Cluster Haptic Texture Dataset: Haptic Texture Dataset with Varied Velocity-Direction Sliding Contacts

Abstract

Haptic sciences and technologies benefit greatly from comprehensive datasets that capture tactile stimuli under controlled, systematic conditions. However, existing haptic datasets collect data through uncontrolled exploration, which hinders the systematic analysis of how motion parameters (e.g., motion direction and velocity) influence tactile perception. This paper introduces Cluster Haptic Texture Dataset, a multimodal dataset recorded using a 3-axis machine with an artificial finger to precisely control sliding velocity and direction. The dataset encompasses 118 textured surfaces across 9 material categories, with recordings at 5 velocity levels (20-60 mm/s) and 8 directions. Each surface was tested under 160 conditions, yielding 18,880 synchronized recordings of audio, acceleration, force, position, and visual data. Validation using convolutional neural networks demonstrates classification accuracies of 96% for texture recognition, 88.76% for velocity estimation, and 78.79% for direction estimation, confirming the dataset's utility for machine learning applications. This resource enables research in haptic rendering, texture recognition algorithms, and human tactile perception mechanisms, supporting the development of realistic haptic interfaces for virtual reality systems and robotic applications.

Figures (12)

• Figure 1: Concept of our Cluster Haptic Texture Dataset. We present a novel approach to haptic texture data collection by integrating controlled sliding velocities and directions using a rubber tip as an artificial finger, thus enhancing the conventional haptic dataset framework with precise motion parameters.
• Figure 2: Haptic recording system overview: (a) Hardware setup for texture data measurement system. (b) The equipment is placed in a soundproof environment during data recording.
• Figure 3: An overview of the images of the 118 materials that comprise the Cluster Haptic Texture Dataset. The texture materials are divided into 9 categories (0-37: Wood, 38-48: Stone, 49-64: Polymer, 65-73 Metal, 74-78: Glass, 79-81: Composite, 82-86: Ceramic, 87-101: Biological Leather, 102-117: Cloth).
• Figure 4: Active noise cancellation performance. (a) Microphone signals before and after noise cancellation. (b) Transfer function of noise cancellation applied to all sound data, showing 20-40 dB attenuation in the low-frequency range below 1000 Hz where mechanical noise is expected.
• Figure 5: Overview of the friction measuring device. (a) A force gauge measures the force required to slide a weight with a urethane-rubber surface (the same material as the rubber tip) across textured samples at a constant speed of 10 mm/min. (b)(c) Static and dynamic friction coefficients of all texture materials included in the dataset.