IEEE Robotics & Automation Magazine - December 2022 - 125

Table 2. The Denavit-Hartenberg parameters of
the robotic rat.
Joint Number ai 11
2
3
4
a1
a2
a3
ai
10
di
ii
i1
i2
i3
Description
Rear
waist
Front waist
Head
- - Whisker base
scope and maintaining the measurement accuracy. In particular,
the contact information is received by the whisker
sensor and transmitted to the motion control system.
After being processed, the signal is split into three
instructions, each of which controls the yawing of the
head, yawing of the waist, and differential rotation of the
wheels, respectively. The movement of the robot chassis is
mainly responsible for maintaining a certain distance
between the robot's centroid and the wall, whereas the
yawing actions are used to confirm that the whisker tip is
in contact with the wall and for making the robotic rat
movement more flexible.
In this scenario, the model that controls the yawing of
the head and waist can be considered a single-input, multiple-output
one, and because the head and waist rotations
have a combined effect on the base position of the whisker
sensor, we simulated the impact of a series of head-waist
correlation functions on the performance of the wall-following
motion in a simulated environment. Each simulation of
the robotic rat passing through a complex environment
while using different wall-following motion strategies were
evaluated. The results show that an inversely proportional
movement relationship performs well with respect to the
desired features of a high degree of followability and low loss
of feature information, as evident in Figure 9(b).
Applying the optimal control strategy, we conducted
experiments on the robotic rat in the real world. In the
experiments, a wall whose contour contained multiple features
was fabricated and fixed in the experimental site, as in
Figure 9(c). Specifically, the contour of the wall was an arc
with a radius of 60 cm and a concave angle of 45° followed
by a convex angle of 45°. By separating these features, we
can assess the GOF of the reconstruction for each one. To
evaluate the wall contour reconstruction similarity, we calculated
the weighted average of the GOF based on the number
of each feature, and the wall contour reconstruction
similarity was computed to be 91.76%. Note that moving
along a smoothly changing contour is more stable, whereas
drastic movement happens when the distance changes suddenly
because the wall contour changes sharply. Especially
for the convex corner, the whisker will snap off the wall
when executing this strategy continuously and lose contact
with the wall. As a result, the robotic rat will move away
from the wall and lose its wall-following capability.
Conclusion
This article presented the development of a small-scale
whisker sensor to establish the tactile perception of a
robotic rat. The sensation ability of the whisker sensor
was thoroughly demonstrated by the results of the experiments
described previously. Feature extraction methods
and an SVM algorithm were used to enable the robotic rat
to effectively identify the different texture surfaces of
common writing paper, sandpaper, tissue paper, and flannelette
by using its whisker sensor, with an accuracy of
about 88.3%. In addition, EBB theory was used to obtain
the distance information of the whisker contact point and
help the robotic rat recognize the contour of objects.
Semicircular contours, which represent shapes with positive
curvature, can be reconstructed well, with a GOF of
97.14% in our experiments in which the robotic rat
sweeps over semicircular objects. As for objects with zero
curvature, such as rectangular ones, the reconstruction
can be completed with a high GOF of around 99% after
geometric optimization.
Taking advantage of
the robotic rat's flexible
DoF configuration, a wallfollowing
control strategy
was proposed. Because
the whisker sensor provides
location information
about the contact
points with the wall, using
contour reconstruction, a
wall-following motion
can be achieved by controlling
the robot's movement
and adjusting its
body posture on the yaw
plane. The positioning
information provided by
the whisker sensor is an input signal to our control system,
which has three outputs that turn the head, turn the
waist, and rotate the wheels to control the motors simultaneously
and maintain a constant distance between the robotic
rat and wall.
In simulation and experiments, the results show that the
Using contour
reconstruction, a wallfollowing
motion can be
achieved by controlling
the robot's movement and
adjusting its body posture
on the yaw plane.
robotic rat can move along the wall by using the whisker
sensor. Moreover, the robot has relatively more stable locomotion
when moving along a smoothly changing contour
than when moving past a corner in the wall contour. In
addition, the high value of the wall contour reconstruction
similarity (91.76%) provides a feasible method for the robot
to perceive its environment under some extreme conditions.
In future work, more application scenarios in which
tactile sensation can be used will be explored. We will also
explore a visual-tactile sensing system in which the visual
system is used to perceive the long-distance environment
and the whisker sensor is able to contribute information
about short distances.
DECEMBER 2022 * IEEE ROBOTICS & AUTOMATION MAGAZINE *
125
IEEE Robotics & Automation Magazine - December 2022

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