A thin and soft optical tactile sensor for highly sensitive object perception

Tactile sensing is crucial in robotics and wearable devices for safe perception and interaction with the environment. Optical tactile sensors have emerged as promising solutions, as they are immune to electromagnetic interference and have high spatial resolution. However, existing optical approaches, particularly vision-based tactile sensors, rely on complex optical assemblies that involve lenses and cameras, resulting in bulky, rigid, and alignment-sensitive designs. In this study, we present a thin, compact, and soft optical tactile sensor featuring an alignment-free configuration. The soft optical sensor operates by capturing deformation-induced changes in speckle patterns generated within a soft silicone material, thereby enabling precise force measurements and texture recognition via machine learning. The experimental results show a root-mean-square error of 40 mN in the force measurement and a classification accuracy of 93.33% over nine classes of textured surfaces, including Mahjong tiles. The proposed speckle-based approach provides a compact, easily fabricated, and mechanically compliant platform that bridges optical sensing with flexible shape-adaptive architectures, thereby demonstrating its potential as a novel tactile-sensing paradigm for soft robotics and wearable haptic interfaces.

💡 Research Summary

The paper introduces a novel thin, soft optical tactile sensor that overcomes the bulk, rigidity, and alignment sensitivity of conventional vision‑based tactile sensors (VBTS). The sensor consists of a 3 mm‑thick transparent silicone elastomer embedded with microscopic glass beads acting as scattering centers, a side‑injected polarization‑maintaining fiber delivering a 635 nm laser, and a compact CMOS camera (ArduCam OV5647) that records the resulting speckle pattern. Mechanical deformation of the elastomer shifts the positions of the scattering centers, thereby modulating the internal light paths and producing reproducible changes in the speckle interference pattern. Because the speckle formation relies on volumetric scattering rather than surface markers or structured illumination, the system requires no lenses, precise optical alignment, or high‑resolution imaging optics.

Data acquisition extracts a central 128 × 128‑pixel region of interest (ROI) from each captured frame. An end‑to‑end lightweight convolutional neural network (CNN) processes these single‑channel images: three convolutional blocks (3 × 3 kernels, batch normalization, ReLU, 2 × 2 max‑pooling) feed a fully‑connected layer of 256 units, followed by a softmax classifier. Inference runs on a Raspberry Pi 5 in approximately 20 ms, enabling real‑time tactile perception.

Three experimental evaluations demonstrate the sensor’s capabilities. (1) Position recognition on both flat and curved (35 mm radius) substrates shows near‑perfect alignment between predicted and ground‑truth contact points, confirming high spatial resolution despite the sensor’s flexibility. (2) Force measurement across three distinct locations (A, B, C) yields root‑mean‑square errors of 0.039–0.040 N and mean absolute errors of 0.026–0.032 N, indicating consistent, location‑independent force sensing in the tens‑of‑millinewton range. (3) Texture classification uses nine classes (eight distinct Mahjong tiles plus a no‑contact state). With 200 training samples per class and 40 test samples, the system achieves 93.33 % overall accuracy; accuracy varies modestly with ROI selection (86–92 %). Even when the training set is reduced to 50 samples per class, accuracy remains above 90 %, highlighting the data‑efficiency of speckle‑based features.

The authors discuss several advantages: (i) elimination of bulky optics and alignment procedures, (ii) reduced dependence on camera pixel count, (iii) inherent interferometric sensitivity to minute deformations, and (iv) geometric flexibility allowing attachment to curved surfaces such as a human hand. Limitations include the need for external power and wiring for the laser and camera, and the potential influence of temperature or humidity on silicone’s optical properties, which were not fully explored. Future work is suggested to integrate light source and detector within the elastomer, and to develop environmental compensation algorithms.

In summary, the speckle‑based soft optical tactile sensor provides a compact, low‑cost, and highly compliant platform capable of precise force measurement and fine texture discrimination. Its alignment‑free architecture and machine‑learning‑driven signal decoding make it a promising candidate for soft‑robotic manipulators, wearable haptic interfaces, and other applications requiring high‑resolution tactile feedback in a form factor compatible with deformable surfaces.