Spike-based tactile pattern recognition using an extreme learning machine

We present a biologically-inspired approach for tactile pattern recognition. Our aim is to develop a low-cost tactile module that can be applied to large areas by integrating sensors with processing circuits. To accomplish this goal a flexible tactile sensor array was developed using piezoresistive fabric material. The output of the tactile array was represented as a spatiotemporal spike pattern to emulate neural signals from mechanoreceptors in the skin. A hardware implemented Extreme Learning Machine (ELM) was used to process the tactile information. The ELM chip is an event-driven system that is massively parallel and energy-efficient. For these reasons, our proposed architecture offers a fast and energy-efficient alternative for processing spatiotemporal tactile patterns. The performance of the system was evaluated during a real-time object classification task, where it achieved 90% accuracy for binary classification.