IEEE Robotics & Automation Magazine - December 2016 - 103

Tr
an

ov
e

Tr
a
C ns
on fe
ve r to
yo
r

sf
er

ra
sp
i

pr
oa

Tr
an

ov
e

Tr
a
C ns
on fe
ve r to
yo
r

sf
er

ov
Po e to
se
R
ec
og
ni
tio
n

Time (s)

ns M
in ov
g e
Po to
se
R
ec
og
ni
tio
n

Time (s)

Containers were packed with 70-kg
45
coffee sacks [Figure 7(a)] in a manner consistent with typical shipping
40
conditions. A video of the system
35
performance can be found on the
30
YouTube channel RobLog (https://
25
youtu.be/U7_lxH-qB5I). The aver20
age cycle time for unloading a sin15.63 (26.6%)
15
10.89 (18.5%)
gle coffee sack was 67 s (σ = 8.2 s,
12.19 (20.8%)
10
median 58.7 s), with all coffee sacks
8.43 (14.4%)
6.56 (11.2%)
successfully unloaded in our tests
4.81 (8.2%)
5
0.23 (0.4%)
(N = 58 for this evaluation). The
0
damage to the goods in this scenario was judged by examining the tear
pattern of the coffee sacks at the
output side and marking the number of sacks that would not be reus(a)
able for transportation. Throughout
100
all the tests of the platform, 7% of
the sacks were deemed unusable
80
after unloading.
The current unloading practice
60
at Vollers GmbH uses four employees, i.e., two workers using two
40
hooks each for unloading, a forklift
37.1 (18.7%)
30.7 (15.4%)
driver, and a weight-scale operator.
41.2 (20.7%) 41.3 (20.8%)
12.0 (6.0%)
A fully packed 20-ft container can
20
16.3 (8.2%)
transport up to 24 tons of coffee
12.13 (6.1%)
with roughly 300 to 400 sacks and is
usually unloaded in approximately
30-45 min, on average. Thus,
human-picking rates are currently
about 6-10 s per sack (with four
(b)
workers), which is an order of magnitude faster than the autonomous
Figure 8. A breakdown of the run times of the two technology demonstrators. (a) The
system. According to our industrial
time taken by each system component in the coffee-sack unloading scenario. (b) The
corresponding component-wise time breakdown for the heterogeneous goods unloadpartner, taking into consideration
ing scenario. Plots are shown as box plots, in which each box is centered at the average
that the autonomous solution can
time value and extends to cover the 25th and 75th percentile; the thick line denotes the
be employed over longer shifts, the
median; and the whiskers extend over the full range of inliers. The scatter plot on the left
of each box is the data-samples distribution.
current system is already economically viable in some use cases. Furthermore, the automation of unloading is highly desirable clusters of grasp run times are also easily identifiable in this
for the improvement of working conditions in the handling plot, with the more time-consuming cluster corresponding
of coffee and similar goods such as cacao-especially in the to the more difficult grasps of sacks near the container
context of labor regulations worldwide becoming increas- floor. The second most time-consuming module is OR, folingly strict with respect to the maximum amount of lifting lowed by the motion execution-related components. Folallowed during the daily shift of a human worker. Coffee- lowing these results, to achieve faster cycle times, future
handling operations in the European Union, for example, improvements of the system would primarily focus on optiwhere most of the coffee roasting and processing takes mizing the grasp-acquisition procedure.
place, will be forced to switch to automation as soon as the
currently practiced operations are no longer compatible Unloading Loose Goods
with labor laws.
As mentioned before, the unloading of heterogeneous
To identify how the system cycle times can be improved, loose goods was selected to evaluate the system perforwe examined a breakdown of the run time per component, mance on more scientifically challenging problems. The
as shown in Figure 8(a). Clearly, the system spends a large relatively light (up to 5 kg) and loosely packed goods in
portion of the cycle time in grasp acquisition. Two separate the container would present little difficulty for a human
December 2016

IEEE ROBOTICS & AUTOMATION MAGAZINE

103

http://https:// http://www.youtu.be/U7_lxH-qB5I

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