IEEE Robotics & Automation Magazine - December 2022 - 136

By measuring the distance
between the prototype of
the support set and the
features of the unlabeled
samples, we can obtain the
contact surface shape and
contact pose classification
results.
between the prototype and sample and the similarity
between each prototype, which is useful for extracting tactile
features. For these two tasks, the proposed method achieves
more than 80 recognition accuracy. The effectiveness of the
proposed method is verified by experiments. In addition, it is
worth noting that the recognition accuracy of the contact
pose is higher than the recognition accuracy of the contact
surface shape. This is because the measuring head used for
contact surface shape recognition is usually smaller than the
measuring head used for contact pose recognition. In the
contact state, the magnetic strength changes caused by
smaller measuring heads
can only be detected by
fewer Hall effect sensors.
As a result, the contact
data generated for smaller
measuring heads contain
fewer tactile features.
Therefore, there is still a
certain gap between the
recognition accuracy of
the two tasks.
Another advantage of
the approach used in this
article is that it requires
fewer contact data. At the
same time, this method
makes the sensor have a
certain generalization.
Compared wi th the
machine learning method
and deep learning method, the metric-based method is more
efficient. The metric-based metalearning method has the
potential to be used for tactile feature extraction.
Conclusion
In this article, a novel magnetic tactile sensor based on the
multipole magnetization method is designed for extracting
contact surface features. The proposed magnetization method
offers the advantages of higher SNR and sensitivity, which
achieve better performance in tactile feature extraction.
Meanwhile, we proposed a method combined with metalearning
that enhances the perception ability and generalization
of the sensor. Two experiments are designed to simulate different
contact shapes and poses, and the experimental results
show that the presented metric-based metalearning method
has better classification and generalization performance compared
with deep learning methods. The Pronet-L model
works by utilizing the prior knowledge learned from previous
contact data to achieve new contact shape and pose classifications,
and the accuracy of the classification results reached
80% and 95%, respectively.
Although the proposed method could enhance the ability
of tactile feature extraction for a magnetic tactile sensor, it
still remains a challenge to perform complex tactile recognition
tasks (e.g., 3D object reconstruction from tactile
136 * IEEE ROBOTICS & AUTOMATION MAGAZINE * DECEMBER 2022
features). Therefore, it is necessary to research a better tactile
feature extraction framework based on a soft magnetic tactile
sensor. In addition, the structure of the sensor still has room
for improvement, including adjusting the distribution of the
Hall effect sensor array and the number of sensors. In future
work, we will further investigate these two aspects.
In conclusion, we introduce the process of magnetic tactile
sensor fabrication in this research. By optimizing the
magnetization for the magnetic layer, the performance of
this sensor is improved. Especially in terms of the SNR, we
collect the magnetic field variation in the same location.
Under the same loading conditions, an optimized magnetic
surface can obtain higher quality tactile signals. In addition,
by using the metric-based metalearning method, the tactile
sensor can extract tactile surface features from a small
amount of data. A series of experiments are carried out to
verify the feasibility of the Pronet-L model.
In future work, we would like to further optimize the
design scheme, including the magnetization mode and
structure, and extend our method to extract more tactile
features. We would also like to combine our magnetic tactile
sensor with dexterous hands to perform complex
manipulation tasks.
Acknowledgment
This work is supported by the National Key Research Program
of China (project 2019YFB1309803) and Tsinghua
University (Department of Computer Science and Technology)-Siemens
Ltd., China Joint Research Center for Industrial
Intelligence and Internet of Things. Bin Fang and Yiyong
Yang are the corresponding authors.
References
[1] J. Tegin and J. Wikander, " Tactile sensing in intelligent robotic
manipulation: A review, " Ind. Robot, vol. 32, no. 1, pp. 64-70, 2005, doi:
10.1108/01439910510573318.
[2] M. Bianchi, J. Bohg, H. Moon, R. Platt, and R. Walker, " Robotic
hands, grasping, and manipulation [TC Spotlight], " IEEE Robot. Autom.
Mag., vol. 28, no. 2, pp. 131-133, 2021, doi: 10.1109/MRA.2021.3071898.
[3] Y. Zhang, Y. Mukaibo, and T. Maeno, " A multi-purpose tactile sensor
inspired by human finger for texture and tissue stiffness detection, " in
Proc. 2006 IEEE Int. Conf. Robot. Biomimetics, pp. 159-164, doi: 10.1109/
ROBIO.2006.340351.
[4] F. Castelli, " An integrated tactile-thermal robot sensor with capacitive
tactile array, " IEEE Trans. Ind. Appl., vol. 38, no. 1, pp. 85-90, 2002,
doi: 10.1109/28.980361.
[5] A. Billard and D. Kragic, " Trends and challenges in robot manipulation, "
Science, vol. 364, no. 6446, p. eaat8414, 2019, doi: 10.1126/science.
aat8414.
[6] H. Yousef, M. Boukallel, and K. Althoefer, " Tactile sensing for dexterous
in-hand manipulation in robotics: A review, " Sensors Actuators
A, Phys., vol. 167, no. 2, pp. 171-187, 2011, doi: 10.1016/j.sna.2011.02.038.
[7] N. F. Lepora and J. Lloyd, " Optimal deep learning for robot touch:
Training accurate pose models of 3D surfaces and edges, " IEEE Robot.
Autom. Mag., vol. 27, no. 2, pp. 66-77, 2020, doi: 10.1109/MRA.
2020.2979658.
IEEE Robotics & Automation Magazine - December 2022

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