Page 14 - Read Online

Page 12 of 26 Jin et al. Soft Sci 2023;3:8 https://dx.doi.org/10.20517/ss.2022.34

Table 2. Tactile sensors for object properties recognition
Object properties Processing methods Practical applications References
Surface texture FFT, bayesian exploration Materials classification [39]
DSWT Surface recognition [124]
Slide monitoring
STAM Textile classification [125]
LSTM Textile classification [66]
Braille reading
Stiffness k-NN, DTW Object recognition [126]
LSTM, RNN [43]
Health monitoring
Variation of the hertz model [127]
Thermal conductivity ANN Materials classification [128]
AC method Chronic wound management [129]
CTD feedback circuit Garbage sorting [29]
Chemical substance LPF Sweat analysis [130]
NN Practical surgical compensation [131]
LPF Human hand proximity [132]
Inertial parameters Cross-correlation analysis Object recognition [46]
CoM line calculation Grasping pose adjustment [133]
FNN Grasping position choosing [134]
Shape GP Object shape reconstruction [135]
SVM Object recognition [136]
iCLAP [137]
k-NN [91]
Pose TIQF Object pose estimation [138]
CNN [139]
CNN Manipulation positioning [140]
Internal properties Sparse GPs Liquid viscosity estimation [141]
GPIOIS Inner-outer shapes estimation [142]
FFT: Fast fourier transform; DSWT: discrete sequence wavelet transform; STAM: spatio-temporal attention model; LSTM: long short term
memory; k-NN: k nearest neighbors; DTW: dynamic time warping; RNN: long short term memory; ANN: artificial neural network; AC: alternating
curret; CTD: constant temperature difference; LPF: low-pass filter; CoM: center-of-mass; FNN: feedforward neural network; GPs: gaussian
processes; SVM: support vector machine; iCLAP: iterative closest labeled point; TIQF: translation-invariant quaternion filter; CNN: convolutional
neural network; GPIOIS: gaussian process inner-outer implicit surface model.

In addition to mechanical tactile sensing, thermal or chemical perception is also significant for robotic
manipulation, and promising applications such as object recognition and health monitoring have been
[129]
[148]
reported. Temperature can be considered another significant parameter when monitoring tactile behavior,
which makes robotic tactile sensing more similar to human beings. Principles such as resistivity, Seebeck
effect, pyroelectricity, and thermochromism have been widely applied in temperature sensing [4,108] .
Furthermore, object thermal conductivity detection relies on monitoring time-varying heat flow, which
combines the temperature sensor with the heat source . Other external local properties such as humidity,
[29]
[149]
gas concentration, or liquid component describe superficial or ambient characteristics of the objects . In
most cases, these properties can be measured directly or extracted after noise elimination processing such as
[51]
[131]
Low-Pass Filter (LPF) . Popular applications such as scenarios like health monitoring , baby care , and
[130]
human proximity detection have been reported.
[132]
Global properties recognition
Global properties can describe the object holistic parameters regardless of partial or superficial
characteristics. In most cases, vision device shows excellent advantages in global properties sensing, but