Telextiles: End-to-end Remote Transmission of Fabric Tactile Sensation

TL;DR

Telextiles tackles the challenge of remote textile touch by learning a latent space that encodes textile proximity via contrastive self-supervised learning on high-resolution DIGIT tactile images. A stable sensing jig, a 512-dimensional latent encoder, and a roller-based actuator enable end-to-end remote transmission, mapping a transmitted tactile impression to the closest pre-trained textile and reproducing its feel. Evaluation shows the latent space can cluster textiles with $80.44\%$ accuracy and that a jig improves clustering performance to $99.69\%$ vs $82.84\%$ without the jig, while a user study reveals partial agreement between human judgments and model in similarity ordering. This approach promises improved online garment evaluation and remote collaboration by enabling continuous tactile recall of textiles, extending tactile communication beyond finite pattern sets.

Abstract

The tactile sensation of textiles is critical in determining the comfort of clothing. For remote use, such as online shopping, users cannot physically touch the textile of clothes, making it difficult to evaluate its tactile sensation. Tactile sensing and actuation devices are required to transmit the tactile sensation of textiles. The sensing device needs to recognize different garments, even with hand-held sensors. In addition, the existing actuation device can only present a limited number of known patterns and cannot transmit unknown tactile sensations of textiles. To address these issues, we propose Telextiles, an interface that can remotely transmit tactile sensations of textiles by creating a latent space that reflects the proximity of textiles through contrastive self-supervised learning. We confirm that textiles with similar tactile features are located close to each other in the latent space through a two-dimensional plot. We then compress the latent features for known textile samples into the 1D distance and apply the 16 textile samples to the rollers in the order of the distance. The roller is rotated to select the textile with the closest feature if an unknown textile is detected.

Figures (7)

• Figure 1: System overview of Telextiles. The training phase is assumed to be performed by the manufacturer, who trains encoders to extract a latent vector representation of the tactile sensation from a large sample of textiles. In the transmission phase, user A acquires tactile data by pressing a tactile sensor against a textile. This tactile data is sent over the network to the server. Using the learned encoder, the server extracts latent representations from the transmitted tactile data. The server then computes the closest training sample to the transmitted data in latent space and sends the raw latent vector or the ID of the closest sample to User B. User B can confirm the received tactile sensation through visual or tactile feedback.
• Figure 2: (a) The DIGIT optical tactile sensor. (b) The jig structure. The jig features a hole for inserting the spring and another hole for accommodating the sensor attachment cord. (c) Inserting the sensor into the jig. The figure demonstrates the custom-made jig tailored to the sensor's dimensions and the incorporated spring mechanism for applying pressure to the sensor.
• Figure 3: (a) The user capturing the tactility of textiles with our tactile sensing device. (b) The sensor is being pressed against the textile. It can be confirmed that the sensor is pressed in with a constant force due to the spring force.
• Figure 4: UMAP 2D visualization of the feature vectors of the training data (119 different textiles). It can be seen that the majority of the samples belong to separate clusters. Textiles with similar weave patterns, rather than similar tactile properties, cluster more closely in latent space, suggesting the limitations of the DIGIT sensor in capturing distinct tactile sensations. For example, the fabrics in the sets have similar weave and yarn thickness, but different materials and yarn stiffness. The three fabrics on the left, from left to right, were made of wool, cotton, and silk, with the left feeling stiffer. The two fabrics on the right, from left to right, were cotton and linen, with the left feeling stiffer.
• Figure 5: The hardware configuration of the actuator, including a stepping motor, a roller, a motor driver, an ESP32 microcontroller, and electronic circuitry.