Robotic tactile sensing is crucial for enabling safe and effective human-robot interaction. While existing flexible tactile sensors excel at detecting pressure, shear force, and strain, they often fall short in real-world object recognition. Human fingertips, conversely, adeptly identify objects through both static pressure and high-frequency vibration detection. This biological system employs slow adaptive (SA) receptors for static pressure and fast adaptive (FA) receptors for dynamic pressure changes. The challenge for artificial systems lies in achieving both high sensitivity to weak stimuli (micron-scale) and rapid response-relaxation speed for high-frequency vibration detection. Current approaches often require multiple sensors and circuits, lacking a clear understanding of the correlation between sensing performance and recognition capability. This research addresses these limitations by introducing a novel sensory system based on a single flexible sensor and defining "spatiotemporal resolution" as a key criterion for evaluating texture recognition performance.

Existing literature highlights the need for improved tactile sensing in robotics, focusing on the precise detection of pressure, shear force, and strain using various flexible sensors. However, these sensors often lack the ability to effectively perceive and recognize objects based solely on tactile interaction. In contrast, the human tactile system efficiently combines static pressure and high-frequency vibration information to identify objects. While several flexible sensor-based artificial sensory systems inspired by biological systems have been developed, they often struggle to balance high sensitivity with rapid response-relaxation speed. This frequently leads to the use of multiple sensors for static and dynamic pressure detection, complicating the system and obscuring the relationship between sensing performance and recognition accuracy. This study aims to address these limitations by presenting a single-sensor solution and establishing a quantitative correlation between spatiotemporal resolution and recognition capability.

The researchers developed a novel slip-sensor consisting of an artificial fingerprint made of polydimethylsiloxane (PDMS), an ionic gel layer of polyvinyl alcohol (PVA)-phosphoric acid, two flexible gold electrodes on polyethylene terephthalate (PET), and a flat PDMS encapsulation layer. The artificial fingerprint mimics human fingerprints, aiding in capturing vibrational stimuli. The PVA-H3PO4 gel's graded microstructure enhances sensitivity and reduces response time. The slip-sensor's high spatiotemporal resolution stems from the combination of materials selection (low-viscosity ionic gel) and microstructural design (graded structure). This design facilitates both high sensitivity (up to 519 kPa) and a fast response to high-frequency vibrations (up to 400 Hz) with a frequency resolution of 0.02 Hz. The spatial resolution is 15 μm in spacing and 6 μm in height. The sensor was integrated into a prosthetic fingertip. Experiments involved sliding the sensor over 20 different commercial textiles at both fixed and random sliding rates. Data acquisition was performed using a custom circuit board with a 24-bit analog-to-digital converter (ADC). Machine learning techniques, including a Bagging ensemble approach combining K-nearest neighbors, random forests, logistic regression, and decision trees, were employed for texture classification. The Tsfresh Python package was used for feature extraction from the time-series data. A real-time visual user interface was developed to display the classification results.

The slip-sensor demonstrated superior spatiotemporal resolution compared to human fingertips, successfully detecting surface features with spacings as small as 15 μm and heights of 6 μm. The sensor's ability to differentiate vibrations with frequencies differing by only 0.02 Hz was verified. Experiments confirmed a strong correlation between sliding rate, feature spacing, and characteristic vibration frequency. A machine-learning-based classification system achieved 100% accuracy in identifying the 20 textiles at a fixed sliding rate and 98.9% accuracy at random sliding rates. The high accuracy is attributed to the sensor's ability to capture subtle differences in texture features and the effectiveness of the Bagging ensemble learning approach. A real-time system with a visual user interface further demonstrated the practical applicability of the technology, achieving an average recognition accuracy of 98.6% at variable sliding rates. The high signal-to-noise ratio (86.79 dB) and effective number of bits (14.12 bits) of the sensor contributed to its precision. The artificial fingerprint is crucial for achieving high recognition accuracy; without it, the accuracy dropped significantly (to 54.5%).

This study successfully demonstrates a highly accurate and robust artificial sensory system for texture recognition using a single, flexible slip-sensor. The system’s high spatiotemporal resolution allows it to mimic the human ability to identify objects through tactile interaction by capturing both static pressure and high-frequency vibrations. The use of a single sensor simplifies the system compared to previous approaches that utilize multiple sensors, improving robustness and reducing complexity. The high accuracy achieved, even at variable sliding rates, highlights the potential of this technology for applications in robotics, prosthetics, and other areas requiring advanced tactile sensing. The real-time visual interface further enhances the usability and practicality of the system. This research contributes significantly to the field of tactile sensing, demonstrating a novel approach that bridges the gap between artificial and biological tactile perception.

This research presents a novel artificial sensory system for texture recognition based on a single, high-resolution slip-sensor. The system achieves remarkable accuracy in classifying textiles, even at variable sliding rates, surpassing previous approaches in accuracy and simplicity. This technology holds significant promise for advancements in robotics, prosthetics, and haptic-based virtual reality.

While the system demonstrates high accuracy in controlled settings, further research is needed to evaluate its performance in more complex and unpredictable environments. The current study focused on textile recognition; the generalizability of the system to other types of surfaces warrants investigation. The long-term stability and durability of the sensor under continuous use also require further assessment.