IEEE Robotics & Automation Magazine - June 2015 - 42

Emerging Opportunities in Cloud Automation
Traditional automation involves single-task mechanisms that
are carefully separated from people, operating in highly structured environments. These systems execute repetitive tasks
with high efficiency and
accuracy without the need
Key automation operations for significant computation. Such solutions are
cost-effective only for
that can greatly benefit
large-scale industrial settings where similar prodfrom the use of cloud
ucts are assembled for
long periods of time. A
computing include motion
transformation is taking
planning and manipulation. place, however, fueled by
financial drivers as well as
technological and foundational developments. The future of automation lies in flexible
factory floors and quick burst manufacturing processes, which
can provide complex, short-life-cycle products fast. This will
allow product types and supplies to be updated frequently,
where the factory floor is reconfigured on demand through interaction with human operators. This requires a new level of
adaptability of the automation infrastructure, effectiveness in
less-structured setups, and safe coexistence with people.
Part of the solution lies in the increasing availability and
affordability of rich sensors [e.g., three dimensional (3-D)
point cloud devices] and new compliant, lightweight arms that
are easy to program (e.g., the Baxter robot, universal arms, and
so on). These solutions can be adopted by small-scale businesses, such as food processing, which have largely missed out
on the benefits of automation. However, these facilities will
also have to deal with high-level and computationally
demanding cognitive processes, such as perception, modeling,
planning, learning, and coordination. Thus, future small-scale,
flexible automation facilities will depend on effective computational solutions, both at a system and an algorithmic level.
At the same time, there is a revolution that is taking place
in computing that can influence the transformation of auto-

Big
Data

Open
Platforms

Parallel
Computing

Broad
Networks

Cloud Automation
Improved Perception

Lifelong Learning

Faster Planning

Large‐Scale Systems

Accurate Modeling

Sophisticated Robots

New Ways of Interacting with People
Figure 1. The aspects of cloud automation and the different ways
it can alter the automation landscape.

IEEE ROBOTICS & AUTOMATION MAGAZINE

June 2015

mation. Technological advances make it easier to access computing as a service over a network, where shared resources,
software, and information are available to individual
machines and devices [1]. This development can alter the
automation landscape in many different ways, giving rise to
the new field of cloud automation [10], [13], [16]. In particular, the use of cloud-based computation can provide effective
access to big data, which can include global knowledge bases
and libraries of object models, metadata, or environmental
maps. Similarly, the cloud can leverage parallel computation
on demand to quickly address the needs of computationally
demanding robot operations, such as motion planning. The
cloud can even act as the intermediary between the automation facility and human operators and can easily access a
global view of industrial operations. The cloud can also
accommodate the interaction between multiple robots, giving
them the ability to share knowledge and learn collectively
from experience over time.
Through these mechanisms, cloud computing can be
applied across a variety of industries to significantly advance
the efficiency, quality, productivity, and reliability of the automation process. Some of the benefits of cloud computing
include improved perception and world knowledge, fast resolution of planning challenges, more accurate models, opportunities for lifelong learning, highly networked large-scale systems, computing for sophisticated robots, and new ways to
interact with people (Figure 1).
Improved Perception and World Knowledge
Feeding data from rich sensors to the cloud can help continuously monitor the location of products, tools, people, and equipment on a flexible, potentially less-structured factory floor [16].
Access to large knowledge bases and databases can provide
quick identification of the type of perceived objects. Furthermore, it allows access to metadata that permits the robot to process how robotic equipment can handle and manipulate objects.
Fast Resolution of Planning Challenges
Some of the most computationally demanding processes in
robotics and automation relate to planning, manipulation,
and grasping [3], [7], [11]. They require searching highdimensional, continuous-configuration spaces and model the
different ways robotic equipment can interact with a physical
object. Massively parallel computation can significantly help
by speeding up the operation of sampling-based processes for
planning [9], [14] and by evaluating multiple hypotheses
about the state of an uncertain world [11].
More Accurate Models
Frequently, robotic algorithms make concessions by simplifying
the underlying model to achieve reasonable solution times.
Access to significant computational power allows the adaptation
of higher-fidelity models for planning purposes, including physics-based simulation [2]. This is a computationally expensive
alternative but can reveal physical interactions, which are difficult to obtain otherwise. This includes reasoning for novel

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2015