IEEE Robotics & Automation Magazine - December 2022 - 128

structure needs to be designed more flexibly to make better
contact with the object. Numerous studies have proposed
new types of sensors to address this issue [7], [8]. Visionbased
tactile sensors adopt an elastomer structure and can
perceive continuous contact features [9]. They have been successfully
applied to robot manipulation tasks [10]. By simply
changing the coating material on the surface, sensors based
on the same principle can even detect other modes of information,
such as temperature [11]. Magnetic tactile sensors
can also be constructed with elastomer and have the same
ability to detect continuous contact forces and contact positions
[12], [13].
For two important criteria for robot tactile sensors, the
resolution and sensing accuracy, magnetic tactile sensors
can achieve the desired performance [14]. In addition, a
flexible surface means
more contact area that
can provide better mechanical
compliance and surface
friction coefficient.
Compared to other kinds
of tactile sensors previously
discussed, soft
magnetic tactile sensors
or skins are also a potential
method for detecting
touch and local deformations
of a soft robotic
hand [15], [16].
However, a major
challenge with this kind
of tactile sensor is the
extraction of tactile features. One way to solve this challenge
is to optimize the sensor magnetic field distribution. In the
early stage, researchers usually used cylindrical magnets and
cast silicone elastomer to manufacture magnetic tactile sensors
[17]. However, magnetic tactile sensors based on this
scheme are usually used for detecting contact force information.
Gradually, sensor schemes began to evolve into mix
permanent particle and elastomer [18]. Compared with the
previous scheme, this scheme can obtain more tactile features
and contact position information. On the other hand,
researchers are also trying to optimize the performance of
sensors by changing the distribution of magnetic fields. By
combining with the Halbach array, Yan et al. [14] improved
the resolution of the magnetic tactile sensor. Gui et al. [19]
designed a Halbach-cylinder-based magnetic skin for tactile
sensing. In this article, we use a multipole magnetization
method to improve sensor performance. Compared to the
Halbach array, the process of our magnetization method is
simpler, and, compared with the general magnetization
method, the sensitivity and signal-to-noise ratio (SNR) of the
sensor are improved.
Meanwhile, for this kind of sensor, the neural network
(NN) method usually calibrates the sensor. It requires many
haptic sample data; for example, Hellebrekers et al. [12] used
128 * IEEE ROBOTICS & AUTOMATION MAGAZINE * DECEMBER 2022
The experimental results
show that our method can
achieve more than 80%
accuracy in contact shape
recognition and more than
95% accuracy in contact
pose recognition.
3,000 × 100 samples to calibrate the contact position, and
Yan et al. [20] used 90 × 90 × 5 sets of data to realize a tactile
superresolution model. For a new object, the result of sensor
output will produce a large error. To solve this limitation, we
combined metric-based metalearning to process sensor data.
It can reduce the amount of required data at the state of
extracting tactile features. It also allows sensors to extract
tactile features based on previous tactile data. We verified
that this method can recognize the shape and position features
of the new contact object based on the existing tactile
features. It is useful to extract the contact features from tactile
data. For example, the shape features of the contact surface
are used for slip detection and in-hand manipulation tasks
[21], [22], and the texture features are essential to material
recognition tasks [23]. Therefore, we attempt to extract the
tactile feature of the contact object from raw tactile information
based on existing data.
To sum up, the presented magnetic tactile sensor includes
the following contributions:
●
●
A new multipole magnetization method is used to optimize
the performance of magnetic tactile sensors.
Combined with the metalearning method, the data collected
by the magnetic tactile sensor are processed. This
allows the tactile sensor to use previous knowledge of contact
features to predict the contact state of the same type of
object. It reduces the time required to collect data when
using sensors.
●
The effectiveness of the sensor and the generalization ability
of the proposed model are verified by a series of experiments.
The sensor has the ability to classify different
contact shapes and poses.
Materials
Sensor Design
Figure 1 shows the configuration of the proposed magnetic
tactile sensor. The sensor consists of four parts: a magnetic
layer, silicone layer, pedestal, and 3D Hall effect sensor array.
The components of the magnetic layer are permanent magnetic
particles [LW-Q-100 (the particle size is 100 meshes),
manufactured by Xinnuode Company] and silicone rubber
(Ecoflex 0030, manufactured by Smooth-on, Inc.). The silicone
layer is made of the same silicone material as the magnetic
layer, which is between the magnetic layer and pedestal
to increase the deformation of the magnetic layer when in
contact with an object. The pedestal is used to hold the Hall
effect sensor and silicone layer, and the sensor array is used to
measure the change in the magnetic field. We selected five
locations to measure the magnetic field values. The five locations
are uniformly distributed on the sensor surface to detect
the changes in the magnetic field of the entire magnetic layer.
We placed an array of four 3D Hall effect sensors
(MLX90393, manufactured by Melexis, Inc.) at each measurement
position.
The principle of this sensor is as follows: first, the magnetic
layer produces deformation when the magnetic sensor is in
IEEE Robotics & Automation Magazine - December 2022

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