IEEE Robotics & Automation Magazine - December 2016 - 95

then transferred onto trucks for further transportation.
Once the containers reach a target warehouse facility, they
are typically unloaded manually, which is a strenuous task
because goods are often heavy and difficult to handle,
resulting in significant health risks to workers. The lack of
employees willing to engage in such exhausting and wearing
labor, combined with an aging population and increasingly
strict labor-law regulations, makes automated solutions for
container unloading desirable, if not necessary.
A robotic system designed for unloading goods from
containers has to deal with several major scientific and
technological challenges. First and foremost, such a system
needs to be able to handle a wide variety of goods of different size, shape, weight, texture, and material. Individual
goods need to be identified, grasped, transported to a
drop-off location, and finally placed securely while maintaining human-like picking success rates and unloading
times. Throughout all operations, it is also imperative that
goods are not dropped or otherwise damaged, since the
contents may be fragile. These constraints, coupled with
the possibly chaotic and cluttered arrangement of goods in
the container, result in a set of difficult interleaved hardware-design, perception, and grasp- and motion-planning
problems. Efficient solutions to these challenges will have
far-reaching applications to a number of domains involving autonomous manipulation in unstructured and uncontrolled environments.
This article delves more deeply into the underlying
challenges on the road toward fully autonomous container-unloading robots by examining two different subscenarios. The first is directly motivated by the real-world
industrial use case of unloading containers packed with
70-kg sacks of coffee. In contrast, the second subscenario
is embedded in the scientifically more challenging
domain of unloading containers packed with heterogeneous loose goods. Although the latter scenario does not
directly correspond to the current industrial practice of
packaging goods in cardboard boxes prior to loading into
a container, it serves as a scientific testbed for research on
autonomous manipulation in complex, cluttered environments, which are very common in the context of logistics.
Generalizing current approaches to such scenarios is at
the forefront of research and has direct applications to
future robot-aided logistics in, e.g., autonomous commissioning systems [i.e., the focus of the widely popular
International Conference on Robotics and Automation
2015 Amazon picking challenge (http://amazonpicking
challenge.org/)].
We propose a common automation framework for handling the unloading automation task, and we present two
different instantiations deployed on two physically different robotic platforms: the Empticon II robot [Figure 1(a)],
targeted at coffee-sack unloading; and the parcel robot
[Figure 1(b)], targeted at unloading heterogeneous loosely
packed goods, thereby investigating the more general
aspects of handling arbitrary goods in logistics applica-

(a)

(b)
Figure 1. The two autonomous container-unloading systems:
(a) the Empticon II robot, designed for unloading heavy sacks of
coffee beans, shown deployed in a storage warehouse owned
by Vollers GmbH, and (b) the parcel robot, targeted at unloading
heterogeneous loose goods. The target application scenarios are
further depicted in Figure 7.

tions. By examining two conceptually different scenarios,
we evaluate both the readiness of the proposed automation
framework to answer current industrial needs and its suitability to addressing wider and more challenging future
application domains.
Approach, Design, and Implementation
Cognitive Approach
To meet the varying requirements of the two application
domains, we opted for implementing two different robots
with significantly different kinematic structures and gripping devices (Figure 1). However, as the chief objective of
this work is to produce generally applicable solutions for the
container-unloading problem, we propose a modular and
reconfigurable framework that allows the reuse of common
components and a unified approach to the two application
domains. All components were implemented as modules in
the Robot Operating System (ROS) [1].
The framework is designed around a classical sense-
plan-act loop that is embedded in a finite-state machine.
This approach allows for seamless transitioning between
fully autonomous unloading and operator-assisted autonomy, as well as for an easy means to substitute different
module instances. At different states of the unloading process, the system uses a variety of processing paradigms,
December 2016

IEEE ROBOTICS & AUTOMATION MAGAZINE

http://amazonpicking http://www.challenge.org/

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2016