IEEE Robotics & Automation Magazine - December 2022 - 116

Tactile Perception
Environmental detection and task execution in narrow spaces
are arduous to achieve. A small-scale robotic rat proposed in
our previous studies [1], [2] has good potential for task execution,
owing to its excellent adaptability to space-constrained
environments. However, environmental perception in robots
is a prerequisite [3] for high-level tasks, such as group collaboration
and biological interaction. Currently, bioinspired sensors
are generally a topic of interest in the field of biomimetic
robots, and a whisker sensor is a reasonable solution for a
robotic rat. There have been several well-designed whisker
sensors [4], [5] developed for contact sensing, but most of
them are fixed on manipulators and have restricted working
space. Besides, their relatively large size also makes their
application scenarios limited. Concerning space-constrained
environments and the miniature size of the robotic rat, these
traditional whisker sensors are not suitable solutions.
Focusing on minisenBioinspired
sensors are
generally a topic of interest
in the field of biomimetic
robots, and a whisker
sensor is a reasonable
solution for a robotic rat.
sors manufactured by
unconventional processing,
there are lots of
whisker sensors being
designed and ultimately
used for weak environmental
perception, such
as airflow sensing [6]
and water flow sensing.
Applying SOI technology,
some whisker sensors [7]
have little utility due
to the fragility of their
microstructures. However,
mimicking the high flexibility of biological whiskers, continuous
and repeated detections are supposed to be executed
by whisker sensors without the risk of damage when colliding
with obstacles. To resolve these problems, we designed a sensing
unit inspired by a rat's follicle-sinus complex (FSC) [8], a
stimuli processing unit under the whiskers, and fabricated a
whisker sensor by combining it with an artificial whisker
shaft. With the ideal mechanical characteristics of small size
and lightweight, the whisker sensor can be integrated into a
robotic rat. Furthermore, we combined cantilever- and flagtype
structures into one to improve the sensor's reliability and
expand its application scenarios. Specifically, on the basis of
an SOI-fabricated whisker sensing array containing four sensing
units [9], we optimized the sensor's structure by developing
one sensing unit with the same functional properties that
is smaller in size.
To establish the tactile perception of whisker sensors, two
main functionalities are commonly developed: texture discrimination
and contour reconstruction. In terms of texture
discrimination, there are several classification methods that
can be applied. However, neural networks [10] rely heavily on
the amount of data, the Gaussian mixture model [11] has a
heavy computational load, and k-means [12] is sensitive to
outliers. Concerning the data set, we used a feature-based
116 * IEEE ROBOTICS & AUTOMATION MAGAZINE * DECEMBER 2022
SVM to create classifiers with high efficiency to improve the
robustness of classification. The data set is also relatively
memory-efficient because we need only to store the parameters
related to the support vectors. In contrast to studies that
provide only some methods for comparing textures [13] or
demonstrate a sensor's ability to discriminate textures [14],
our work successfully classified four different textures with an
accuracy of 88.3%. As for contour reconstruction, much
research relies on image comparison. In contrast to other
methods, which are based on visual similarity between tomographic
reconstruction [15] and object outlines, we calculated
the contours of scanned objects by using EBB theory and
compared the counters with the real ones to prove the accuracy
of reconstruction via the standard metric, the GOF.
In addition, it is commonly argued that one key unresolved
issue for whisker sensors is the low scope of reconstruction.
Only a few features or a small part of the objects
can be scanned and detected through the traditional one-way
scanning process. To solve the problem, we take advantage of
the robotic rat, which has a high number of degrees of freedom
(DoF) to implement highly flexible movement according
to real-time contour reconstruction. Using the perception
capability of a whisker sensor and with the corresponding
control strategy, the scope of reconstruction is increased, and
the precision is improved. To take this approach a step
beyond the detection of objects with regular shapes, a reconstruction
experiment for a complex wall was executed, and
the result has a high reconstruction similarity, demonstrating
the robotic rat's capacity for environmental perception
through tactile sensation.
Similar work was conducted by Pearson et al. [16]. Shrewbot
was developed to achieve simultaneous localization and
mapping. Even though Shrewbot had a whisker module of an
array with 18 whiskers, there were still sharp, concave corners
that had not been explored. Comparably, applying the robotic
rat, we demonstrated that more features can be acquired, and
a high contour reconstruction accuracy of 91.76% can be
achieved using one whisker only. This ability also provides an
approach to the problem of vision failure in some extreme
environments, such as those with low illumination.
In this article, we propose a rat-inspired whisker sensor for
a robotic rat and present its superior tactile perception performance.
The main contributions of the work are as follows:
1) We design a biomimetic whisker sensor for a small-scale
robotic rat to realize ratlike perception and fabricated it
with SOI technology to overcome shortages caused by the
size limitation of traditional sensors.
2) We demonstrate the effectiveness of the whisker sensor
with comparably high accuracy (88.3%) for the classification
of four textures and the desired GOF (above 97.14%)
for the contour reconstruction of obstacles.
3) We experiment with the robotic rat in a designed scene
with walls of multiple contour features and demonstrate
that the robotic rat has a great ability to pass through complex
environments and achieves high contour reconstruction
similarity (91.76%) for the intricately contoured wall.
IEEE Robotics & Automation Magazine - December 2022

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