robot. Res. (IJRR), vol. 30, no. 7, pp. 895913,
Jun. 2011.
[155] L. von Ahn, Human computation, Ph.D. dissertation, Carnegie
Mellon Univ., Pittsburgh, PA, USA, 2009.
[156] M. Waibel, M. Beetz, J. Civera, R. D'Andrea, J. Elfring, D.
Glvez-Lpez, K. Hussermann, R. Janssen, J. Montiel, A. Perzylo,
B. Schiele, M. Tenorth, O. Zweigle, and R. De Molengraft,
RoboEarth, IEEE Robot. Autom. Mag., vol. 18, no. 2, pp.
6982, Jun. 2011.
[157] H. Wang, Y. Ma, G. Pratx, and L. Xing, Toward real-time
Monte Carlo simulation using a commercial cloud computing
infrastructure., Phys. Med. Bio., vol. 56, no. 17, pp. N175N181,
Sep. 2011.
[158] J. Wang, S. Krishnan, M. J. Franklin, K. Goldberg, T. Kraska, and
T. Milo, A Sample-and-clean framework for fast and accurate query
processing on dirty data, in Proc. ACM SIGMOD Int. Conf. Manage.
Data, Snowbird, UT, USA, 2014.

KEHOE et al.: A SURVEY OF RESEARCH ON CLOUD ROBOTICS AND AUTOMATION

409

[159] L. Wang, M. Liu, and M. Q.-H. Meng, Real-time multi-sensor data

retrieval for cloud robotic systems, IEEE Trans. Autom. Sci. Eng.
(T-ASE): Special Issue Cloud Robot. Autom., vol. 12, no. 2, pp.
507518, Apr. 2015.
[160] We Robot, We robot conference, 2012. [Online]. Available: http:
//robots.law.miami.edu/
[161] R. H. Weber, Internet of things-new security and privacy challenges,
Comput. Law Security Rev., vol. 26, no. 1, pp. 2330, 2010.
[162] J. Weisz and P. K. Allen, Pose error robust grasping from contact
wrench space metrics, in Proc. IEEE Int. Conf. Robot. Autom. (ICRA),
May 2012, pp. 557562.
[163] E. Wilhelm, J. Siegel, S. Mayer, J. Paefgen, M. Tiefenbeck, M. Bicker,
S. Ho, R. Dantu, and S. Sarma, CloudThink: An open standard for
projecting objects into the cloud, 2013. [Online]. Available: http://
cloud-think.com/
[164] Willow Garage, Personal service robotics with tiered human-in-theloop assistance, NSF SBIR Award 1256643 (PI:Matei Ciocarlie),
2013.
[165] Q. Zhang, L. Cheng, and R. Boutaba, Cloud computing: State-of
the-art and research challenges, J. Internet Services App., vol. 1, no.
1, pp. 718, 2010.

Pieter Abbeel received the B.S. and M.S. degree in

electrical engineering from KU Leuven, Leuven, Belgium, and the Ph.D. degree in computer science from
Stanford University, Stanford, CA, USA, in 2008.
He joined the faculty at the University of California, Berkeley, CA, USA, in the Fall of 2008,
with an appointment in the Department of Electrical
Engineering and Computer Sciences. His current
research focuses on robotics and machine learning
with a particular focus on challenges in personal
robotics, surgical robotics and connectomics. He
Dr. Abbeel has won numerous awards including Best Paper Awards at ICML
and ICRA, the Sloan Fellowship, the Air Force Ofce of Scientic Research
Young Investigator Program (AFOSR-YIP) Award, the Ofce of Naval Research Young Investigator Program (ONR-YIP) Award, the DARPA Young Faculty Award (DARPA-YFA), the Okawa Foundation Award, the TR35, the IEEE
Robotics and Automation Society (RAS) Early Career Award, and the Dick Volz
Best U.S. Ph.D. Thesis in Robotics and Automation Award.

Ben Kehoe received the B.A. degree in physics and

mathematics from Hamline University, Saint Paul,
MN, USA, in 2006. He is now working towards
the Ph.D. degree at the Department of Mechanical
Engineering, University of California, Berkeley, CA,
USA.
His research interests include cloud robotics, medical robotics, controls, and grasping.

Ken Goldberg (S84M90SM98F05) received

the Ph.D. degree in computer science from Carnegie
Mellon University, Pittsburgh, PA, USA, in 1990.
He is a Professor of Industrial Engineering and
Operations Research with the University of California, Berkeley, CA, USA, with appointments
in Electrical Engineering, Computer Science, Art
Practice, and the School of Information. He was appointed Editor-in-Chief of the IEEE TRANSACTIONS
ON AUTOMATION SCIENCE AND ENGINEERING
(T-ASE) in 2011 and served two terms (20062009)
as Vice-President of Technical Activities for the IEEE Robotics and Automation Society. He has published over 200 refereed papers and awarded eight
U.S. patents.
Prof. Goldberg is Founding Co-Chair of the IEEE Technical Committee on
Networked Robots and Founding Chair of the (T-ASE) Advisory Board. He
is the recipient of the NSF Presidential Faculty Fellowship (1995), the Joseph
Engelberger Award (2000), and the IEEE Major Educational Innovation Award
(2001).

Sachin Patil received the B.Tech. degree in computer

science and engineering from the Indian Institute of
Technology, Bombay, India, in 2006, and the Ph.D.
degree in computer science from the University of
North Carolina at Chapel Hill, NC, USA, in 2012.
He is now a Postdoctoral Researcher with the
Department Electrical Engineering and Computer
Science, University of California, Berkeley, USA.
His research interests include motion planning,
cloud robotics, and medical robotics.