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
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.
Figures & tables
Fig. 1: Overview of the proposed tissue-structured biomimetic joint-covering skin system. a. Concept of the tissue-structured biomimetic skin. The layered structure of biological skin (epidermis, dermis, and subcutaneous tissue) and its mechanoreceptors are mimicked by a multilayer soft structure embedding receptor-like sensors, including strain gauge sensors corresponding to Merkel cells and conductive filament structures corresponding to Ruffini endings. b. Implementation on the musculoskeletal humanoid Musashi-W. (1) The conventional joint module is replaced with a biomimetic joint structure adapted from Miki et al. [ 5 ] under the CC BY 4.0 license. (2) The joint region is enclosed by the tissue-structured biomimetic skin. (3) The resulting joint-skin structure is implemented on the musculoskeletal humanoid Musashi-W for enhancing proprioception.
Fig. 2: Fabrication and integration process of the tissue-structured biomimetic joint-covering skin. a. Strain gauge sensors used as Merkel cell-like elements. b. Conductive chainmail sensors used as stretch-sensitive elements corresponding to Ruffini endings. c. Illustration of the sensor arrangement in the biomimetic skin, showing the spatial distribution of strain gauge sensors and conductive chainmail sensors. Sensors are placed with reference to biological tissue organization and embedded as densely as space permits. d. Fabrication procedure: 3D printing of the mold, embedding strain gauge sensors during dermal layer fabrication, and embedding conductive chainmail sensors during subcutaneous layer fabrication. A pair of skin components is then fabricated and combined. The epidermal layer is applied by thin coating, the biomimetic joint is enclosed, and the structure is finally mounted onto the musculoskeletal humanoid. The biomimetic joint is adapted from Miki et al. [ 5 ] under the CC BY 4.0 license.
Fig. 3: Process of sensorimotor data collection using the musculoskeletal humanoid equipped with the tissue-structured biomimetic joint-covering skin. During robot motion, muscle-related signals (muscle length and muscle tension), cutaneous signals from distributed strain gauges (Merkel-cell-like sensors) and conductive sensors (Ruffini-like sensors), and motion capture measurements are recorded simultaneously. These signals are used to construct a sensorimotor dataset for evaluating proprioceptive estimation.
Fig. 4: Comparison of single-modality joint angle estimation performance for the elbow (3DOF). RMSE are averaged across the three joint angles and paired by random seed ( n=20 ). ML: muscle length, SKIN-R: Ruffini-like skin sensor, SKIN-M: Merkel-like skin sensor, MT: muscle tension.
Modality
n
Mean RMSE
SD
Median RMSE
Muscle length (ML)
20
1.70
0.12
1.70
Ruffini-like Skin Sensor (SKIN-R)
20
2.56
0.14
2.52
Merkel-like Skin Sensor (SKIN-M)
20
2.69
0.44
2.58
Muscle tension (MT)
20
4.84
0.25
4.78
TABLE I: Summary of elbow joint angle estimation errors for single-modality conditions.
Comparison
Mean difference (ML – X)
p
pHolm
ML vs MT
−3.14
<10−5
<0.001
ML vs SKIN-M
−0.99
<10−5
<0.001
ML vs SKIN-R
−0.86
<10−5
<0.001
TABLE II: Paired statistical comparison between muscle length and other modalities (Wilcoxon signed-rank test, n=20 ).
Fig. 5: Comparison of multimodal proprioception estimation performance for the elbow (three degrees of freedom). RMSE values are averaged across targets and paired by random seed ( n=20 ). Results are shown for different modality combinations and fusion architectures.
Method
n
Mean RMSE
SD
Median RMSE
M_fused (concat)
20
2.03
0.19
2.01
M_fused (enc)
20
1.76
0.14
1.74
T_fused (concat)
20
2.48
0.22
2.45
T_fused (enc)
20
2.15
0.20
2.12
M+T (concat)
20
2.05
0.17
2.02
M+T (enc)
20
1.62
0.11
1.60
TABLE III: Summary of elbow joint angle estimation errors for multimodal fusion conditions.
Comparison
Mean diff. (ML–Fusion)
p
pHolm
ML vs M_fused (concat)
−0.334
2.0×10−6
1.1×10−5
ML vs T_fused (concat)
−0.781
2.0×10−6
1.1×10−5
ML vs T_fused (enc)
−0.446
2.0×10−6
1.1×10−5
ML vs M+T (concat)
−0.353
2.0×10−6
1.1×10−5
ML vs M_fused (enc)
−0.061
1.0×10−3
1.0×10−3
ML vs M+T (enc)
+0.077
3.2×10−4
6.5×10−4
TABLE IV: Paired statistical comparison between the muscle length modality and multimodal fusion conditions (Wilcoxon signed-rank test, n=20 ).
Fig. 6: External mechanical stimulation experiment. a. External perturbation applied to the elbow region using a rod during robot motion. b. External perturbation applied manually by hand. c. Comparison of joint angle estimation errors (RMSE) across sensing modalities under external stimulation. RMSE values are averaged across the three elbow joint angles and paired by random seed ( n=20 ). ML: muscle length, SKIN-M: Merkel-like skin sensor, SKIN-R: Ruffini-like skin sensor.
Modality
No-stim
Stim
pHolm
Muscle length (ML)
1.70±0.12
4.18±0.07
<0.001
Ruffini-like Skin Sensor (SKIN-R)
2.56±0.14
53.63±1.82
<0.001
Merkel-like Skin Sensor (SKIN-M)
2.69±0.44
40.81±1.99
<0.001
TABLE V: Effect of external mechanical stimulation on elbow joint angle estimation error (RMSE, deg).
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.
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.
In neuroscience, joint receptors have traditionally been viewed as limit detectors, providing positional information only at extreme joint angles, while muscle spindles are considered the primary sensors of joint angle position. However, joint receptors are widely distributed throughout the joint capsule, and their full role in proprioception remains unclear. In this study, we specifically focused on mimicking Type I joint receptors, which respond to slow and sustained movements, and quantified their proprioceptive potential using a biomimetic joint developed with robotics technology. Results showed that Type I-like joint receptors alone enabled proprioceptive sensing with an average error of less than 2 degrees in both bending and twisting motions. These findings suggest that joint receptors may play a greater role in proprioception than previously recognized and that the relative contributions of muscle spindles and joint receptors are differentially weighted within neural networks during development and evolution. Furthermore, this work may prompt new discussions on the differential proprioceptive deficits observed between the elbows and knees in patients with hereditary sensory and autonomic neuropathy type III. Together, these findings highlight the potential of biomimetics-based robotic approaches for advancing interdisciplinary research bridging neuroscience, medicine, and robotics.
Akihiro Miki, Shun Hasegawa, Sota Yuzaki +4
Mechano-Informatics, Graduate School of Information Science and Technology, the University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, 1138656, Tokyo, Japan
We present blind, whole-body manipulation skills on a Unitree G1 humanoid using only onboard proprioception, without cameras, markers, force-torque, or tactile sensors. Despite this minimal sensing, the trained policies exhibit surprising capability across qualitatively different tasks: push-resilient bipedal walking without IMU feedback, active soccer ball trapping with a foot, seeking and lifting a suitcase by its handle, and mounting a randomly positioned skateboard. We argue that these capabilities arise from a key underappreciated signal: the way the joint encoder readouts evolve under purposeful compliant contact, effectively forming a whole-body tactile channel. By generating contact-rich motions, the trained policies actively probe the environment; as a result, task-relevant object state (e.g., pose) becomes increasingly decodable from short proprioceptive histories. We expose this information using compact task-specific state estimators trained alongside, but fully separately from, the policies; their prediction errors decrease rapidly after informative contact. Our results indicate that joint encoder-based proprioception, combined with compliant actuation (now widely available on commercial robots and low-cost motors) is already a strong, practical substrate for whole-body dexterous manipulation and interactive perception, and therefore a natural foundation on which richer sensing can be layered.
Aditya Bhatt, Oleg Kaidanov, Puze Liu +1
Intelligent Autonomous Systems Lab, TU Darmstadt · German Research Center for AI (DFKI) · Tongji University, Shanghai Research Institute for Autonomous Intelligent Systems +1