IEEE Robotics & Automation Magazine - December 2018 - 63

xb
yb
zb
xsp
ysp
zsp

Pose and Set-Point of the Aerial Platform
1.2
1
0.8

1.4
Position (m)

1
0.8

0.6
0.4
0.2
0

0.4

-0.2

0.2

40

0

30
0.4
0.2

Position
on y (m)

0
-0.2
-0.4
-0.6

Contact
Engage
0

20

40

60

1.5
1
Position x (m)

0.5
0
5

φb
θb
ψb
µsp

20
10
0
-10
0

Brush Activated

20

40

60

80

Engage

Disengage
Disengag

80 100
Time (s)

120

140

Brush
Disengage
Activated

160

Figure 6. The end-effector position over time in a surfacecleaning experiment. Note that the end effector is positioned on
a single point from t . 60 s to t . 110 s.

z sp and z e, which affected i b and resulted in temporary
contact loss for the bottom wheel. In this experiment, the
operator manually lowered z sp to reduce this effect. After
the system stabilized, the operator started the contact phase
at t = 67.8 s.
During the contact phase, the controller depended
entirely on angular measurements, and the position setpoints were reinitialized to accommodate the disengage
phase. The operator gradually increased i sp to the desired
40° and activated the brush at t . 90.7 s. From t = 110 s to
t = 150 s, the operator actively controlled the locomotion
of the ground system, moving it across the surface in the
pattern illustrated in Figure 6, effectively cleaning the wall.
In Figure 8, we observe that, when pure horizontal velocity
commands are given, slight vertical displacements of the
end effector also occur and vice versa. These displacements
are caused by occasional slipping of individual wheels and
by undesired rotations of the end effector on the surface.
Although these displacements were seemingly small, the
operator in several experiments had to compensate for
them. After successfully cleaning the wall, the operator initiated the disengage phase at t = 154 s, returned the system
to the free-flight phase, and safely landed it. We observe
that, over the entire period in the contact, phase, z b and } b
never exceeded 7.5°.
One of the results of the second set of experiments is
shown in Figure 9. In this experiment, 18 consecutive trials of the multimodal locomotion procedure were

100 120 140 160 180

Time (s)
Contact

Figure 7. The pose of the multirotor and the angle of the
manipulator plotted over time in the surface-cleaning
experiment of Figure 6.

Displacement (m)

0.6

Angle (°)

0.6

Velocity Command

Position z (m)

1.2

xe
ye
ze

0.2
0
-0.2
-0.4
0.5

vysp
vzsp

0
-0.5
100

110

120

130
Time (s)

140

150

160

Figure 8. The relative displacement of the end effector and the
associated velocity commands during the ground-locomotion
phase in the experiment of Figure 6.

performed over 20 min. Figure 9(a) shows the surface
motion trajectories of each trial. Except for trials 7 and 16,
the motion profiles were quite similar. However, we
noticed a sideways drift in the first point of attachment
over the course of the experiment. We were unable to
identify the cause of this drift. The controller effectively
withstood disturbances and maintained stability in all 18
trials. In Figure 9(b), the orientation of the multirotor and
the pose of the end effector are shown for trials 6, 7, 15,
and 16. In trial 7, a coincidental bounce with the
December 2018

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

63
IEEE Robotics & Automation Magazine - December 2018

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