Insect-inspired, efficient event-based classification of tactile features

Abstract

Tactile sensing enables humans and animals to detect and discriminate features during exploration and guide context appropriate actions. Compared to conventional touch sensors, sensing of tactile features in animals is fundamentally event-based through spikes. Yet how sensor mechanics shape spike activity for tactile perception is not well understood. Inspired by the American cockroach--an insect touch specialist--we developed a neuromechanical framework that linked antenna passive mechanics, mechanosensory encoding, and spike-based computation. A physics-based model of antenna bending simulated spatiotemporal strain patterns during contact, which were encoded into spike trains through a strain-to-firing mapping calibrated against electrophysiological recordings. The model captured antennal nerve activity observed in vivo by reproducing key features of population-level neural responses across multiple contact locations and speeds. Compared with conventional threshold-based encoding, the insect-inspired spike encoder preserved the spatiotemporal structure of tactile signals while achieving sparser activity. To establish a link between spiking activity and perception, we trained a spiking neural network to classify contact location and speed directly from the predicted spike trains. The network achieved >95% accuracy with reduced computational demands and enabled rapid discrimination within the first 170 ms of contact, indicating that sparse, event-based codes support fast and reliable tactile perception. Together, these results establish a mechanistic bridge between sensor mechanics and neural computation, revealing how physical interactions shape efficient sensory coding. This integrative framework advances our understanding of tactile perception and provides design principles for energy-efficient, neuromorphic tactile systems. Author SummaryAnimals use touch to explore their surroundings, identify objects, and make rapid decisions. Unlike most engineered touch sensors, which continuously transmit data, biological touch systems communicate through brief electrical signals called spikes. However, how the physical properties of a touch sensor influence these signals remains poorly understood. In this study, we used the antenna of the American cockroach as a model system to investigate how mechanics and neural activity work together during touch. We developed a computational framework that links the way an antenna bends during contact to the neural signals generated by touch-sensitive sensors. By comparing our model with neural recordings from living insects, we showed that it can reproduce key patterns of neural activity observed during tactile interactions. We found that the insect-inspired encoding strategy produces sparse signals that retain important information about where and how contact occurs. These signals enabled a neural network to rapidly and accurately identify contact location and speed while using fewer computational resources. Our results suggest that tactile perception emerges from a close interaction between sensor mechanics and neural processing. Beyond advancing our understanding of animal sensation, this work provides principles for designing energy-efficient touch sensors and neuromorphic robotic systems.