The TIP Benchmark: A Tactile Image-Based Psychophysics-Inspired Benchmark for Artificial Tactile Sensors

Tianyi Liu, Benjamin Ward-Cherrier

Abstract: We introduce a comprehensive benchmarking method, the TIP benchmark, to assess the spatial acuity of tactile sensors. The TIP benchmark is made up of 4 stages, in which data output from a tactile sensor performing a psychophysics-inspired task is converted into images. The Structural Similarity Index Measure (SSIM) and support vector machine (SVM) classifier are then used to evaluate sensor performance across 4 metrics, representing sensor accuracy (Accuracy, Distance), stability (IQR) and generalisability (Margin). The TIP benchmark is validated to determine an ideal indentation depth and evaluate noise degradation on a tactile task, and is then employed for a grid search hardware optimization of a neuromorphic tactile sensor (9 configurations, 2 hardware design parameters). Sensors with shorter, denser internal pins are shown as having greater spatial acuity on a grating orientation discrimination task, demonstrating the TIP benchmark’s potential to quantitatively compare tactile sensors, paving the way for establishing unified methods for hardware design and benchmarking for real-world applications.