IEEE Robotics & Automation Magazine - June 2011 - 79

second robot was able to generate an open-loop trajectory
and successfully open the cupboard's door. (Videos of the
demonstrator can be accessed at http://www.roboearth.org.)
Open Issues
It is not only difficult to predict the benefits and opportunities but also the risks and challenges created by sharing
data between robots. The global, open-source network outlined in this article will likely exacerbate current challenges
concerning legal [59], [60], moral [61], [62], privacy [63],
and safety [64], [65] aspects and create new and unexpected risks in the future. However, these risks are currently dwarfed by the technical challenges posed by
creating a World Wide Web for robots.
Historically, robotics has been plagued by a lack of reusable components [66], [67]. Today, the code and algorithms used by robots are typically highly hardware
dependent and difficult to reuse across platforms [68]. Not
only the much-cited lack of a widely accepted robot operating system (ROS) [69], [70] but also the lack of standard
interfaces to robot hardware components [71] continue to
be great challenges in all areas of robotics. Such compatibility between robots is a necessary condition for many
aspects of a truly World Wide Web for robots. It remains
to be seen if the potential benefits of a World Wide Webtype database for robots can provide sufficient incentive to
speed up the necessary process of interoperability and
standardization, and to what extent a worldwide storage
and an exchange of robot data can capitalize on the existing Internet standards.
Robots, unlike humans, excel at systematic and repetitive data collection. Although this provides unprecedented
opportunities for obtaining consistent and comparable
data as well as performing large-scale systematic analysis
and data mining, the sheer amount of data collected by
robotics systems far exceeds that processed by current
information systems. Despite rapid progress [72], the
technical feasibility of a system that connects robots worldwide remains unknown.
Another related issue is that of generating linked data
[42]. Humans perceive objects in relation to their environment and other objects. However, allowing robots to learn
the semantic (e.g., a bottle is a container to store liquid
food), contextual (e.g., a bottle is more likely to be found
on a cupboard than on the floor), and factual links (e.g., a
bottle is not very heavy and can be picked up quite easily)
between data remains a challenging research question.
Finally, despite a large number of potential benefits, it
is difficult to predict to what extent sharing data between
robots will be beneficial. Although data seem fundamental to robot learning and to extending robot performance
beyond preprogrammed tasks, more information does
not automatically translate into better robot performance
[73]. Future experiments will help reveal currently unexpected drawbacks and the benefits of sharing data
between robots.

(a)

(b)
Figure 9. The third demonstrator. (a) A compliant robot arm is
used to estimate an articulation model of a cupboard's door by
learning the right kinematic trajectory to open it. The robot then
shares the model using RoboEarth. (b) A second, different robot
arm, that has to interact with the same kind of cupboard
downloads this model, reuses it to generate an open-loop
trajectory, and successfully opens the cupboard. (Photo courtesy
of RoboEarth Consortium.)

Conclusions
Connecting robots worldwide is a challenging task beyond
the scope of any individual project. One main contribution
of this article is conceptual, outlining the key requirements
and potential benefits of the Internet for robots.
Another contribution of this article is a first demonstration
of the feasibility of a World Wide Web for robots. In the first
demonstrator, robots were able to successfully execute hardware-independent action recipes and could autonomously
improve their task performance throughout multiple iterations of execution and knowledge exchange using a simple
learning algorithm. This demonstrator illustrates how sharing
data between multiple robots can lead to faster learning. In
addition, it shows an example of how a robot can efficiently
execute a task, such as navigating in a new environment, without explicit planning at design time. Once a good solution to
navigate through the maze had been found, all robots could
successfully navigate it even without a preprogrammed routefinding algorithm. The final result of this demonstrator was
an improved, task-specific but platform-independent action
recipe stored in the RoboEarth database.
The second demonstrator shows how prior knowledge
can greatly increase the speed of performing complex tasks,
such as serving a drink in the semistructured environment of
JUNE 2011

IEEE ROBOTICS & AUTOMATION MAGAZINE

http://www.roboearth.org

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2011