IEEE Robotics & Automation Magazine - December 2022 - 87

directly actuated by the servo motor, an additional holder is
located, which is connected by a universal joint. This
mechanical configuration results in all holders being actuated
in synchrony [see Figure 7(c)].
A feather shape is fabricated by cutting a polypropylene
sheet on a laser cutter. Each feather is bolted to an acrylic
connector, which can be slotted into the feather holder. Each
holder can be equipped with a solenoid that drives the
detachment mechanism [Figure 7(b) shows how feathers
can be attached to or detached from the layer]. The feather
detachment mechanism is described in Figure 7(d). When a
solenoid attached on the feather holder is energized, a pin
holding the feather in place is unlocked, allowing a leaf
spring (a 3D-printed flexure) to push out the feather from
the holder. This allows feathers to be detached rapidly and
on demand.
In this experiment, two layers were fabricated and assembled
(see Figure 1). Electrical wires form the servo motor, and
solenoids connect the waterproofed robot to the Arduino
microcontroller located outside of the water. Solenoids are
energized through MOSFETs, which can be turned on and off
via the Arduino. The Arduino is connected to a PC, where the
feather detachment and the servo motion profile are commanded
by a human operator.
Experimental Results
Feather Optimization
To demonstrate and experimentally validate the optimization
of a single feather on the robot, a single ring of the
robot was tested with different feathers and controllers. The
single ring was tested with the " optimized " feathers (120 ×
40 mm) and feathers with the same area but longer (192 ×
25 mm) to provide a baseline. For these two robot configurations,
three different controllers were tested: the two top
controllers from the experimental validation of the optimized
feather (Figure 5), i.e., controllers 1 and 2, and controller
3, which was poorly performing in both simulation
and the single feather experiment. The neutrally buoyant
robot was placed in a downward-facing configuration, with
a thin rod placed through the center of its body to constrain
its motion to allow for comparison among experiments. For
each combination of feather geometry and controller, the
robot was recorded swimming downward until the bottom
of the tank was reached (or the time exceed 1 min) for five
repetitions, where the average swimming velocity was
obtained (Figure 8).
The results, in Figure 5, show that the optimized feathers
significantly outperform the baseline feathers, with the
speeds generated by the top two controllers offering more
than a 60% increase relative to the baseline feathers. The
experiments reinforce that controller 1 shows the greatest
thrust generation and, hence, velocity. However, the relative
performance of controller 2 is poorer than expected, with
controller 2 showing a velocity, on average, 10% lower than
controller 1. Compared to the single-feather case, the servo
motor on the robot must exert a higher torque to move all
the feathers. Without a sufficiently long hold time, the true
position of the servo motor in some cases does not reach the
desired position (for both up and down strokes). Hence, we
believe that controller 2, with a longer hold time, is less susceptible
to environmental disturbances and noise, which
cause the up and down strokes to not meet the desired
angles. In addition, it was clearly observed that when using
controller 1, the true feather amplitude is less than the servo
motor demand position (A = 40º). This is because of the
lack of hold time in controller 1, which means the servo
demand position direction is reversed before the motor has
time to reach it. In comparison to the top-performing controllers,
controller 3 struggled to produce any downward
thrust for both feather geometries.
Single-Ring Control of Motion
To demonstrate and explore the degree of controllability,
we first present one single layer of the robot. While the
range of motion is lower than for multiple layers, it allows
the concepts to be explored on lower complexity hardware.
To demonstrate how varying the body structure leads to
different behaviors, we first show the motion of the robot
with different fixed feather configurations [Figure 9(a)-
(d)]. For these and all following experiments we aimed to
perform them in the center of the tank as much as possible,
and if the sides of the tank influenced the experiment,
the experiment was repeated. These results show that
different robot configurations lead to different motions
and directions of swimming. Fully symmetric designs
[Figure 9(c)] and partially symmetric designs that produce
a net thrust at the center of the layer [Figure 9(d)]
lead to the robot swimming in a straight line, whereas
asymmetric designs [Figure 9(a) and (b)] allow for gradual
turning arcs.
To demonstrate how we can use this variation in motion
with body shape, we extend the experiments to include the
detachment of feathers to transition among different
motions. As shown in Figure 9(e)-(h), the detachment of
feathers results in a change in heading of the robot. By altering
both the starting configuration and the selection of
feathers to be detached, the motion before and after detachment
can be controlled. Figure 9(e)-(h) shows how the
robot transitions from one stable motion to another stable
motion, where the change is caused by feather detachment.
These experiments demonstrate that even with the same
controller, the change in body (detachment) leads to a
change in motion. In some cases, the robot moves close to
the edge of the tank, where the fluid behavior may be more
complex. Although the dominant effect on the maneuverability
is still from the feather detachment, in open water,
the exact trajectory may differ.
Two-Layer Robot Demonstration
The full two-layer robot allows for a complete demonstration of
the analytical results described by the optimization problems in
DECEMBER 2022 * IEEE ROBOTICS & AUTOMATION MAGAZINE *
87
IEEE Robotics & Automation Magazine - December 2022

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2022
