Organizations: Department of Mechano-Informatics, Graduate School of Information Science and Technology, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan. · AI Center, Graduate School of Information Science and Technology, The University of Tokyo, Japan.
Human skin contains diverse mechanoreceptors, including Merkel cells and Pacinian corpuscles, each localized in specific regions. Although their spatial organization is known to be shaped by molecular signaling and neural induction during development, experimentally examining its functional significance under controlled conditions in living systems remains challenging. In particular, physically comparing alternative receptor arrangements that do not occur biologically is difficult. In this study, we developed a layered biomimetic robotic skin designed to mimic key structural features of human skin tissue by embedding receptor-analog sensors within compliant soft media. Using this platform, we implemented both biologically motivated and counterfactual sensory arrangements and quantitatively evaluated their performance across three passive tactile tasks: detection of weak contacts, a two-point discrimination-like assessment, and texture identification. Arrangements reflecting key features of the biologically motivated sensory organization exhibited higher performance than the alternative configurations examined. This study addresses a biological question concerning how spatial organization influences tactile function by presenting a biomimetic robotic approach that enables physical and empirical comparison of biologically motivated and counterfactual sensory arrangements that are difficult to investigate in living systems.
Figures & tables
Figure 1: Structural and functional overview of the biomimetic skin system designed to replicate human tactile sensing. SA and RA denote slowly adapting and rapidly adapting responses, respectively. In human skin, Merkel cells exhibit SA responses, whereas Pacinian corpuscles exhibit RA responses. In this study, strain gauge sensors correspond to SA units, while accelerometers correspond to RA units. (a) Structure of human skin: A schematic diagram showing the layered anatomy of human skin—epidermis, dermis, and hypodermis—as well as mechanoreceptors (e.g., Merkel cells and Pacinian corpuscles) embedded at different depths. (b) Structure of the biomimetic skin: The biomimetic skin is constructed from soft, elastic materials, layered to mimic the structure of the epidermis, dermis, and hypodermis. Strain gauge sensors and accelerometers are embedded to replicate the distribution and depth of biological tactile receptors. As shown in (e), three different sensor embedding patterns were implemented for comparative analysis. (c) Tactile sensing in human skin: Static pressure is mainly detected by densely distributed Merkel cells, while rubbing stimuli are primarily sensed by Pacinian corpuscles, which respond to high-frequency vibrations. (d) Tactile sensing in the biomimetic skin: Distributed strain gauges emulate the function of Merkel cells to detect static pressure, while embedded accelerometers, mimicking Pacinian corpuscles, respond to vibratory stimuli caused by rubbing. (e) Sensor arrangement patterns: To compare different sensory organ arrangements, multiple sensor embedding patterns were implemented inside the biomimetic skin. The depth of the accelerometers was varied and used as a reference, around which strain gauge sensors were vertically embedded as densely as possible. These sensor arrangement patterns are evaluated in later experiments to investigate how the spatial pattern of sensors affects tactile sensitivity and discrimination performance.
Figure 2: Experimental setup for applying point stimuli to biomimetic skin using the HIRO-NX robot. (A) Overview of the setup. The HIRO-NX robot’s fingertip is equipped with a custom end-effector that delivers one- or two-point tactile stimuli to the biomimetic skin laid on a flat surface. The inter-point distance is varied by replacing the end-effector. (B) Experimental timeline. The robot applies stimuli with randomized variations in position and orientation, repeating the indentation cycle throughout the experiment.
Mean
Standard Deviation
A (pressed C)
4.83
2.54
B (pressed C)
8.02
3.55
C (pressed B)
36.39
16.32
Table 1: Mean and standard deviation of sensor responses in non-target patterns.
Figure 3: Experimental setup for rubbing stimuli with the PR2 robot. (a) Overview of the experiment. A material sample is attached to the fingertip of the PR2 robot and rubbed against the biomimetic skin placed on a horizontal plane. (b) The material is rubbed along the centerline of the sensory organ arrangement on the biomimetic skin (indicated by the red arrow). (c) Time sequence of the experiment. The robot lowers its fingertip, waits 3 seconds after contact, performs the rubbing motion, and then lifts off. This cycle is repeated.
Figure 4: (a) Experimental setup of the indentation test using the tabletop universal testing machine. (b) Averaged force–displacement curve obtained from 10 loading–unloading cycles. The shaded region represents the standard deviation. The dashed line indicates the estimated contact point ( xc=2.797\text{,}\mathrm{mm}$$ ).
Figure 5: Comparison of strain gauge sensor responses to deformation in the biomimetic elastic skin. The figure presents representative sensor outputs from embedded strain gauges under a two-point indentation with a 15mm inter-point distance. Three sensor embedding patterns (Pattern A, B, and C) are compared. For each pattern, responses from four surface-level sensors are plotted to assess how sensor placement affects sensitivity to localized pressure. The plots show a selected time segment from the experimental recordings. Different colors are used to distinguish individual sensors within each pattern for visual clarity and do not imply correspondence across patterns or differences in sensor function.
Condition
Sample size
Mean
SD
Median
Q1
Q3
A
397
7.298464
3.272834
7.195545
5.230198
9.537129
B
397
29.237356
14.931334
28.960396
17.573020
38.785891
C
388
21.651800
12.197818
20.090965
12.208540
29.542698
Table 2: Mean and standard deviation of strain gauge responses for the three sensor arrangement patterns (A, B, and C).
Figure 6: Classification accuracy for single- vs. multi-touch discrimination at varying inter-point distances. This plot illustrates the classification performance of a machine learning model distinguishing single-point and two-point stimuli at distances of 5mm15mm , simulating a two-point discrimination threshold test. Curves correspond to sensor arrangement patterns: black (A), blue (B), red (C); line styles indicate sensor usage—solid: sensors in A; dashed: only shallow sensors in B/C; dotted: all sensors in B/C.
Pattern
Detection Count (count/20)
A
0 / 20
B
17 / 20
C
3 / 20
Table 3: Detection rates of very light touch stimuli for each sensor arrangement pattern. The values indicate the percentage of correct detections out of 20 trials.
Figure 7: Sensor responses and spectral analysis during rubbing experiments. (a) Comparison of acceleration sensor waveforms for four materials (a1:metal, a2:cloth, a3:cork, a4:rough cork) rubbed against the biomimetic skin using the robot fingertip. The data shown here corresponds to a rubbing speed of 3cms−1 . (b) Example of frequency-domain analysis using Fourier transform. The plot shows the FFT result for the cloth material at a speed of 3cms−1 .
Speed
Pattern A
Pattern B
Pattern C
2cms−1
0.900
0.950
0.950
3cms−1
0.900
0.950
0.750
6cms−1
0.900
0.900
1.000
Table 4: Accuracy for each pattern across different speeds
Figure 8: Visualization of biomimetic skin responses to four materials across three sensory organ arrangements at different rubbing speeds. This figure shows the distribution of tactile sensor responses when four materials (cloth, cork, metal, and rough cork) are rubbed against the biomimetic skin by the PR2 robot at three different speeds: (a) 6 cm/s, (b) 3 cm/s, and (c) 2 cm/s. For each condition, spectral centroid (x-axis) and RMS amplitude (y-axis) are plotted to represent the frequency and intensity characteristics of the responses. Data from all three sensory organ arrangements (Pattern A, B, and C) are included, illustrating how both material type and sensor layout affect the tactile response under varying dynamic conditions.
Pattern
Speed
metal
cloth
cork
rough-cork
A
1000
36.041
9.493
4.589
3.432
2000
51.41
4.955
11.364
0.314
3000
16.843
8.494
1.842
0.585
B
1000
4.091
8.238
6.822
2.046
2000
1.796
3.893
2.83
1.006
3000
1.012
1.233
0.383
0.404
Table 5: Covariance Trace by Speed, Pattern, and Material (Top 20% Highlighted)
Figure 9: TDPI-based selection of the best sensory pattern under varying task weight combinations. (a) Heatmap showing the best-performing pattern (A, B, or C) based on the Tactile Discrimination Performance Index (TDPI) when all three patterns are compared. (b) Heatmap comparing only patterns A and C. The axes represent task weights: w1 (vertical) for texture classification, w2 (horizontal) for light touch detection, and w3=1−w1−w2 for two-point discrimination. The variability penalty coefficient for texture classification is fixed at λ=0.2 in both panels.
Figure 10: Classification accuracy for single- vs. multi-touch discrimination at varying inter-point distances ( 5mm15mm ) using Pattern A sensors. The curve is shown in black, solid line. Error bars indicate the standard deviation across trials. This corresponds to the Pattern A condition in the Main Figure (Fig. 6 ).
Figure 11: Classification accuracy for single- vs. multi-touch discrimination at varying inter-point distances using Pattern B (All Sensors). The curve is shown in blue, dotted line. Error bars indicate the standard deviation across trials. This corresponds to the Pattern B (All Sensors) condition in the Main Figure (Fig. 6 ).
Figure 12: Classification accuracy for single- vs. multi-touch discrimination at varying inter-point distances using Pattern B (Shallow Sensors Only). The curve is shown in blue, dashed line. Error bars indicate the standard deviation across trials. This corresponds to the Pattern B (Shallow Sensors Only) condition in the Main Figure (Fig. 6 ).
Figure 13: Classification accuracy for single- vs. multi-touch discrimination at varying inter-point distances using Pattern C (All Sensors). The curve is shown in red, dotted line. Error bars indicate the standard deviation across trials. This corresponds to the Pattern C (All Sensors) condition in the Main Figure (Fig. 6 ).
Figure 14: Classification accuracy for single- vs. multi-touch discrimination at varying inter-point distances using Pattern C (Shallow Sensors Only). The curve is shown in red, dashed line. Error bars indicate the standard deviation across trials. This corresponds to the Pattern C (Shallow Sensors Only) condition in the Main Figure (Fig. 6 ).
Figure 15: TDPI-based comparison between Pattern A and C under varying weight combinations and penalty coefficients λ . Each panel shows the preferred pattern (A or C) based on the TDPI for a fixed value of λ . The axes represent task weights: w1 (vertical) for texture classification, w2 (horizontal) for light touch detection, and w3=1−w1−w2 for two-point discrimination. The results illustrate how the relative preference between Pattern A and C changes as the variability penalty coefficient λ is varied.
Robotic systems increasingly demand tactile sensing that approaches the adaptability and resolution of human skin to enable dexterous manipulation and safe interaction. OptiTac is a biomimetic tactile sensor that emulates the mechanoreceptor-to-nerve architecture of human touch by pairing each mechanical pin on a soft skin with an optical fiber acting as an artificial nerve. This design demonstrates an architectural principle for routing tactile information away from the sensing surface while preserving high spatial resolution, establishing a practical route toward distributed tactile sensing in future robotic systems. By treating tactile signals as images, simple analytical methods, rather than opaque deep-learning models, are used to infer contact location, size, and shape, providing interpretable and scalable tactile intelligence. This work demonstrates how evolutionary principles from biology can guide the development of artificial nerve systems for robots, offering a pathway toward human-like tactile perception in next-generation robotic platforms. More broadly, OptiTac establishes an artificial nerve-inspired sensing framework for interpretable robotic touch and a scalable route toward future distributed tactile systems.
Laura E. Butcher, Chris J. Ford, Nathan F. Lepora +1
School of Engineering Mathematics and Technology, University of Bristol, Bristol, BS8 1QU, UK
Proprioception in musculoskeletal humanoids is typically estimated primarily from muscle sensing, while the role of cutaneous deformation around joints remains insufficiently explored. In biological systems, mechanoreceptors distributed within soft tissue complement muscle feedback and support reliable joint state estimation. This study presents the design of a biomimetic joint-covering skin with a tissue-like layered structure that integrates pressure- and stretch-sensitive elements within the joint-covering tissue. The proposed skin is implemented on the musculoskeletal humanoid Musashi-W, and its independent proprioceptive capability as well as its integration with muscle sensing are evaluated. Experimental results show that the proposed skin alone achieves joint angle estimation with an average error of approximately 3 degrees. Furthermore, integration with muscle sensing improves estimation accuracy. Owing to its joint-covering structure, the skin may mechanically mitigate the influence of external disturbances on the muscles, and the integration of multiple modalities suggests the possibility of contributing to the identification of external stimuli that are difficult to interpret using muscle sensing alone. This work presents a design methodology for biomimetic joint-covering skin and demonstrates that such tissue-structured skin can serve as an effective approach for extending proprioceptive systems in musculoskeletal humanoids.
Department of Mechano-Informatics, Graduate School of Information Science and Technology, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan · AI Center, Graduate School of Information Science and Technology, The University of Tokyo, Japan
High-fidelity tactile feedback is essential for robotic teleoperation, enabling precise manipulation and critical decision-making. While biomimetic fingertip sensors can capture surface-texture features, how their design shapes the tactile transparency of rendered feedback remains poorly understood. This paper presents a biomimetic fingertip replicating the human finger's multilayer mechanical gradient and fingerprint morphology, with an embedded high-sensitivity inertial measurement unit capturing texture-induced vibrations for remote vibrotactile rendering. Three variants are compared with the human fingertip through temporal and spectral analyses and a user study spanning three perceptual dimensions, establishing a transmission chain from fingertip design through signal characteristics to tactile transparency. The multilayer mechanical gradient yields the clearest improvements in signal intensity and texture-feature representation, whereas a stiffer skin layer amplifies vibration intensity at the expense of feature representation. The perceptual dimensions draw on distinct signal attributes: roughness on energy scaling, granularity on spectral shape, and repetitivity on characteristic-peak representation. These findings offer design principles for application-specific fingertips in robotic teleoperation.
Shuang Yang, Fuyuan Liu, Yitian Shao
School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, Shenzhen 518055, China. · State Key Laboratory of Smart Farm Technologies and Systems, Harbin 150006, China.