"Rats navigate and perceive their environment largely through whisker‑mediated touch, especially in poorly lit, confined spaces," explains Professor Lo. "We asked: could we replicate that sensory capability to help endoscopists feel what they cannot see?" The system uses slender, flexible whisker shafts–0.16‑mm acupuncture needles – each instrumented with a strain gauge at its base. When a whisker deflects against tissue, the strain gauge transduces that mechanical deformation into an electrical signal. Eight independent channels, arranged in task‑specific mounts, provide spatial coverage for both fine surface interrogation (a 4‑whisker underside array) and radial collision detection (a 4‑whisker peripheral configuration angled at 20°).
A major challenge was ensuring consistency across channels. "Manufacturing tolerances and mounting variations mean each channel has slightly different sensitivity and baseline," notes lead author Zeyu Wang. The team developed an affine transformation‑based calibration framework, enhanced with robust fitting, that compensates for gain and offset mismatches. This low‑computational‑overhead correction improves cross‑channel consistency and makes downstream tasks–such as shape reconstruction and force estimation–interpretable and reliable.
The system was evaluated in three controlled experiments. First, texture discrimination: using five grades of sandpaper (P80 to P2000), the whisker array was scanned across each surface. Time‑domain, frequency‑domain, and time‑frequency analyses all showed clear distinctions–coarser textures produced higher‑amplitude, lower‑frequency signals, while finer textures yielded stable, higher‑frequency responses. "This tells us the system is sensitive to fine surface irregularity, which is relevant for detecting subtle mucosal abnormalities," says Dr. Thompson.
Second, shape reconstruction: 3D‑printed benchmarks with wavy, rectangular, and triangular profiles–incorporating 1.5‑mm stepped height changes–were scanned. Uncalibrated reconstructions showed substantial mismatch with ground truth due to inter-channel variability and platform leveling errors. After applying the affine calibration, mean Pearson correlation improved from 0.081 to 0.796, and Z‑axis mean absolute error decreased from 2.22 mm to 0.81 mm – a 63% reduction. "We can now recover local geometric features at millimetre scale, which is clinically meaningful for detecting low‑profile protrusions or early narrowing," emphasizes Professor Yeatman.
Third, radial force sensing: using a 15‑mm‑diameter peripheral mount, each channel was calibrated against an ATI force sensor. Validation showed millinewton‑level errors and Pearson correlations above 0.91 across all channels. In a handheld colon phantom demonstration, the system produced cylindrical force maps that visualized contact distribution. In a simulated endoscopic insertion (two trials), the system clearly distinguished high‑collision events–where whiskers were pushed backward against the wall – from smooth, stable navigation. "This capability could provide early warning of excessive wall loading, potentially reducing perforation risk and improving procedural safety," comments Dr. Thompson.
The current prototype uses a strain‑gauge‑based transducer with a Wheatstone bridge preamplifier, a 24‑bit Σ‑Δ analog front‑end, and wireless data transmission. Inherent noise is exceptionally low (RMS 0.41 μV, peak‑to‑peak 2.91 μV), ensuring that even subtle tactile signals are captured cleanly. The system's dominant resonance is around 69 Hz – well above the task‑relevant frequency range under the scan conditions used.
However, the team acknowledges that the distal mount (15 mm outer diameter) is still too large for routine integration with standard gastrointestinal endoscopes. Future work will focus on miniaturization through MEMS‑based strain gauge fabrication, more compact packaging, and modular or reconfigurable mounts. Biocompatibility, sterilization compatibility, and calibration stability under repeated use will also need systematic assessment.
"Our system is not a replacement for vision," clarifies Professor Lo. "It is a complementary tactile layer that could be integrated into future robotic endoscopes–providing force‑aware feedback during navigation and contact‑informed interrogation of visually ambiguous regions." The authors envisage two operational modes: a navigation‑oriented configuration for radial contact monitoring during advancement, and a structure‑sensitive configuration for short‑range, controlled scanning of suspicious local areas.
Looking further ahead, the whisker system could be combined with positional tracking (electromagnetic, optical, or fibre‑based shape sensing) and integrated with learning‑based models to interpret tactile signatures in the context of disease. This study establishes the foundational hardware and calibration framework – the next step is to translate these tactile primitives into clinically actionable information..
Authors of the paper include Liping Dong, Xingxuan Zhang, Meng Li, Yi Li, Jingyi Guo, Shaorong Lu, Zhenyu Peng, Chenxi Zheng, Yufei Hui, Xiaoping Shao, Feiyan Wang, Weikang Jiang, Maoyao Li, Xianshuai Wu, Xu Guo, Yingchuan Li, Yuanyi Zheng, and Liping Zhang.
This work was supported by the Multiscale Medical Robotic Center (MRC) InnoHK, Hong Kong Innovation, and Technology Commission.
The paper "A Bionic Multichannel Whisker System for Assisting Endoluminal Intervention" was published in the journal Cyborg and Bionic Systems on Aug. 8, 2026, at DOI: 10.34133/cbsystems.0616.
