IEEE Robotics & Automation Magazine - June 2018 - 56

attenuate the other colors. The images
are sent to a classification algorithm
based on a convolutional neural network that determines if glue is present
on the parts.
Safety
To achieve a safe interaction, the Kuka
KR 5 SI robot implemented by MRKSysteme was used for the gluing process at Audi. Although different robots
(a)
(b)
were suitable for the use case, such as
the Kuka lightweight robot or the UniFigure 6. (a) A camera view and (b) a segmented depth image of the glued reinforcement
versal Robots UR5, we were limited to
plates. White pixels represent near points and dark pixels represent distant ones. (Photo
courtesy of the Audi Brussels factory.)
the KR 5 SI (SafeInteraction) from
MRK-Systeme. It was the only robot
during the project's preparatory phase
certified by the Berufsgenossenschaft Holz und Metall (the
Wood and Metal Professional Association) and, therefore,
the only one allowed on the Audi manufacturing floor for
safe human-robot interaction.
Figure 7 shows the MRK-Systeme robot gluing the reinforcement plates in the Audi Brussels factory. The robot is a
suitable platform, since it can operate without separate safety
equipment. Safety is based on various functional mechanisms. The work envelope and speed of the robot are controlled according to the safety requirements of the standard
DIN EN ISO 10218-1 [14], such that the applied forces stay
below the biomechanical limits. Also, integrated cushioning
Figure 7. The MRK-Systeme robot during the gluing process.
The safety skin of the robot detects human contacts. The end
elements are used to reduce contact forces. The robot stops
effector is framed by a plastic case. (Photo courtesy of the Audi
in case of collisions with humans. These are detected by a
Brussels factory.)
safety skin, as shown in Figure 7. The glue gun, which is a
potentially hazardous part because of its pointed shape, at the
F18 scanner. This was done once a day per person, allowing end effector is framed by a plastic case so the required safety
for checking if the worker was authorized to work with level is achieved.
the robot.
Improved Quality
Glue and Part Detection
The integration of the cobot Walt allowed a quality improveTo give the robot visual feedback during the gluing process, ment of the produced cars. The gluing quality of the reinRoboSense cameras were integrated with the setup to detect forcement plates of the assembled car is determined
the reinforcement plates on a rack. These cameras are mount- according to flatness measures. The number of cars that do
ed on an unactuated beam that moves loosely together with not fulfill the flatness criterion decreased by 15%. Moreover,
the robot to prevent robot collisions. Therefore, this intelli- the total quantity of glue used during the process, i.e., the
gent camera system was not only used for parts and glue quantity used for a determined number of cars, decreased
detection but also to find the rack that holds the parts, inde- by 20%.
pendent of its position and pose.
Object detection is performed by segmentation, using Foolproof Design
3-D images and heuristics to find the table pattern formed by The cobot system was designed to reduce undesired faults as
the placeholders of the parts. First, a threshold is applied to much as possible, using robust technologies. For instance, for
the depth image provided by the structured-light camera to gesture recognition, a confirmation with a thumbs-up is
identify the metal parts. These are located at the same height required to validate the command gesture. In the case of a
above the table. Figure 6 shows the camera view, along with robot misinterpretation, the operator can negate the comthe obtained segmented depth image. The latter is mapped to mand with a fist gesture, as shown in Figure 5. Moreover, to
the two-dimensional image of the RGB camera to extract reduce misuse by new operators, information regarding the
image patches for every reinforcement plate. A color fil- robot usage is displayed in the factory. Markers are also
ter is applied to accentuate the black color of the glue and placed on the floor to indicate the interaction zone. During
56

IEEE ROBOTICS & AUTOMATION MAGAZINE

June 2018

IEEE Robotics & Automation Magazine - June 2018

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2018