IEEE Robotics & Automation Magazine - September 2015 - 10

COMPETITIONS

The Amazon Picking Challenge 2015
By Peter R. Wurman and Joseph M. Romano

he first Amazon Picking Challenge (APC) was held at the
2015 IEEE International Conference on Robotics and Automation (ICRA) in Seattle, Washington,
26-27 May. The objective of the competition was to provide a challenge
problem to the robotics research community that involved integrating the
state of the art in object perception,
motion planning, grasp planning, and
task planning to manipulate real-world
items in industrial settings. To that end,
we posed a simplified version of
Amazon provided
the task that
many humans
US$26,000 in prize
face in waremoney for the
houses all over
the world, i.e.,
winning teams.
picking items
from shelves
and putting them into containers. In this
case, the shelves were prototypical pods
from Kiva Systems, and the picker had
to be a fully autonomous robot.
The items were a preselected set of
24 products that were commonly sold
on Amazon.com and that we expected
would pose varying degrees of difficulty for the contestants. On the easier end
were simple cuboids like a box of
straws or a spark plug. Some items were
chosen because they were easy to damage, like the two soft-cover books or the
package of crushable Oreo cookies.
Others were harder to perceive and
grasp, like the unpackaged dog toys or
Digital Object Identifier 10.1109/MRA.2015.2452071
Date of publication: 11 September 2015

IEEE ROBOTICS & AUTOMATION MAGAZINE

the black mesh pencil holder. The box
of Cheez-Its posed a challenge because
it could not be removed from the bin
without twisting it sideways.
Each pod had 12 bins, and the 25
products were distributed among the
bins in such a way that each competitor
had the same challenges. Each bin had
one target item to be picked, with a base
score of ten, 15, or 20 points depending
on how many other items were in the
bin. In addition, some items that were
projected to be more difficult to pick
were given one to three bonus points.
Damaging an item incurred a five-point
penalty, while picking the wrong item
incurred a 12-point penalty. Each competitor had 20 min to pick as many of
the 12 target items as possible and could
score as many as 190 points.
The competition was announced
1 October 2014. Through a series of
video submissions, the organizers
selected 25 teams to receive equipment

grants (sample pods and products) and
travel grants to help defray the costs of
travel to the venue. In addition, Amazon provided US$26,000 in prize
money for the winning teams.
Representing 11 different countries, 26 teams made the trip to Seattle
to try their robot's hand at picking out
of Kiva pods. The success of the teams
was mixed, but the enthusiasm and
excitement was contagious. The competition was won by RBO from the
Technical University of Berlin, Germany. Its device, with a Barrett arm, a
Nomadic Technologies mobile platform, and a suction cup attached to a
commercial vacuum cleaner, was able
to successfully pick ten of 12 correct
items in under 20 min. Their score of
148 points put them well into the
lead. The Massachusetts Institute of
Technology (MIT) team placed second, with seven items picked and 88
points (Figure 1). The MIT entry used

Figure 1. A crowd gathers to watch Team MIT's entry. MIT's robot placed second after
picking seven items in 20 min. (Photo courtesy of Joseph Romano.)

September 2015

http://www.Amazon.com

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