IEEE Robotics & Automation Magazine - December 2018 - 62

stabilizes the rotational dynamics of the system. This results
in gain matrices of the following shape:
RK 1, 1 K 1, 2 0
0 K 1, 5 K 1, 6VW
S
0 K 2, 3 K 2, 4 0
0 W
S 0
K=S
,
(9)
K 3, 1 K 3, 2 0
0 K 3, 5 K 3, 6W
SS
WW
0 K 4, 3 K 4, 4 0
0 X
T 0

which indicate a distinct separation between the roll and yaw
states, stabilized using the roll torque and yaw torque, and
the pitch state, stabilized by the pitch torque and the variation in thrust.

Experiments
Experiments were performed to evaluate the multimodal
approach toward surface cleaning (presented in the "Multimodal Aerial Locomotion Approach" section) and the control
strategy (presented in the "Control Strategy" section), which
are reported here. We have included a supplementary multimedia file showing the experiments. This is available at http://
ieeexplore.ieee.org.
Experimental Setup
The experimental setup consisted of an aerial manipulator, a
ground-control station, an Optitrack motion capture system,
and a vertical surface. As shown in Figure 1, the aerial
manipulator comprises a hexarotor platform equipped with a
single actuator manipulator, which carries the end effector
presented in the section "End-Effector Design." The system
was controlled by an onboard Intel NUC i5 computer that
communicates with the ground-control station over a wireless network. The ground-control station provided the user
interface to the operator. The motion capture system
obtained absolute pose measurements, which were used for
the multirotor's state estimation algorithm and to obtain
experimental measurement results.
The hexarotor we used is illustrated in Figure 1. Its diameter (excluding propellers) is 80 cm, and it weighs 2.1 kg. A
frame with long arms was chosen deliberately to increase
the gap between the front propellers and thus allow the
manipulator to pass through. The hexarotor was controlled
by a Pixhawk 2.1 flight controller, running PX4 Firmware
[17]. Its propulsion system consisted of Cobra ESC CM2217
950-Kv motors and 10 # 4.5 -in dual-blade propellers. The
aerial system was powered by a tethered 16-V power supply.
At this voltage, the configuration can provide a maximum
Fu of 78 N.
The manipulator consisted of a Dynamixel MX106R servo
motor, which rotates along yt b, and a hollow carbon-fiber tube
connected on top of this servo. The tube had a length of 60 cm,
an inner diameter of 10.5 mm, and an outer diameter of 12 mm,
which provided a sufficient stiffness for us to assume negligible
deflection of the tube, given the weight of the end effector. At the
end of this rod, a 3-D-printed bend is attached that applies an
angular correction because the rod is not exactly aligned with
the vector vp be . This bend connects to the elastic component of
the end effector. The manipulator weighs 0.23 kg.
62

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

December 2018

Experiment Description
Two sets of experiments were conducted in a confined flying
arena where a wall was placed at x w = 1.6 m in the first set of
experiments and x w = 1.75 m in the second set. In both
cases, the wall was aligned with the plane (yt w, zt w) such that,
in interaction, W w and W e have identical orientations.
In the first set of experiments the goal was to clean a
scribble from the wall. This scribble was drawn on a patch
measuring 5 cm wide and 10 cm high on the wall. An operator controlled p sp during the free-flight and engage phases
and i sp during the engage and contact phases. To establish
quick and reliable contact, i sp was set to 20° before entering
the engage phase. Regarding the end effector, the operator
activated the brush and controlled the lateral movement by
giving lateral velocity commands. The operator initiated the
engage, contact, and disengage phases in the experiment.
The disengage distance C was set to 0.75 m.
In the second set of experiments, the repeatability of the
approach was qualitatively evaluated. The switching control
procedure was automated and repeated for several runs, each
lasting 64 s. This procedure was as follows. First, the surface
was approached by incremental adjustments of the p sp in the
free-flight phase. Then, the engage phase was initiated. Rather
than using an immediate 20° set-point, i sp was gradually
increased from 12° to 20° over a period of 4 s to reduce the
shock of impact. Then, the contact phase was initiated, and
i sp was gradually increased further to 27°. An up-down locomotion was performed by sending a velocity command of
0.125 in both directions for 6 s, with a pause of 2 s in between,
after which the disengage phase was initiated.
Results
In the first set of experiments, several trials were performed
under contact angles varying between 25° and 40° to qualitatively evaluate the performance and reliability of the system in
interaction. In all of the experiments, the operator was
able to remove the scribbles on the wall by controlling the
ground locomotion.
The results of one trial of the first set are displayed in
Figures 6, 7, and 8. Figure 6 shows the spatial position of
the end effector during the experiment; a part of the surface
is illustrated for clarity. In Figure 7, the pose of the multirotor and the n sp are plotted. Figure 8 shows the position and
velocity commands of the end effector. The start of different events are annotated in Figures 6 and 7.
The experiment started with the system lifting off at
t = 10 s and ascending to approximately 1 m. A small
steady-state error between z b and z sp was visible due to a
small error in estimating the system's mass and the lack of
integral action in the altitude control loop. The multirotor
was moved toward the surface, and at t = 57.4 s the operator started the engage phase: the system pitched forward
and successfully established contact. Before contact with the
surface was made, a small drop in height occurred, caused
by the end-effector mass not being taken into account in the
contact controller. This drop caused a mismatch between
http://ieeexplore.ieee.org http://ieeexplore.ieee.org
IEEE Robotics & Automation Magazine - December 2018

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